Posts | SUMAN

Environmental Data — and Where Does It Come From | GIS

Mon, 27 Apr 2026 00:00:00 +0000

If you have ever read a report warning that a river is shrinking, a wetland is disappearing, or a city’s air quality is dangerous, you were reading the conclusions drawn from environmental data. But what exactly is environmental data, and who collects it in the first place?

A Simple Definition

Environmental data is any measured, observed, or recorded information about the natural world and how human activity interacts with it. It can describe the physical state of land, water, and air; the health of ecosystems and species; or the pressures that human settlements, industry, and agriculture place on natural resources.

The key word is measured. Environmental data is not opinion or estimate — it is grounded in observation, whether that observation comes from a satellite 700 km above Earth or a farmer describing how long floodwater stayed on her fields last monsoon.

The Main Types

Environmental data generally falls into a few broad families.

Physical and geographic data describes the landscape itself — elevation, soil type, river discharge, rainfall, temperature, and land cover. This is the baseline against which everything else is measured.

Ecological and biodiversity data records which species live where, in what numbers, and in what condition — fish populations in a beel, migratory birds using a haor, or the presence of a plant species that has not been recorded in a district for twenty years.

Pollution and contamination data captures what humans are releasing into the environment — agrochemical residues in soil, effluent in rivers, particulate matter from brick kilns, or arsenic in groundwater.

Socioeconomic and land use data links the natural environment to human activity — how much agricultural land was converted to settlement last decade, how many households depend on open-water fisheries, or which communities flood most frequently and why.

Hazard and disaster data documents events — flood years, drought duration, storm damage — and helps identify which areas are most vulnerable and why.

No single report uses all of these. But a serious environmental assessment draws from at least three or four types together, because environmental problems are almost never caused by one factor alone.

Where Environmental Data Comes From

Sources fall into two broad categories: primary data, which you collect yourself, and secondary data, which institutions have already collected and published.

Secondary Sources — the Foundation

For most environmental reports, secondary data from government agencies and research institutions does the heavy lifting. In Bangladesh, for example, the Bangladesh Bureau of Statistics (BBS) holds population, agricultural production, and livelihood data at the upazila level. The Soil Resource Development Institute (SRDI) has decades of soil survey records. The Bangladesh Water Development Board (BWDB) maintains river discharge measurements and flood frequency histories. The Department of Environment (DoE) holds pollution monitoring records and the legal register of Ecologically Critical Areas.

Internationally, satellite platforms make an enormous amount of environmental data freely available. Sentinel-2 and Landsat imagery from the European Space Agency and NASA respectively allow anyone with a computer to map land cover, measure water body area, and track vegetation health across time — without setting foot in the field.

Primary Sources — what only you can find

Secondary data has limits. It may be out of date. It may not exist at the upazila or union level. And it almost never captures the lived knowledge of local communities.

This is where primary data collection becomes essential. Focus Group Discussions (FGDs) with farmers, fishers, and local government members can reveal fish species that have disappeared from a beel within living memory, specific roads whose construction blocked a drainage channel, or the precise years that particular storms destroyed crops. That knowledge is irreplaceable and cannot be retrieved from any database.

Field observation — visiting a beel, walking a river bank, photographing a brick field — adds ground truth that satellite imagery alone cannot provide.

The Rule: never mix the two up

A common mistake in environmental reporting is presenting estimated or locally-observed information with the same confidence as rigorously measured institutional data. Good practice is to always signal the source and its certainty — phrases like “according to BWDB records,” “based on community consultation,” or “estimated from regional satellite analysis” give the reader the information they need to judge how much weight to place on each finding.

Why It Matters for Land Zoning

Environmental data is not collected for its own sake. Its purpose is to inform decisions — about where to protect agricultural land from development, which wetlands deserve legal conservation status, where brick fields should not be permitted, and which communities need flood resilience investment most urgently.

A land zoning plan built without solid environmental data is essentially a map drawn in the dark. It may look authoritative, but it cannot account for what is actually there — the soil that takes a generation to regenerate, the beel that is the last remaining spawning ground for five fish species, the drainage channel that, if blocked by a new road, will flood three unions every monsoon.

Getting the data right is not a bureaucratic formality. It is the difference between a zoning plan that protects a landscape and one that accelerates its destruction.

This post is part of a series on environmental planning methodology for Bangladesh. The next post covers how to analyse land use change using free satellite imagery.

Framework

A comprehensive breakdown across two categories: LLMs for GIS reasoning/coding and specialized GIS/geospatial AI platforms.

🤖 Best General LLMs for GIS Tasks

Based on the GeoBenchX benchmark — which specifically tested LLMs on multistep geospatial tasks (spatial reasoning, coordinate systems, geodata, mapping):

Claude Sonnet 3.5 and GPT-4o achieved the best overall performance across commercial LLMs tested (which included Claude Sonnet 3.5, 3.7, Haiku 3.5, Gemini 2.0, GPT-4o, GPT-4o mini, and o3-mini). Claude models excelled on solvable GIS tasks, while OpenAI models were better at identifying unsolvable scenarios.

Model	Provider	GIS Strength
Claude Sonnet 3.5/3.7	Anthropic	Best on solvable geospatial tasks
GPT-4o	OpenAI	Strong overall, good at edge cases
Gemini 2.0	Google	Solid, benefits from GEE integration
o3-mini	OpenAI	Good reasoning for spatial logic

🗺️ Specialized GIS & Geospatial AI Platforms

For Geodatabase & Full GIS Workflow

ArcGIS Pro / ArcGIS Enterprise (Esri) — Industry standard GeoAI is embedded throughout ArcGIS across geoprocessing and analysis tools. ML algorithms handle clustering, prediction (classification & regression), and spatiotemporal forecasting. Deep learning is used for pixel classification, image segmentation, object detection and tracking, change detection, and extracting geospatial data from unstructured text.

QGIS — Free & open source QGIS integrates with GRASS GIS and SAGA GIS for enhanced analytical capabilities, and supports publishing via QGIS Server through OGC-compliant web services. It’s completely free and open-source under the GNU General Public License.

For Satellite Imagery & Remote Sensing

Google Earth Engine (GEE) GEE provides access to an extensive archive of satellite imagery and environmental datasets. Its AI-driven features include detecting landscape changes using ML algorithms — tracking deforestation, urban sprawl, and natural disasters in near real-time. The cloud-based platform allows processing of enormous datasets without expensive hardware.

Planet (now Planet Labs) Planet runs the largest fleet of Earth observation satellites, imaging the entire planet’s landmass every single day. Their APIs and web interface make imagery accessible for monitoring applications needing frequent revisits.

ICEYE — For SAR (all-weather imaging) ICEYE runs the world’s largest commercial SAR (Synthetic Aperture Radar) constellation with over 60 satellites. Their Gen4 satellites hit 16cm resolution, and SAR works through clouds, rain, and darkness — something optical satellites can’t do.

For Raster & Vector AI Analysis

GeoAI (Python package) GeoAI is a Python package offering a unified framework for processing satellite imagery, aerial photographs, and vector data using deep learning models. It integrates PyTorch, Transformers, and geospatial libraries, and includes vector-to-raster and raster-to-vector conversion utilities, automated training dataset generation, and workflows for segmenting buildings, water bodies, wetlands, and solar panels.

GRASS GIS GRASS GIS provides tools for raster and vector data management, spatial modeling, and visualization. It includes over 500 modules for processing raster, vector, and 3D formats, and can connect to spatial databases and interface with various third-party systems.

SAGA GIS SAGA GIS specializes in raster image processing with a strong emphasis on image classification and calculation of vegetation indices from satellite and aerial imagery, with modules for terrain analysis derived from digital elevation models.

🧭 Quick Recommendation by Use Case

Use Case	Best Tool/Model
GIS coding & scripting help	Claude Sonnet / GPT-4o
Geodatabase management	ArcGIS Enterprise / QGIS
Satellite imagery analysis	Google Earth Engine / Planet
Raster processing	GRASS GIS / SAGA GIS / ArcGIS
Vector data & feature extraction	GeoAI (Python) / Mapflow
All-weather SAR imagery	ICEYE
No-code GIS AI analysis	GeoWGS84.ai / Picterra
Large-scale cloud geospatial	Google Earth Engine / HEAVY.AI

For most GIS professionals, the winning combo is ArcGIS or QGIS for the GIS platform + Claude or GPT-4o for AI-assisted scripting and analysis + Google Earth Engine for large-scale satellite work.

Make Chrome Feel New Again: 8 Speed Tweaks for 2026

Mon, 20 Apr 2026 00:00:00 +0000

Chrome is still the fastest browser on paper, but after a few months of tabs, extensions, and cached junk, it starts to crawl. Google has been shipping real speed improvements — Chrome 112 alone brought a 10% boost in benchmarks — but you have to turn on the right settings.

Here is the exact checklist I use when Chrome feels sluggish (tested on Windows, Mac, and low-RAM laptops in 2026).

1. Update First, Always

Google ships performance patches in the background. If you never close Chrome, you never get them.

Click ⋮ > Help > About Google Chrome
Let it update, then hit Relaunch

2. Stop the Tab Hoard

Each open tab eats RAM and a bit of CPU. More tabs = slower load times and freezes.

Press Shift+Esc to open Chrome’s Task Manager and kill memory hogs
Enable Memory Saver: Settings > Performance > Memory Saver > Maximum
Add Gmail, YouTube Music, or WhatsApp Web to “Always keep these sites active” so they never freeze

3. Prune Extensions

Extensions add features, but too many slow everything down.

Go to chrome://extensions
Toggle off anything you don’t use daily
Keep one good ad-blocker (uBlock Origin) — blocking ads is the single biggest speed win

Managed browser? You may not be able to remove work extensions, but you can still disable personal ones.

4. Clear the Cache (Not Everything)

Clearing browsing data frees disk space and fixes sites broken by old files.

⋮ > Settings > Privacy and security > Clear browsing data
Time range: Last 4 weeks
Check only “Cached images and files” — keeps you logged in

Do this once a month.

5. Turn On the Hidden Speed Switches

Preload pages: Settings > Privacy and security > Preload pages > Extended preloading. Chrome loads links before you click them.

Hardware acceleration: Settings > System > Use hardware acceleration when available > ON. Offloads graphics to your GPU.

Stop background apps: Settings > System > turn OFF “Continue running background apps when Google Chrome is closed”

6. Two Flags Worth Enabling

Type chrome://flags in the address bar:

Parallel downloading → Enabled (splits large files into chunks — huge help on slower connections)
GPU rasterization → Enabled (paints pages faster, smoother scrolling)

Relaunch Chrome after.

7. Fix Energy Saver

Energy Saver saves battery but throttles performance.

Desktop: Settings > Performance > Energy Saver > OFF
Laptop: set to “Turn on only when battery is 20% or less”

8. Reset If Nothing Else Works

Settings > Reset settings > Restore settings to their original defaults. Keeps bookmarks and passwords, removes extensions and bad tweaks.

My 2-Minute Daily Habit

Keep tabs under 10 — use OneTab to park the rest
Restart Chrome every 2–3 days to apply updates
Check Shift+Esc once a week for runaway tabs

Do steps 1–4 and you will feel the difference immediately. The rest is for power users who want Chrome to feel like it did on day one.

Best LLMs for Coding in 2026: Ranked from Number 1 to Last

Fri, 17 Apr 2026 00:00:00 +0000

Not all AI models are created equal, especially when it comes to writing, editing, and reasoning about code. With dozens of models now available across cloud APIs and local runtimes like Ollama, it is easy to get lost. This post ranks the best coding LLMs in 2026 based on real benchmarks: SWE-Bench (real GitHub issue resolution), LiveCodeBench (unseen competitive programming), and BFCL (tool/function calling accuracy).

Tier 1 - Cloud / API Models (Best Overall)

These are the top-performing models available via API. If output quality is your priority and cost is secondary, start here.

1. Claude Opus 4.7 - Best for Agentic Coding

SWE-Bench: 87.6% (1st in the world)
Best model for multi-step, autonomous coding tasks (Claude Code, Cursor, agentic loops)
Dominates real-world GitHub issue resolution, not just toy benchmarks
Use when you need the model to think, plan, and execute across files

2. Claude Sonnet 4.5 - Best Daily Driver

SWE-Bench: 82%
Faster and cheaper than Opus 4.7, with nearly the same agentic performance
The sweet spot for production coding workflows, great for CI/CD integration
Ideal for developers who run dozens of tasks a day

3. Gemini 3 Pro - Strong All-Rounder

LiveCodeBench: 79.7% | SWE-Bench: 76.2%
Google’s most capable coding model, competitive across all benchmark types
Excellent context handling and multimodal support
Good alternative if you are already in the Google ecosystem

4. GPT-5.1 / GPT-5 - Reliable OpenAI Option

SWE-Bench: 76.3% (GPT-5.1) / 74.9% (GPT-5)
Solid across the board, well-integrated with GitHub Copilot and OpenAI’s ecosystem
Strong instruction following and function calling

5. Kimi K2 Thinking - Best Raw Code Generation

LiveCodeBench: 83.1% (1st on this benchmark)
Moonshot AI’s reasoning model, excels at algorithm writing and competitive programming
Strong SWE-Bench score of 71.3%
New contender worth watching closely

6. Grok 3 (Beta) - Competitive Coding Beast

LiveCodeBench: 79.4%
xAI’s flagship, particularly strong on pure code generation tasks
Less proven on agentic/real-world benchmarks but improving fast

7. OpenAI o3-mini - Best Budget Reasoning Model

LiveCodeBench: 74.1% | SWE-Bench: 61% | BFCL: 65.1%
The most cost-efficient reasoning model for coding
Great for developers who need chain-of-thought reasoning without paying for full o3

8. DeepSeek R1 - Best for Logic-Heavy Code

LiveCodeBench: 64.3% | MATH-500: 97.3%
Open-source model with exceptional mathematical and logical reasoning
Great for algorithmic problems, data structures, and ML code
Available on free tier via DeepSeek API, accessible from Bangladesh

Tier 2 - Local / Ollama Models (Free and Self-Hosted)

These models run fully locally via Ollama, LM Studio, or similar. No API costs, no internet required after download.

1. DeepSeek V3 0324 - Best Open Source Overall

LiveCodeBench: 41% | SWE-Bench: 38.8% | BFCL: 58.5%
The best locally runnable model for general coding tasks
MIT-licensed, fast on consumer hardware with quantized versions
Recommended first choice for Ollama users

2. Qwen2.5-Coder / Qwen2.5-VL-32B - Best Tool Use Locally

BFCL Tool Use: 62.8% (highest among open models)
Alibaba’s Qwen series is purpose-built for code
Excellent for function calling, structured output, and agentic tasks locally
Available in 7B, 14B, 32B sizes, runs well on mid-range GPUs

3. Llama 4 Behemoth - Best for Reasoning (High VRAM)

LiveCodeBench: 49.4% | MATH-500: 95%
Meta’s most capable open model, requires significant hardware
Best choice if you have a high-end GPU setup (48GB+ VRAM)

4. Llama 4 Maverick - Balanced Performance

LiveCodeBench: 41%
Good balance between model size and coding performance
Multimodal support and very long context window (up to 1M tokens)

5. Gemma 3 27B - Lightweight and Capable

MATH-500: 89% | BFCL: 59.1%
Google’s open model, surprisingly strong tool use for its size
Runs comfortably on consumer hardware (16-24GB VRAM)

6. GLM-4 / GLM-5 (ZhipuAI) - Decent but Outclassed

Usable for general coding, especially Chinese-language documentation
Significantly behind DeepSeek V3 and Qwen2.5 on every major benchmark
Consider only if you have specific GLM integration requirements

Best by Use Case

Use Case	Best Model
Agentic coding (Claude Code, Cursor)	Claude Opus 4.7
Everyday coding assistant	Claude Sonnet 4.5
Raw algorithm / competitive coding	Kimi K2 Thinking
Tool and function calling (API)	GPT-4.5 (69.9% BFCL)
Budget reasoning (API)	OpenAI o3-mini
Local - best overall	DeepSeek V3 0324
Local - tool/function calling	Qwen2.5-Coder 32B
Local - low VRAM (8GB)	Qwen2.5-Coder 7B or Gemma 3
Free API (accessible in Bangladesh)	DeepSeek API or Groq (Llama/Qwen)

Key Takeaways

Claude dominates agentic coding. If you are using any coding agent (Claude Code, Cursor, Continue), Claude Opus 4.7 or Sonnet 4.5 will outperform everything else on real-world tasks.
For local/Ollama, skip GLM and use DeepSeek V3 or Qwen2.5. These are objectively stronger on every benchmark and have better English instruction-following.
The agent matters as much as the model. The same model produces meaningfully better outputs in a well-engineered agent vs. a naive one.
Free options exist. DeepSeek’s free tier and Groq’s API (running Llama/Qwen) give you near-frontier model quality at zero cost, both accessible from Bangladesh without VPN.

Benchmark data sourced from Vellum AI Coding LLM Leaderboard, updated March 2026. Rankings reflect SWE-Bench Verified, LiveCodeBench, and BFCL scores where available.

Redmi Note 15 Pro+ 5G — Complete Debloat & Ad Removal Guide

Fri, 17 Apr 2026 00:00:00 +0000

📌 Device Info

Field	Details
Device	Redmi Note 15 Pro+ 5G
Purchase Price	৳60,000 (3-month EMI)
OS	HyperOS (Global)
PC	Windows 10 Pro + Android Studio installed

⚠️ IMPORTANT — PayJoy Warning (EMI Device)

The app com.payjoy.access was found pre-installed. PayJoy is a device-financing lock app. Since this phone was bought on 3-month EMI, do NOT remove this app until all installments are paid — the finance company can remotely lock your device.

✅ Once EMI is fully paid, remove it safely:
adb shell pm uninstall -k --user 0 com.payjoy.access

🔧 ADB Methods — Choose One

Method 1 — ADB via PC (Windows)

Step 1 — Enable Developer Options

Settings → About Phone → tap "HyperOS Version" 7 times fast You’ll see: “You are now a developer!”

Step 2 — Enable USB Debugging

Settings → Additional Settings → Developer Options

Turn ON USB Debugging
Turn ON USB Debugging (Security Settings) ← Critical for HyperOS!

Step 3 — Get ADB on PC

Option A — From Android Studio (already installed):

C:\Users\YourName\AppData\Local\Android\Sdk\platform-tools\

Shift + Right-click in that folder → Open PowerShell / CMD here

Option B — Download standalone:

https://developer.android.com/tools/releases/platform-tools
Extract to C:\platform-tools\

Option C — Add to System PATH (permanent):

Win + S → “Environment Variables” → Path → Edit → New
Paste: C:\Users\Dell\AppData\Local\Android\Sdk\platform-tools
OK → OK → OK → reopen terminal

Step 4 — Connect & Verify

Plug phone with original data cable → tap ALLOW on phone popup
Change USB mode to File Transfer (MTP)

# PowerShell — use .\adb prefix
.\adb devices

# CMD — use adb directly
adb devices

Expected output:

List of devices attached
XXXXXXXXXXXXXXXX device

Output	Fix
`unauthorized`	Check phone popup, tap Allow
`offline`	Try different USB port or cable
empty list	Set USB mode to MTP; run `adb kill-server` then `adb start-server`

Method 2 — Wireless ADB via Termux (No PC Needed) ✅ Recommended

Shizuku is not available for HyperOS. Use Termux with Wireless ADB instead — full ADB control, no PC required.

Step 1 — Enable Wireless Debugging on Phone

Settings → Additional Settings → Developer Options → Wireless Debugging → Turn ON

Tap “Pair device with pairing code” — note the IP, pairing port, and pairing code.

Step 2 — Install ADB in Termux

pkg update && pkg install android-tools

Step 3 — Pair Termux with Phone

adb pair <IP>:<pairing-port>
# Example:
adb pair 192.168.1.5:41234
# Enter the pairing code when prompted: 123456

Step 4 — Connect

Back in the Wireless Debugging screen, note the main IP:Port (different from pairing port):

adb connect 192.168.1.5:5555

Step 5 — Run Commands Normally

All commands below work exactly the same — no .\ prefix needed in Termux:

adb shell pm uninstall -k --user 0 com.package.name

🗑️ Removal Results (Actual Session via PC)

Package	App	Result
`com.xiaomi.mipicks`	GetApps Store	✅ Removed
`com.miui.videoplayer`	Mi Video	✅ Removed
`com.xiaomi.joyose`	Ad/Tracking Service	✅ Removed
`com.miui.yellowpage`	Yellow Pages	✅ Removed
`com.miui.cleanmaster`	Cleaner	❌ Not installed
`com.xiaomi.gamecenter`	Game Center	❌ Not installed
`com.mi.globalbrowser`	Mi Browser	❌ Not installed
`com.miui.android.fashiongallery`	Fashion Gallery	❌ Not installed
`com.miui.systemAdSolution`	Ad System	❌ Not installed
`com.miui.weather2`	Mi Weather	❌ Protected [-1000]

[!IMPORTANT]
Use .\adb instead of adb (Quick Fix); Just add .\ before every command:

🔴 Priority — Ad & Tracking Apps (Remove First)

adb shell pm uninstall -k --user 0 com.miui.msa.global
adb shell pm uninstall -k --user 0 com.miui.analytics
adb shell pm uninstall -k --user 0 com.mi.globalminusscreen
adb shell pm uninstall -k --user 0 com.xiaomi.discover
adb shell pm uninstall -k --user 0 com.miui.cleaner
adb shell pm uninstall -k --user 0 com.miui.bugreport
adb shell pm uninstall -k --user 0 com.xiaomi.barrage
adb shell pm uninstall -k --user 0 com.miui.misightservice

Package	Why Remove
`com.miui.msa.global`	Main ad engine — serves ads system-wide
`com.miui.analytics`	Tracks usage, sends data to Xiaomi
`com.mi.globalminusscreen`	Left-swipe news feed full of ads
`com.xiaomi.discover`	“Content discovery” = ads in disguise
`com.miui.cleaner`	Fake cleaner, shows lots of ads
`com.xiaomi.barrage`	Floating danmaku feature, useless
`com.miui.bugreport`	Sends bug reports to Xiaomi servers
`com.miui.misightservice`	Sight AI tracking service

🟠 Facebook Bloatware (Remove)

These run silently in the background even if you never use Facebook:

adb shell pm uninstall -k --user 0 com.facebook.appmanager
adb shell pm uninstall -k --user 0 com.facebook.system
adb shell pm uninstall -k --user 0 com.facebook.services

🟠 Microsoft Bloatware (Remove)

adb shell pm uninstall -k --user 0 com.microsoft.appmanager
adb shell pm uninstall -k --user 0 com.microsoft.deviceintegrationservice
adb shell pm uninstall -k --user 0 com.microsoftsdk.crossdeviceservicebroker

🟡 Xiaomi Optional Apps

Package	App	What It Does	Remove If…
`com.miui.player`	Mi Music	Built-in music player	You use VLC, Auxio, etc.
`com.mi.healthglobal`	Mi Health	Fitness/step tracking	You don’t use Mi Band
`com.miui.cloudbackup`	Mi Cloud Backup	Backs up to Xiaomi cloud	You use Google Backup
`com.miui.cloudservice`	Mi Cloud Service	Core Xiaomi cloud engine	You don’t use Mi Cloud
`com.miui.micloudsync`	Mi Cloud Sync	Syncs contacts to Mi Cloud	You use Google Contacts
`com.miui.misound`	Mi Sound	Audio tuner/equalizer	You don’t use Xiaomi sound
`com.miui.touchassistant`	Touch Assistant	Floating virtual home button	You use gesture navigation
`com.xiaomi.aiservice`	Xiaomi AI Service	Powers Xiaomi AI features	You don’t use Xiaomi AI
`com.xiaomi.cameramind`	Camera Mind AI	AI scene detection	You prefer manual camera
`com.xiaomi.aicr`	AI Call Recording	Call transcription	You don’t record calls
`com.xiaomi.aiasst.vision`	AI Vision	Camera “scan to search”	You don’t use this
`com.miui.extraphoto`	Extra Photo Formats	RAW/HEIF support	You only use JPG
`com.miui.aod`	Always On Display	Clock on locked screen	You don’t use AOD
`com.mi.appfinder`	App Finder	Swipe-down app search	You use Google search
`com.miui.backup`	Mi Backup	Local + cloud backup	You back up via Google/ADB
`com.miui.phrase`	Phrases	Text snippet shortcuts	You never use this
`com.miui.qr`	Mi QR Scanner	Built-in QR scanner	You use Google Lens

adb shell pm uninstall -k --user 0 com.miui.player
adb shell pm uninstall -k --user 0 com.mi.healthglobal
adb shell pm uninstall -k --user 0 com.miui.cloudbackup
adb shell pm uninstall -k --user 0 com.miui.cloudservice
adb shell pm uninstall -k --user 0 com.miui.micloudsync
adb shell pm uninstall -k --user 0 com.miui.misound
adb shell pm uninstall -k --user 0 com.miui.touchassistant
adb shell pm uninstall -k --user 0 com.xiaomi.aiservice
adb shell pm uninstall -k --user 0 com.xiaomi.cameramind
adb shell pm uninstall -k --user 0 com.xiaomi.aicr
adb shell pm uninstall -k --user 0 com.xiaomi.aiasst.vision
adb shell pm uninstall -k --user 0 com.miui.extraphoto
adb shell pm uninstall -k --user 0 com.miui.aod
adb shell pm uninstall -k --user 0 com.mi.appfinder
adb shell pm uninstall -k --user 0 com.miui.backup
adb shell pm uninstall -k --user 0 com.miui.phrase
adb shell pm uninstall -k --user 0 com.miui.qr

🟡 Google Apps (Optional)

Package	App	Remove If…
`com.google.android.apps.tachyon`	Google Meet	You use Zoom/Jitsi
`com.google.android.apps.youtube.music`	YouTube Music	You use VLC/local music
`com.google.android.videos`	Google TV	You don’t buy from Google Play
`com.google.android.apps.bard`	Google Gemini	You use Claude 😄
`com.google.android.apps.subscriptions.red`	YouTube Premium	You don’t have YT Premium
`com.google.android.apps.wellbeing`	Digital Wellbeing	You don’t track screen time
`com.google.android.apps.safetyhub`	Safety Hub	You don’t use Google safety
`com.google.android.gm`	Gmail	You use K-9 Mail or webmail
`com.google.android.apps.docs`	Google Docs	You use LibreOffice, etc.
`com.google.android.apps.messaging`	Google Messages	You use Xiaomi Messages
`com.google.android.apps.photos`	Google Photos	You use local/other backup
`com.google.android.apps.restore`	Google Restore	Phone is already set up

adb shell pm uninstall -k --user 0 com.google.android.apps.tachyon
adb shell pm uninstall -k --user 0 com.google.android.apps.youtube.music
adb shell pm uninstall -k --user 0 com.google.android.videos
adb shell pm uninstall -k --user 0 com.google.android.apps.bard
adb shell pm uninstall -k --user 0 com.google.android.apps.subscriptions.red
adb shell pm uninstall -k --user 0 com.google.android.apps.wellbeing
adb shell pm uninstall -k --user 0 com.google.android.apps.safetyhub
adb shell pm uninstall -k --user 0 com.google.android.gm
adb shell pm uninstall -k --user 0 com.google.android.apps.docs
adb shell pm uninstall -k --user 0 com.google.android.apps.messaging
adb shell pm uninstall -k --user 0 com.google.android.apps.photos
adb shell pm uninstall -k --user 0 com.google.android.apps.restore

⚠️ Skip Gmail if you actively use it. Skip Google Photos if you rely on it for backup.

🟢 DO NOT TOUCH — Critical System Apps

Package	Why Keep
`com.android.systemui`	Core UI — removing = black screen
`com.google.android.gms`	Google Play Services
`com.android.vending`	Play Store
`com.miui.home`	Launcher
`com.miui.securitycenter`	Core security
`com.miui.securitycore`	Core security
`com.android.phone`	Phone calls
`com.android.bluetooth`	Bluetooth
`com.android.settings`	Settings app
`com.xiaomi.account`	Mi Account (Find Device)
`com.qualcomm.*`	Chipset drivers
`com.android.providers.*`	Data providers

🔍 Find Any App’s Package Name

From Phone Settings: Settings → Apps → [find the app] → ⋮ → Additional Info

From ADB:

# List all system apps
adb shell pm list packages -s

# Filter by keyword
adb shell pm list packages -s | grep "miui" # Termux / Linux
adb shell pm list packages -s | findstr "miui" # Windows CMD

🔄 Restore a Mistakenly Removed App

adb shell cmd package install-existing com.package.name

🛡️ Block Remaining Ads — Private DNS

Even after debloating, some ads slip through system apps. Block them system-wide with no root:

Settings → More Connection Settings → Private DNS → Private DNS provider hostname

Option	Hostname	Notes
✅ AdGuard DNS	`dns.adguard.com`	No account needed, just works
NextDNS	`<your-id>.dns.nextdns.io`	Needs free account at nextdns.io — your unique ID is shown on the dashboard

⚠️ Entering just nextdns.io will break internet. You must use your personal subdomain like abc123.dns.nextdns.io from your NextDNS dashboard.

Recommendation: Use dns.adguard.com — simpler, zero setup, equally effective.

📱 Open-Source Ad-Free Alternatives (F-Droid)

Removed App	Replacement	Source
Mi Browser	Mull / Fennec F-Droid (Firefox-based)	F-Droid
Mi Video	VLC / mpv for Android	F-Droid
Mi Music	Auxio / Music Player GO	F-Droid
Mi File Manager	Material Files	F-Droid
Mi Calendar	Etar Calendar	F-Droid
GetApps	F-Droid	f-droid.org
Mi Weather	Breezy Weather	F-Droid
Mi Notes	Markor (Markdown notes)	F-Droid
Mi Gallery	Aves	F-Droid / GitHub
Mi Cleaner	❌ Not needed — Android manages RAM itself	—
Game Center	❌ Just uninstall	—

💡 After OTA Updates

Apps removed via ADB are hidden for your user account, not fully deleted. A major HyperOS OTA may restore some of them. Simply re-run the relevant commands — takes about 2 minutes.

Reference guide generated from a personal Claude conversation — April 2026.
For a full community-maintained package list, see UAD-ng on GitHub.

Converting Your Web App to an Android App Using WebView

Fri, 27 Mar 2026 00:00:00 +0000

Converting Web App to an Android App Using WebView

DEPLOY YOUR REACT APP (Vercel)

✅ Build your project (LOCAL) 📂 Open your React project folder G:\lzproject\lzapp-v2 🔹 Open Terminal in this folder

👉 In VS Code:

Terminal → New Terminal 🔹 Run this command: npm run build ✅ Expected result:

You will see a new folder:

dist/

👉 This is your production build

🌐 Deploy to Vercel 🔹 Step 1 — Open Vercel

Go to: 👉 https://vercel.com

🔹 Step 2 — Login

Use:

GitHub (recommended) OR Google 🔹 Step 3 — Import Project

Click:

Add New → Project 🔹 Step 4 — Select your repo

If your project is already on GitHub:

👉 Select:

lzapp-v2 🔹 Step 5 — Configure

Vercel auto-detects Vite 👍

Check:

Framework: Vite Build Command: npm run build Output Directory: dist

👉 Click:

Deploy ⏳ Wait 1–2 minutes ✅ STEP 6 — Get your URL

After deploy, you’ll get:

https://your-project-name.vercel.app

✅ Any update- TEST BEFORE DEPLOY

Run locally:

npm run dev

👉 Open: http://localhost:5173

THEN DEPLOY AGAIN

npm run build

Every time you change UI:

1. Edit React (App.jsx / CSS)
2. npm run build
3. Push or redeploy Vercel
4. Open Android app → updated automatically, See updated UI

🚀 1. Install Android Studio (Setup)

🔹 Download & Install

Head over to the official Android Developer website: https://developer.android.com/studio

Download and install the latest version of Android Studio. The installation process is straightforward and includes everything you need to get started.

🔹 During Setup

When prompted during installation, make sure to select:

✅ Standard installation
✅ Install Android SDK
✅ Install Emulator (optional, but useful for testing)

🔹 After Installation

Launch Android Studio and click “New Project” to begin.

📱 2. Create New Android Project

In the New Project wizard, select:

Template: Empty Views Activity
Language: Java (simpler for beginners)
Minimum SDK: API 21 (Android 5.0) - this covers about 95% of active Android devices

Fill in the project details:

Name: LzApp
Package: com.yourname.lzapp

Click Finish and let Android Studio set up your project.

🌐 3. Add Internet Permission

Your web app will need internet access, so we need to add the necessary permission.

Navigate to: app > src > main > AndroidManifest.xml

Add this line just above the <application> tag:

<uses-permission android:name="android.permission.INTERNET" />

🌍 4. Add WebView in Layout

Now let’s set up the user interface. Open: res > layout > activity_main.xml

Replace the entire contents with:

<?xml version="1.0" encoding="utf-8"?>
<WebView xmlns:android="http://schemas.android.com/apk/res/android"
 android:id="@+id/webView"
 android:layout_width="match_parent"
 android:layout_height="match_parent" />

This creates a full-screen WebView that will display your web app.

⚙️ 5. Load Your Vercel App in WebView

Open: MainActivity.java

Replace the entire file with:

package com.yourname.landzoning;

import android.os.Bundle;
import android.webkit.WebView;
import android.webkit.WebViewClient;
import androidx.appcompat.app.AppCompatActivity;

public class MainActivity extends AppCompatActivity {

 WebView webView;

 @Override
 protected void onCreate(Bundle savedInstanceState) {
 super.onCreate(savedInstanceState);

 webView = new WebView(this);
 setContentView(webView);

 webView.setWebViewClient(new WebViewClient());

 webView.getSettings().setJavaScriptEnabled(true);
 webView.getSettings().setDomStorageEnabled(true);

 webView.loadUrl("https://your-vercel-app.vercel.app");
 }
}

Important: Replace https://your-vercel-app.vercel.app with your actual deployed Vercel URL.

🔙 6. Handle Back Button (IMPORTANT)

To provide a native app experience, we need to handle the back button properly. Add this method to your MainActivity.java:

@Override
public void onBackPressed() {
 if (webView.canGoBack()) {
 webView.goBack();
 } else {
 super.onBackPressed();
 }
}

This allows users to navigate back through your web app’s history before exiting the app.

⚡ 7. Improve WebView (Production Ready Basics)

Let’s enhance the WebView settings for better performance and user experience. Add these lines after the existing settings in onCreate():

webView.getSettings().setLoadWithOverviewMode(true);
webView.getSettings().setUseWideViewPort(true);
webView.getSettings().setBuiltInZoomControls(true);
webView.getSettings().setDisplayZoomControls(false);

These settings ensure your web app displays properly on different screen sizes and provides zoom controls.

🎬 8. Add Splash Screen

A splash screen provides a professional first impression. Let’s add one.

Step 1: Create splash layout

Create a new file: res > layout > splash.xml

<?xml version="1.0" encoding="utf-8"?>
<LinearLayout
 xmlns:android="http://schemas.android.com/apk/res/android"
 android:orientation="vertical"
 android:gravity="center"
 android:background="#ffffff"
 android:layout_width="match_parent"
 android:layout_height="match_parent">

 <TextView
 android:text="Land Zoning BD"
 android:textSize="24sp"
 android:textStyle="bold"
 android:textColor="#000"/>

</LinearLayout>

Step 2: Create SplashActivity

Right-click on your package → New → Activity → Empty Activity Name: SplashActivity

Replace the code with:

package com.yourname.landzoning;

import android.content.Intent;
import android.os.Bundle;
import android.os.Handler;
import androidx.appcompat.app.AppCompatActivity;

public class SplashActivity extends AppCompatActivity {

 @Override
 protected void onCreate(Bundle savedInstanceState) {
 super.onCreate(savedInstanceState);
 setContentView(R.layout.splash);

 new Handler().postDelayed(() -> {
 startActivity(new Intent(SplashActivity.this, MainActivity.class));
 finish();
 }, 2000); // 2 seconds
 }
}

Step 3: Make Splash First Screen

In AndroidManifest.xml, modify the activity declarations:

<activity android:name=".SplashActivity">
 <intent-filter>
 <action android:name="android.intent.action.MAIN" />
 <category android:name="android.intent.category.LAUNCHER" />
 </intent-filter>
</activity>

<activity android:name=".MainActivity" />

📦 9. Build APK (Production)

🔹 Build Debug APK

From the top menu: Build → Build APK(s)

After the build completes, click Locate to find your APK at: app/build/outputs/apk/debug/app-debug.apk

🔐 10. Build Release APK (IMPORTANT)

For distribution, you need a signed release APK.

Go to: Build → Generate Signed Bundle / APK

Choose:

APK
Create new keystore

Fill in:

Password
Key alias
Validity: 25 years

Then:

Select release
Finish

Your signed APK will be at: app-release.apk

🚀 11. Install on Phone

You can install the APK in two ways:

USB Method: Copy the APK to your phone and install it directly.
ADB Method: Connect your phone via USB and run:
```
adb install app-release.apk
```

⚠️ Important Tips for Your GIS App

Since you’re working with Leaflet, Firebase, and Vercel:

🔹 Security & Performance

✅ Ensure HTTPS (Vercel handles this automatically)
❌ Avoid mixed content (no HTTP resources in HTTPS pages)

🔹 Map Performance Optimization

Add these settings for better map rendering:

webView.getSettings().setAllowFileAccess(true);
webView.getSettings().setAllowContentAccess(true);

🎯 Next Steps

Once you have the basic WebView wrapper working, consider these enhancements:

Offline Caching: Implement PWA features for offline functionality
Native Loading Indicators: Add progress bars and loading states
Pull-to-Refresh: Implement swipe-to-refresh functionality
Push Notifications: Integrate Firebase Cloud Messaging
Custom Toolbar: Add navigation controls that feel native
Play Store Release: Prepare your app for Google Play Store distribution

⚠️ If Build Issues Occur

Sometimes after moving, you may see errors like:

Gradle path issue
SDK not found

Fix:

🔹 1. Re-sync Gradle

File → Sync Project with Gradle Files

🔹 2. Clean & Rebuild

Build → Clean Project
Build → Rebuild Project

🔹 3. Check SDK Path

File → Project Structure → SDK Location

Make sure it points to something like:

C:\Users\Dell\AppData\Local\Android\Sdk

👉 Summary

You’re essentially creating a native Android wrapper around your web application:

React App (Vercel) → Android WebView Wrapper → APK App

web app development

Fri, 27 Mar 2026 00:00:00 +0000

Web App

You’ve already built a solid Web + GIS dashboard system. Now we’ll upgrade it into a full ecosystem:

✅ Android App (Data Control Panel) ✅ Realtime Database (replace Google Sheets) ✅ Web App (auto-sync updates)

🧠 System Architecture (Upgraded)

🔄 Current (Your Setup) Google Sheet → Web App (Vercel) → Map UI

🚀 New System (Recommended) Android App → Firebase / Supabase DB → Web App (Vercel)

✔ No more manual Google Sheet editing ✔ Real-time sync ✔ Mobile control panel ✔ Scalable + faster

⚙️ STEP 1: Choose Backend (VERY IMPORTANT)

🥇 Best Option: Firebase (Recommended)

Realtime database
Easy Android integration
Free tier enough

Alternative:

Supabase (PostgreSQL)
Node.js API + MongoDB (more control)

👉 I’ll proceed with Firebase Firestore (best for your case)

🔥 STEP 2: Database Structure

Replace your Excel with this structure:

Collection: upazila_progress { “id”: “123”, “division”: “Dhaka”, “district”: “Gazipur”, “upazila”: “Kaliganj”, “status”: “done”, // done | ongoing | todo “updated_at”: “2026-03-24T10:00:00” }

🌐 STEP 3: Update Your Web App (IMPORTANT)

Replace Google Sheets API with Firebase fetch.

Install Firebase in your web project npm install firebase

Firebase config (JS) import { initializeApp } from “firebase/app”; import { getFirestore, collection, getDocs } from “firebase/firestore”;

const firebaseConfig = { apiKey: “XXXX”, authDomain: “XXXX”, projectId: “XXXX”, };

const app = initializeApp(firebaseConfig); const db = getFirestore(app);

export async function fetchUpazilaData() { const querySnapshot = await getDocs(collection(db, “upazila_progress”)); return querySnapshot.docs.map(doc => doc.data()); }

👉 Replace your Google Sheet logic with this.

📱 STEP 4: Android App (Core Part)

We build a Minimal Admin App

Features:

Upazila list
Status change (Done / Ongoing / ToDo)
Search
Dashboard summary

🏗 Tech Stack

Kotlin
Firebase Firestore
Jetpack Compose (or XML if you prefer)

📲 Android UI Structure Screens:

Dashboard
Upazila List
Edit Status

🧩 Example Code

Firebase Setup (Android) val db = FirebaseFirestore.getInstance()
Fetch Data fun getUpazilas(onResult: (List) -> Unit) { db.collection(“upazila_progress”) .get() .addOnSuccessListener { result -> val list = result.map { doc -> Upazila( id = doc.id, name = doc.getString(“upazila”) ?: “”, status = doc.getString(“status”) ?: “todo” ) } onResult(list) } }
Update Status (🔥 KEY PART) fun updateStatus(id: String, newStatus: String) { db.collection(“upazila_progress”) .document(id) .update(“status”, newStatus) }

👉 This is what replaces Google Sheets editing.

Simple UI (XML idea)

🔄 STEP 5: Real-Time Sync (Important Upgrade)

Instead of getDocs(), use listener:

Web + Android both onSnapshot(collection(db, “upazila_progress”), (snapshot) => { // auto update map });

👉 Now: ✔ Android change → instantly updates web map ✔ No refresh needed

📊 STEP 6: Dashboard Logic (Same as Web)

Count status:

val done = list.count { it.status == “done” } val ongoing = list.count { it.status == “ongoing” } val todo = list.count { it.status == “todo” }

🎯 FINAL RESULT Android App

✔ Change status ✔ View progress ✔ Control system

Web App

✔ Map auto-updates ✔ No Google Sheets ✔ Real-time sync

🚀 OPTIONAL ADVANCED FEATURES

If you want next-level:

🔐 Login system (Admin only) 📍 GPS tagging per update 📸 Upload field images 📶 Offline sync (Firestore supports) 📊 Analytics dashboard

⚠️ Important Advice

👉 Do NOT keep Google Sheets now 👉 Firebase = faster + scalable 👉 Your current map code can stay mostly same

🧱 STEP 1: Add Firebase to Your Project

Go to: https://console.firebase.google.com

→ Create project: landzoning

Then:

Enable: ✅ Firestore Database ✅ (Optional) Authentication later

🔑 Get Config

You’ll get something like:

const firebaseConfig = { apiKey: “XXXX”, authDomain: “XXXX”, projectId: “XXXX”, };

📦 STEP 2: Modify Your GitHub Code

In your repo 👉 Landzoning-main

Install Firebase npm install firebase
Create New File 📁 /src/firebase.js import { initializeApp } from “firebase/app”; import { getFirestore } from “firebase/firestore”;

const firebaseConfig = { apiKey: “YOUR_KEY”, authDomain: “YOUR_DOMAIN”, projectId: “YOUR_PROJECT_ID”, };

const app = initializeApp(firebaseConfig); export const db = getFirestore(app);

🔁 STEP 3: Replace Google Sheet Logic

Find your current file where you load Excel/Sheet data (Probably something like fetchData.js, App.js, or map.js)

❌ REMOVE (Old) // Google Sheet / XLSX loading fetch(“bdupazila.xlsx”)

✅ ADD (New Firebase Fetch) import { collection, onSnapshot } from “firebase/firestore”; import { db } from “./firebase”;

export function listenToUpazilaData(callback) { const unsub = onSnapshot(collection(db, “upazila_progress”), (snapshot) => { const data = snapshot.docs.map(doc => ({ id: doc.id, …doc.data() })); callback(data); });

return unsub; }

🗺 STEP 4: Connect with Your Map

Where you currently update map colors:

Replace your data source with:

listenToUpazilaData((data) => { updateMap(data); // your existing function updateStats(data); // your dashboard logic });

🎨 STEP 5: Keep Your Map Styling (NO CHANGE)

Your existing:

Leaflet map
GeoJSON boundaries
Color logic

👉 KEEP EVERYTHING SAME 👉 Only change data source

📊 STEP 6: Status Color Mapping (Ensure This)

function getColor(status) { switch (status) { case “done”: return “#22c55e”; // green case “ongoing”: return “#facc15”; // yellow default: return “#9ca3af”; // gray } }

🧪 STEP 7: Upload Initial Data to Firebase

Go to Firestore → Collection: upazila_progress

Add documents manually OR use script:

import { collection, addDoc } from “firebase/firestore”;

const data = [ { upazila: “Kaliganj”, status: “done” }, ];

data.forEach(async (item) => { await addDoc(collection(db, “upazila_progress”), item); }); 🔄 RESULT Now your system becomes: Android App → Firebase → Web Map (Auto Update)

✔ No refresh needed ✔ Real-time updates ✔ Centralized control

📱 NEXT (Android Integration)

Your Android app will:

db.collection(“upazila_progress”) .document(id) .update(“status”, “done”)

👉 Web updates instantly 🔥

🚀 BONUS: Performance Upgrade

Use indexing:

Firestore → Indexes → Add:

status ASC district ASC ⚠️ Common Mistakes

❌ Using getDocs() (not realtime) ✅ Use onSnapshot()

❌ Keeping Excel logic ✅ Fully remove it

Best ai model for coding for developer

Thu, 26 Mar 2026 00:00:00 +0000

For most developers in 2026, the “best” AI model for coding is OpenAI’s GPT‑5 series (especially GPT‑5.2 Pro / Thinking), closely followed by Anthropic’s Claude Opus 4.5 / Sonnet 4.5 and Google’s Gemini 3 Pro.¹²³⁴

Top all‑round pick

OpenAI GPT‑5.2 (Pro / Thinking)
- Very strong at solving algorithmic problems, generating clean syntax‑correct code, and writing tests and docs.³¹
- Works well via VS Code, Cursor, and other IDEs, so it fits into a typical developer workflow.²⁴

Best for reasoning and design

Claude Opus 4.5 / Sonnet 4.5
- Excellent at system‑design, large refactors, and keeping consistency across big codebases.⁴²
- Good for tasks where you need deep reasoning, not just autocomplete.²⁴

Best for context‑heavy projects

Google Gemini 3 Pro
- Huge context window lets it “see” many files at once, so it’s strong for debugging interactions across large projects.³²
- Works well for quick prototyping and when you want to upload entire directories or long logs.²³

If you want free or local

Mistral, CodeGemma, DeepSeek‑Coder / Qwen‑Coder
- Good for local/self‑hosted setups or budget‑constrained teams; they cover many languages and are strong on repetitive tasks.⁵¹³

Practical recommendation for you

Given your comfort with Python, JavaScript, and advanced tooling (VS Code, web, and cloud), you’re likely best served by:

Primary model: GPT‑5.2 Pro (via VS Code / Cursor) for daily coding, debugging, and quick prototyping.¹⁴²
Secondary model: Claude Opus 4.5 or Gemini 3 Pro for heavier design and refactoring work.⁴³²

If you tell your main stack (backend, web, data‑science, etc.), I can narrow this down to one best‑fit model and a concrete setup (IDE + extensions). ⁶⁷⁸⁹¹⁰

⁂

How to free use these models as longer complex coding require

You can use powerful coding models for free by mixing free tiers, standalone tools, and local‑run open‑source models—you just need to be strategic about when you use each.¹¹¹²¹³¹⁴

1. Use free tiers of big models

Big vendors offer limited free usage that works well for long but not “all‑day” coding:

GPT‑5 / GPT‑4.1‑based tools
- Many IDEs and wrappers (Cursor, GitHub Copilot’s chat, Warp, etc.) give you free tiers or credits (e.g., hundreds of completions/chat requests per month).¹²
- Example: GitHub Copilot free tier includes ~50 chat requests + ~2,000 completions/month without a credit card.¹²
Claude models (Opus / Sonnet)
- Some platforms (Warp, Kilo Code, Rovo Dev CLI, Amazon Q, etc.) bundle Claude Sonnet/Opus with free monthly credits or signup bonuses (e.g., 10–150 “fast” requests/month).¹²
Gemini 3 Pro / 2.5 Pro
- Google’s free tier of Gemini 3 Pro is very limited for deep reasoning (often only ~3–5 heavy prompts per day), but you can still use it for light help; Gemini 2.5 Pro is more generous in some tools.¹⁵¹²

Practical tip: rotate models (GPT‑5 → Claude → Gemini) across different tasks instead of hammering one, so you stay inside free limits.¹⁶¹¹

2. Join “unified” free portals

Several platforms let you switch between GPT‑5, Claude, Gemini, and others in one place for free, usually with a small monthly quota:

Tools like Admix or similar “vault”‑style apps front‑end GPT‑5, Claude, Gemini, and others with free plans (often after a quick signup).¹⁷¹¹
These are good for long‑form coding sessions where you can try one model until it throttles, then switch to another.¹¹¹⁷

3. Run local open‑source coding models (truly free)

For long, complex coding that you want to run without hitting API limits, install local coding‑optimized models:

Popular free models you can run on a decent laptop or cloud VM:

CodeLlama 13B / 34B – strong pure‑code model, works well via Ollama or Hugging Face.¹³
Mistral / Codestral‑22B – efficient code‑first models that fit on a single high‑end GPU.¹⁴¹³
Qwen‑Coder / Qwen3‑Coder – 30B–480B MoE models with large context, good for repo‑level refactors.¹³¹⁴

Tools to run them:

Ollama (simple, works on Linux/macOS/Windows):

ollama pull codellama:13b
ollama run codellama:13b

¹³

LM Studio / Jan – GUIs for local models that can plug into VS Code or standalone editors.¹³

Downside: local models usually need at least 16–24 GB RAM / 8–24 GB VRAM for best performance, but they’re free to run forever and keep your code private.¹⁴¹³

4. Practical workflow for you (Python + web dev)

For your stack (Python, JS, web, some data‑science / GIS), you can:

Daily coding:
- Use free tiers of GPT‑5‑enabled IDEs (Cursor, VS Code with Copilot, etc.) for quick refactors, boilerplate, and debugging.¹⁶¹²
Heavy design / refactors:
- Save your “heavy” prompts for Claude Sonnet or Gemini 3 Pro when you get free credits, e.g., re‑architecting a module or writing a long spec.¹⁵¹⁶
Long‑run / offline projects:
- Run a local CodeLlama or Qwen‑Coder model for tasks that might blow cloud free limits (large‑file refactors, generating many functions, docs).¹⁴¹³

If you tell your setup

If you share your machine (RAM/VRAM, OS, whether you use VS Code / JetBrains), I can give you a step‑by‑step plan like:

which model to install locally,
which free web tools to use, and
how to structure prompts so you don’t waste your limited free‑tier tokens on long coding tasks. ¹⁸¹⁹²⁰

⁂

how to structure prompts so I don’t waste my limited free‑tier tokens in different model. Also structure a crazy plan with tricks using smartly all free plan tire for a whole project

To avoid blowing your free‑tier tokens, you need both prompt discipline and a “free‑tier rotation” plan that lets you use GPT‑5, Claude, Gemini, and local models strategically across an entire project.²¹²²²³²⁴²⁵

Below is a two‑part answer:

How to structure prompts smartly
A “crazy but smart” free‑tier plan for a whole coding project.

1. How to structure prompts to save tokens

Use these rules for every model.

a) Set clear roles and constraints

Tell each model what it is and what it must not do:

Header pattern:
- “You are a senior Python/JS developer.
  - Task: [refactor / debug / design].
  - Constraints:
    - No long explanations unless I ask.
    - Output only code + minimal comments.
    - Respect PEP 8 / best practices.”²²²⁵

b) Split big tasks into small atomic prompts

Instead of:

“Help me build a whole API backend with auth, DB, and tests.”

Break it into:

“Design a high‑level architecture for a REST API in Python (FastAPI) with auth, Postgres, and tests.”
“Generate a minimal FastAPI main.py with a /users endpoint and auth stubs.”
“Now write only the auth middleware, no extra explanation.”

Smaller prompts = fewer tokens per request and less chance of model drifting off‑topic.²³²²

c) Re‑use contexts and “remember this”

Always keep your sessions short but consistent:

Start each coding session with:
- “Remember:
  - Project: [name]
  - Stack: Python 3.11, FastAPI, PostgreSQL, Docker.
  - Style: PEP 8, no comments unless necessary.”

Then, follow up with:

“Continue from previous context, but only write the next file: auth.py.”

This keeps the model in the same “mental state” without re‑sending your whole codebase every time.²⁵²²

d) Use “return only code” mode

Add this line at the end of code‑focused prompts:

“Return only code. Do not explain or repeat instructions.”

This cuts down ~30–50% of the token usage in each reply.²²²³

e) Let one model do “one job” well

Assign roles:

GPT‑5 / GPT‑4.1 → fast code completion, small refactors, Python/JS boilerplate.²³²²
Claude Opus/Sonnet → design, big refactors, docs, and tests.²⁶²²
Gemini 3/2.5 Pro → quick debugging, log analysis, or lighter design.²⁷²²

This way you don’t waste your best‑quality model tokens on trivial tasks (like “convert nested loop to list comprehension”).

2. “Crazy but smart” free‑tier plan for a whole project

Assume:

You have free tiers of GPT‑5, Claude, Gemini, and one local model (e.g., CodeLlama or Qwen‑Coder).²⁴²¹²⁵²³
You’re building a medium‑size backend (e.g., Python API + tests + Docker + some frontend glue).

Phase 0: Setup and token‑budgeting

Decide:
- “I’ll spend ~20% of my free tokens on GPT‑5, 30% on Claude, 30% on Gemini, 20% on local models.”²¹²³
Always:
- Use local model for heavy‑rework tasks (multi‑file refactors, docs, bulk generation).
- Use cloud models for high‑quality, short‑cycle tasks (quick fixes, small features).

Phase 1: Project scaffolding (use GPT‑5)

Goal: Get structure fast, under tight token budget.

Steps:

Architecture prompt (1 long prompt)
- “Design a Python FastAPI project with auth, PostgreSQL, async, Docker, and tests.
  - Output: directory structure as JSON, plus a short README.md sketch.”
- You paste this once; re‑use it as a “single source of truth.”
Generate skeleton files (small prompts)
- “Based on that structure, write only main.py for a minimal API with /health.”
- Then: “Now write only models.py for a User model (FastAPI + SQLAlchemy).”
Use GPT‑5 for “first passes” Each file is one small prompt; review and trim its output locally.²²²³

Token‑saving trick:

After each file, copy‑paste only the changed lines into your local repo, then delete the chat history and start fresh in the next model to avoid context bloat.

Phase 2: Heavy design and refactoring (use Claude Sonnet)

Goal: Create a solid, maintainable design and large‑scale refactors.

How to use Claude:

Feed it only the important parts
- Paste only:
  - README.md choices.
  - Core modules (auth.py, models.py, main.py).
- Prompt:
  - “Analyze this API design.
    - Suggest 3 refactors to improve maintainability and testability.
    - Return only bullet points and file names to change.”
Let Claude do “one refactor at a time”
- Prompt:
  - “Now, refactor only auth.py to be modular and testable.
    - Return only updated auth.py code.”
Use it for docs and tests plans
- “From this project, write a high‑level test plan and a short Sphinx doc structure.”

This keeps expensive Claude tokens focused on high‑value design, not trivial edits.²⁶²²

Phase 3: Debugging and small‑scale fixes (use Gemini)

Goal: Use Gemini as your “cheap” debugger and quick‑fix worker.

Strategy:

Logs and errors
- Paste only the error stack trace plus a few surrounding lines.
- Prompt:
  - “Here’s an error in my FastAPI app.
    - What is the most likely cause and a minimal fix?”
Micro‑tasks only
- “Fix only this SQL query in models.py to support async.”
- “Change all snake_case variable names to kebab_case in this function.”

Avoid using Gemini for full‑file rewrites; save that for Claude or your local model.²⁷²²

Phase 4: Long‑running / heavy work (use local models)

Goal: Do bulk, multi‑file, or repetitive work that would burn through cloud free tiers.

Examples:

Bulk documentation
- Local model prompt:
  - “Annotate each function in auth.py with a short docstring.”
Multi‑file refactors
- “Rewrite all PostgreSQL‑related functions to use async SQLAlchemy.”
Code migration
- “Convert this old Flask‑style API to FastAPI‑style endpoints.”

Because you run this locally, you can make as many iterative prompts as needed without worrying about quotas.²⁴²⁵

Phase 5: Final polish and review (rotate models)

Ask each model to review the same final file, but with different focus:
- GPT‑5: “Review this Python file for bugs and style issues. Return only line numbers and fixes.”
- Claude: “Review this file for maintainability and suggest 2–3 improvements.”
- Gemini: “Find any obvious performance issues or slow queries.”

This spreads your final‑review cost across multiple free tiers instead of overloading one.²³²²

3. Extra “tricks” to stretch free tiers

Cache and reuse:
- Save model outputs in a notes.md or model‑output/ folder.
- When you switch models, paste the last 5–10 lines of that file, not the entire project.²⁵
Use “diff” style prompts
- Instead of pasting full files:
  - “Here’s the diff of my auth.py (I added these lines):

...

 - Suggest improvements to only the new code.”

Work offline first
- Write your own first draft, then ask the AI to “fix” or “polish” it instead of generating from scratch; this massively reduces token usage.²⁵²²

What to do next for your setup

If you tell:

your OS, RAM/VRAM,
which IDE you use (VS Code / JetBrains), and
your current project (Python API, web app, GIS tool, etc.),

I can give you a concrete “token‑budget sheet” plus a step‑by‑step prompt‑template set tailored for your exact stack and hardware, so you can execute this “crazy plan” as a real workflow.

⁂

Which Ai offers delete specific chat. I can’t see options to remove immediate chat.

Also tell me how can I use ollama cloud in mobile?

Several AI services let you delete specific chats, and you can absolutely use Ollama Cloud / local‑hosted Ollama models on mobile with a few tweaks.²⁸²⁹³⁰³¹

1. Which AI lets you delete specific chats?

Most major AI chat platforms now let you delete individual threads/conversations, not just “clear all”.

Perplexity (this app)

You can delete specific threads but not “one line” inside a chat; you delete the whole thread.
- Go to Library at https://www.perplexity.ai/library.
- Click the three‑dots menu next to any thread and choose “Delete thread”.³¹³²
- You can also delete all threads at once from the top‑right menu.³¹

ChatGPT / OpenAI

In the web or desktop app:
- Hover over a chat in the sidebar, click the three dots, and choose “Delete”.³¹
You cannot delete a single message inside a chat; only the whole thread.

Claude (Anthropic)

On the web app:
- Open the History sidebar, click the three dots next to a project / chat, and select “Delete”.³³

Gemini (Google)

In the Gemini web or mobile app:
- Long‑tap a chat (mobile) or click the three dots (web) and choose “Delete”.³⁴

If you tell me which specific AI app you’re using (e.g., “the Perplexity mobile app” or “Claude Android”), I can give exact taps for that UI.

2. How to use Ollama Cloud / Ollama‑hosted models on mobile

There are two main ways:

Option A: Use Ollama Cloud via a mobile client (Android/iOS)

Ollama Cloud itself is accessed via the Ollama API (/api/generate, /api/chat), so you connect to it from a mobile app that supports Ollama servers.³⁵²⁸

Popular choices:

Off Grid (Android & iOS)
- Lets you scan and connect to a local or cloud Ollama server on your network.
- Steps:

Run Ollama on your PC/laptop (or self‑hosted server).
On your phone, open Off Grid → Remote Models → Scan Network.
Tap the detected Ollama server and start chatting.²⁹

LMSA for LM Studio & Ollama (Android)
- Supports Ollama servers and also cloud APIs like OpenRouter.
- Steps:

Enable Ollama’s local server and allow network access (OLLAMA_HOST=0.0.0.0:11434).³⁰
In LMSA, add your server IP (http://YOUR_IP:11434) and pick a model.³⁶

Third‑party Ollama‑style apps
- Apps like “MyOllama” or other open‑source clients let you connect to an Ollama server and talk to CodeLlama, Qwen‑Coder, etc. directly from your phone.³⁷

Option B: Use a web UI on mobile (no install)

Run Open WebUI or similar on your PC/server, pointing at Ollama.
Access it from your phone browser at http://YOUR_IP:port (same network) or via a secure tunnel (e.g., Pinggy, Cloudflare Tunnel).³⁰

This is the “cheapest” way if you already have a laptop running Ollama at home.

Concrete mobile setup example (you + Ollama)

If you’re on Android and already run Ollama on your laptop:

On laptop:
- Ensure Ollama listens on the network (e.g., OLLAMA_HOST=0.0.0.0:11434).³⁰
On Android:
- Install Off Grid from GitHub Releases or the Play Store.
- Open Remote Models → Scan Network → pick your Ollama machine.
- Select a coding model (e.g., codellama:13b, qwen:32b) and start chatting.²⁹

If you tell your OS (Android/iOS) and whether Ollama is on laptop vs cloud server, I can give you a step‑by‑step command + exact settings for your exact setup. ³⁸³⁹⁴⁰⁴¹

⁂

Are there any way in online/web I can run ollama free

Yes, there are ways to use Ollama‑style models for free on the web, but you cannot fully “run Ollama” in the browser the same way you run it locally—instead, you either:

Use Ollama‑compatible web UIs that talk to your own Ollama server, or
Use platforms that integrate Ollama‑compatible models and let you interact with them via a web UI.

1. Free Ollama‑compatible web UIs (you host, you browse)

These are tools that let you browse Ollama models in a browser, but you still need Ollama running somewhere (laptop, home server, or cheap cloud VM):

Open WebUI
- A popular web‑based UI that connects to your Ollama instance and lets you chat like ChatGPT.
- You install it once (Docker or native) and then open it in your browser at http://localhost:3000.
- Works free forever; you only pay for the VM if you host it in the cloud.
Ollama UI / “LlamaPen”‑style lightweight web GUIs
- Some folks have built simple no‑install‑needed web‑based GUIs that connect to a running Ollama server.
- You run Ollama on your machine, enable network access, then open the web UI in your phone browser and use it anywhere.

These are free to run, but the compute cost only vanishes if you run Ollama on your own hardware.

2. Ollama‑style cloud models (limited free tier)

Ollama now offers cloud‑hosted models that you can use via their API or some integrated tools:

Ollama Cloud models
- You can log in to Ollama and use a few cloud‑hosted models for free (limited usage).
- Heavier usage or many requests will push you into paid territory.
- You can access these via CLI or connect them to tools like Open WebUI or Lobe Chat so they feel like a web app.
Third‑party “free‑tier‑first” platforms
- Some tools let you wire Ollama‑compatible models (including Ollama Cloud) and give you a web UI on top, with a free tier.
- Examples include certain LLM‑frontend dashboards that support Ollama‑style backends.

3. Truly “free run Ollama” in a browser (no install)?

At the moment there’s no mainstream, fully‑hosted “Ollama in the browser” where you get free heavy GPU usage forever. What you can do:

Spin up a cheap cloud VM (e.g., $5–10/month) with Ollama + Open WebUI, then use it via your phone or laptop browser; effectively “online Ollama” for coding.
Use Ollama‑fronted web apps that expose a small number of models for free, but usually with rate limits.

Practical path for you (free‑focused)

Given you’re comfortable with Python and cloud tools:

Option A – DIY “online Ollama”:
- Rent a cheap Linux VM, install Ollama + Open WebUI, then access it from your browser like a web app.
- Run CodeLlama / Qwen‑Coder / Mistral for free (you pay only for the VM).
Option B – Use Ollama‑fronted web tools:
- Pick a free‑tier‑friendly Ollama‑compatible web UI (e.g., Open WebUI or Lobe Chat) and connect it to your Ollama instance or Ollama Cloud.

If you tell your preferred cloud (e.g., AWS, DigitalOcean, or home‑laptop‑only), I can give you a step‑by‑step command‑line setup tailored to your hardware and budget. ⁴²⁴³⁴⁴⁴⁵⁴⁶⁴⁷⁴⁸⁴⁹⁵⁰⁵¹

⁂

Can I host ollama in GitHub and run in mobile freely, or already exist such feature?

Short answer: You cannot host Ollama “on GitHub Pages” (the static site host), but you can run Ollama inside a GitHub Codespace and then access it from your mobile browser—effectively a “free (within limits) online Ollama you can use from phone”.⁵²⁵³⁵⁴⁵⁵

There is no official “Ollama on GitHub” button that gives you a ready‑made mobile‑friendly web UI for free forever, but you can build this pattern yourself.

1. Why you can’t put Ollama on GitHub Pages

GitHub Pages is for static sites only (HTML, JS, CSS, no backend).⁵²
Ollama is a server process that needs CPU/GPU, RAM, and a persistent runtime; Pages can’t run that.⁵²

So you cannot host Ollama itself directly on GitHub Pages and expect it to run models there.

2. What you can do: Ollama + GitHub Codespace + mobile

You can:

Run Ollama inside a GitHub Codespace (cloud‑based dev environment)
- GitHub Codespaces give you a Linux VM with CPU/RAM and a VS Code–like editor in the browser.⁵³⁵⁴
- You can install Ollama in the Codespace and run models like llama3‑8b, phi‑3, etc. inside it.⁵⁴⁵³
Expose it as a web endpoint
- Install a simple web UI (e.g., Open WebUI or a minimal chat server) in the Codespace.⁵⁶⁵⁷
- GitHub Codespace provides a public URL (e.g., your‑codespace‑xxx-443.github.dev), which you can open on your phone browser.⁵⁵⁵⁴
Connect your phone to that URL
- From your Android/iOS browser, go to the Codespace URL → open the web UI → chat with Ollama, just like a web app.⁵⁴⁵⁵

Limits:

GitHub’s free Codespace tier has limited hours and compute; heavy Ollama usage will either exhaust your free tier or push you into paid.⁵⁵⁵⁴
You run small‑to‑medium models (7B–13B) best; big 30B+ models will be slow or fail in Codespaces.⁵³⁵⁴

3. “Already exist such feature?” (community workarounds)

GitHub Codespace + Ollama + web UI
- Guides like “Run Ollama in GitHub Codespace” and community repos (Ollama‑in‑GitHub‑Codespaces) show exactly how to set this up.⁵³⁵⁴⁵⁵
Distributed Ollama APIs
- Projects like OllamaFreeAPI provide a free, community‑run API gateway to Ollama‑style models, but this is not “hosted on GitHub” and still has usage limits.⁵⁸

4. Practical “free mobile‑friendly” flow for you

If you want to use Ollama freely from mobile via GitHub‑adjacent tech:

Create a repo on GitHub and enable Codespaces.
In the Codespace .devcontainer.json, add the Ollama feature so it installs on startup.⁵⁴⁵³
Run:

ollama run llama3:8b

and optionally install a web UI (e.g., Open WebUI) that listens on a port. 4. Open the Codespace URL in your phone browser and use the web UI as your “mobile Ollama client”.

If you tell me:

your OS (Windows/Linux/macOS) and
whether you already use GitHub Codespaces,

I can give you a copy‑paste commands + devcontainer.json snippet so you can spin this up in under 10 minutes and access it from your phone. ⁵⁹⁶⁰⁶¹⁶²

⁂

Free alternatives to host Ollama for mobile access

Yes—there are several free‑tier ways to host Ollama and access it from your mobile browser without paying for a full‑time VPS. Here are the best options:

1. Free cloud VPS / “Always Free” tiers

These give you a real Linux server where you install Ollama + a web UI and then open it in your phone’s browser.

Oracle Cloud “Always Free” ARM VM
- 4–6 ARM cores, 24 GB RAM, 200 GB storage, free forever if you don’t exceed limits.
- Perfect for 7B–13B models (CodeLlama, Qwen‑Coder, Mistral) with Ollama + Open WebUI.⁶³
- Steps:

Create an Oracle Cloud account and launch an Always‑Free ARM VM.
Install Ollama and a web UI (Open WebUI / Lobe Chat).
Open the server’s IP or domain in your phone browser.

Railway.app
- Gives new users $5 free monthly credits, enough to run a small Ollama‑style app or Ollama + web UI for light usage.⁶⁴
- Good if you want a managed Docker‑style setup rather than managing a raw VM.
Other “free‑trial‑first” hosts (Kamatera, Koyeb, etc.)
- Services like Koyeb and Kamatera offer free credits or short‑term trials (e.g., up to $200 in credit).⁶⁵⁶⁶
- You can deploy Ollama as a one‑click app or custom service and stop it when you hit your free cap.

2. Developer‑cloud free tiers (GitHub‑adjacent)

These are not “Ollama as a service” but places where you can run Ollama‑compatible setups for free within limits:

GitHub Codespaces
- Free‑tier credits let you spin up a Linux VM in the browser, install Ollama, and run 7B–13B models for light work.
- You can then expose a web UI (e.g., Open WebUI) and access it from your phone via the Codespace URL.⁶⁷⁶⁸
Fly.io / Heroku‑style platforms with free tiers
- Some platforms let you deploy Ollama‑compatible backends (often via Docker) and give you a free subdomain.
- Once running, you can open that domain in your phone’s browser like a web chat app.

3. “Free” community‑run Ollama‑style APIs

These are not Ollama per se, but Ollama‑compatible ways to get models for free on mobile:

OllamaFreeAPI
- Community‑run distributed API that lets you query Ollama‑style models without paying for infrastructure yourself.⁶⁹
- You connect a mobile client (e.g., Off Grid, LMSA, or your own web UI) to this API instead of a self‑hosted server.
Other Ollama‑fronted hubs
- A few platforms aggregate open‑source models behind an Ollama‑compatible API and offer a free tier for light usage.⁷⁰⁶⁴

4. Local‑hosted + “free” mobile access

You can also run Ollama on your own hardware and make it reachable from your phone:

Home‑laptop or mini‑PC
- Install Ollama + Open WebUI, enable network access (OLLAMA_HOST=0.0.0.0:11434), and open http://your‑laptop‑ip:3000 from your phone on the same Wi‑Fi.⁷¹⁷²
Use a secure tunnel (e.g., Pinggy, Cloudflare Tunnel)
- Forward your local Ollama‑web‑UI to a public URL, then open that URL from anywhere on your phone.⁷⁰⁷¹

Simple “free‑mobile” recommendation for you

Given you’re comfortable with Linux and cloud tools, the cheapest and most “free” path is:

Pick a free‑tier host:
- Oracle Cloud Always Free (long‑term free) or Railway (low‑cost, easy).⁶³⁶⁴
Deploy:
- Ollama + Open WebUI (or similar web UI) on that machine.⁷³⁶³
Access from mobile:
- Open the server IP/domain in your phone’s browser; use it as a mobile‑friendly Ollama chat app.⁶⁴⁶³

If you tell me:

which host you prefer (Oracle / Railway / GitHub Codespaces / home‑PC), I can give you a step‑by‑step command set (what to run on the server and what to type in your phone) so you can be up and running in 10–15 minutes. ⁷⁴⁷⁵⁷⁶⁷⁷⁷⁸

⁂

If I HOST OLLAMA IN my office pc, how can I access it at home in pc/mobile?

Yes—you can host Ollama on your office PC and securely access it from home on your PC and mobile. There are two main safe patterns: same‑network tunneling and public‑accessible‑only‑via‑tunnel.

1. Basic setup on your office PC

Make Ollama listen on the network
- Set the environment variable:
  - Windows:

set OLLAMA_HOST=0.0.0.0
ollama serve

 - Linux/macOS:

export OLLAMA_HOST=0.0.0.0
ollama serve

- Now Ollama listens on `0.0.0.0:11434` instead of just `127.0.0.1`.[^8_1][^8_2][^8_3]

Optionally add a web UI
- Install Open WebUI or Lobe Chat on the same machine and point it to http://your‑office‑ip:11434; it will act as a nice chat frontend.⁷⁹⁸⁰

2. Access from home (PC + mobile)

Option A: Via a secure tunnel (easiest and safe)

You don’t expose Ollama directly to the internet; instead you use a tunnel service that forwards your office 11434 to a public URL:

Tools like Pinggy, Cloudflare Tunnel, or localtunnel
- Example with Pinggy (free tier exists):⁸¹⁷⁹

On your office PC (behind any firewall), run:

ssh -R 0:localhost:11434 qr@free.pinggy.io

Pinggy gives you a public URL (like https://xxxxx.pinggy.io).
From home: - On PC: open https://xxxxx.pinggy.io → use Ollama API or web UI at that URL. - On mobile: open the same URL in your phone browser or an Ollama‑compatible app (e.g., Off Grid, LMSA).⁸²⁸¹

This is the safest option because:

Your office PC is not exposed to the open internet.
Only the tunnel endpoint is public.

Option B: Port‑forwarding your router (more advanced, riskier)

If your office network allows it, you can expose Ollama more directly:

In your office router, set up port forwarding:
- External port: 11434 → Internal IP of your office PC: 11434.⁸³⁷⁹
From home, access:
- http://YOUR_OFFICE_PUBLIC_IP:11434 (API) or http://YOUR_OFFICE_PUBLIC_IP:3000 if you run Open WebUI on port 3000.⁸⁴⁷⁹

Security warnings:

This exposes Ollama to the entire internet unless you add:
- A reverse proxy (e.g., Nginx) with basic auth or API‑key protection.⁸⁵⁸⁶
- A VPN or IP‑whitelist (e.g., Cloudflare Access or Tailscale) so only your home IP can reach it.⁸⁶⁸⁵

3. Practical workflow for you (safe + simple)

Given your tech comfort, here’s a recommended setup:

At the office:
- Install Ollama + Open WebUI on your office PC.
- Run Ollama with OLLAMA_HOST=0.0.0.0 and keep it on your internal network.⁸⁷⁸⁴
At home:
- Use Pinggy (or similar tunnel) from your office PC so it gives you a public URL.⁸⁸⁸¹
- On your home PC / laptop: open https://xxxxx.pinggy.io like a normal web app.
- On your mobile:
  - Install an Ollama‑compatible client (e.g., Off Grid or LMSA) and point it to the Pinggy URL.⁸⁹⁸²

If you tell me:

your OS at the office (Windows/Linux) and
whether you can install SSH‑style tunnels on that machine, I can give you the exact commands for both the office PC and your home/mobile side so you can be up and running in 10 minutes. ⁹⁰⁹¹

⁂

Pinggy free tire 60 minutes tunnel timeout, are there alternatives

Yes, there are several free tunnel alternatives that either give you longer‑lived or persistent free tunnels than Pinggy’s 60‑minute limit, or are free‑forever but with small trade‑offs.⁹²⁹³⁹⁴

Here are the best options for keeping your Ollama‑on‑office‑PC tunnel active longer.

1. Cloudflare Tunnel (free + “no time limit” feel)

Pros:
- Free tier has no hard time limit (ephemeral or named tunnels can stay up as long as you run cloudflared on your office PC).
- High‑performance global network, HTTPS by default, good for serving Ollama + a web UI.⁹⁵⁹³⁹²
Setup:
- Install cloudflared on your office PC.
- Run something like:

cloudflared tunnel --hostname myollama.your‑domain.com --url http://localhost:11434

- You access `https://myollama.your‑domain.com` from home on PC/mobile.[^9_4][^9_1]

This is one of the strongest free‑forever options if you own a domain (or can use a free one via Cloudflare).

2. Localtunnel (free, simple, short‑lived)

Pros:
- Very simple: npx localtunnel --port 11434 and it assigns a public URL.
- No account or payment required.⁹³⁹⁴
Limit:
- URLs are ephemeral; the tunnel dies when you stop the process or your PC reboots.
- Not ideal for “always on”, but great for sporadic remote Ollama access.

3. Serveo (free, no‑setup style)

Serveo is a free SSH‑based tunnel like Pinggy, with no signup and no hard 60‑min cap in the free tier (just resource‑based eviction).⁹⁴⁹³
Example:

ssh -R 80:localhost:11434 serveo.net

- Gives you a public URL you can open in your phone browser.

Trade‑off: less stable than Pinggy/Cloudflare, and may be taken down or throttled if heavily used.⁹⁴

4. Tunnelmole (free + DIY‑friendly)

Tunnelmole is a self‑hostable tunnel server; you can run it yourself on a cheap VM and point your office PC at it.
The public instance has a free tier with reasonable uptime, and you can build your own “forever free” setup if you rent a minimal VPS.⁹⁶⁹⁴

5. Self‑hosted “tunnel gateways” (Zrok, frp, etc.)

Tools like Zrok or frp let you deploy a tunnel gateway on a cheap VM (or even always‑free Oracle), then connect your office Ollama to it.⁹⁶⁹⁴
You get no time‑out limit as long as the gateway is up.
Best if you’re comfortable with Linux + Docker/VMs and want a true “forever‑free” tunnel stack.⁹⁴⁹⁶

Practical recommendation for you

For your use case (Ollama on office PC → PC + mobile at home):

Best “free & stable”:
- Use Cloudflare Tunnel if you can configure a domain (even a cheap/free one). It avoids Pinggy’s 60‑min limit and still stays free.⁹²⁹³⁹⁵
If you don’t want a domain + need simple:
- Use Serveo or Localtunnel for on‑demand Ollama access from home.⁹³⁹⁴

If you tell me:

whether you have a domain (even a cheap .ml / free CF‑hosted one) and
your OS at the office (Windows/Linux),

I can give you exact commands for Cloudflare Tunnel or Serveo‑style setup so you can keep your Ollama tunnel alive much longer than 60 minutes. ⁹⁷⁹⁸⁹⁹¹⁰⁰¹⁰¹

⁂

How to set up Cloudflare Tunnel for Ollama access

Here’s a step‑by‑step plan to set up Cloudflare Tunnel so you can access your Ollama server (on your office PC) securely from home on PC and mobile, via HTTPS.¹⁰²¹⁰³¹⁰⁴

Assumptions:

Ollama is already running on your office PC at http://localhost:11434.
You have a Cloudflare account and a domain (e.g., yourdomain.com).

1. Set up Ollama to listen on the network

On your office PC:

Set Ollama to listen on all interfaces:
- Linux/macOS:

export OLLAMA_HOST=0.0.0.0
ollama serve

- Windows:

set OLLAMA_HOST=0.0.0.0
ollama serve

Ollama now accepts connections on http://127.0.0.1:11434 and http://YOUR_OFFICE_IP:11434.¹⁰⁵¹⁰⁶

2. Install `cloudflared` on your office PC

Download cloudflared
- Go to: https://github.com/cloudflare/cloudflared/releases
- Pick the binary for your OS (e.g., cloudflared-windows-amd64.exe, cloudflared-linux‑amd64).¹⁰³¹⁰²
Put it in PATH / a folder
- Example on Linux:

sudo install cloudflared-linux-amd64 /usr/local/bin/cloudflared

- On Windows: place `cloudflared.exe` in a folder like `C:\cloudflared` and add that folder to your PATH.[^10_1]

3. Authenticate `cloudflared` with your domain

Run this on your office PC once:

cloudflared tunnel login

A browser window will open; choose the domain (e.g., yourdomain.com) you want to use.
cloudflared saves a credentials file (e.g., ~/.cloudflared/cert.pem).¹⁰³¹⁰²

4. Create a tunnel for Ollama

Create a named tunnel (e.g., ollama-tunnel):

cloudflared tunnel create ollama-tunnel

This generates a ollama-tunnel.json file with an ID.¹⁰⁴¹⁰²

Now configure it:

Create a config file (e.g., ~/.cloudflared/config.yaml or C:\Users\YourUser\.cloudflared\config.yaml):

tunnel: ollama-tunnel
credentials-file: C:\Users\YourUser\.cloudflared\ollama-tunnel.json
ingress:
 - hostname: ollama.yourdomain.com
 service: http://localhost:11434
 - service: http_status:404

- `ollama.yourdomain.com` is the public URL you’ll use.
- `http://localhost:11434` is your Ollama API.[^10_1][^10_2]

Add DNS record in Cloudflare
- In Cloudflare dashboard → DNS → add:
  - Type: CNAME
  - Name: ollama
  - Target: (Cloudflare will show a random ID like xxxxx.cfargotunnel.com after you create the tunnel.)
- Or let cloudflared auto‑configure DNS:

cloudflared tunnel route dns ollama-tunnel ollama.yourdomain.com

¹⁰³¹⁰²

5. Start the tunnel (keep it running)

On your office PC:

cloudflared tunnel run ollama-tunnel

This keeps the tunnel open as long as the process runs.
For “always‑on”, you can turn it into a system service (Windows service or Linux systemd).¹⁰⁴¹⁰²

6. Access Ollama from home (PC + mobile)

Once the tunnel is up:

From home PC:
- Open: https://ollama.yourdomain.com/v1/models or https://ollama.yourdomain.com (if you also run Open WebUI there).
From mobile:
- In your browser, open https://ollama.yourdomain.com and use it like a web app, or
- In an Ollama‑compatible mobile app (e.g., Off Grid, LMSA), set the server URL to https://ollama.yourdomain.com.¹⁰⁷¹⁰⁸

7. Optional: Add a web UI in front of Ollama

If you want a nicer chat interface:

Run Open WebUI (or similar) on your office PC, pointed to http://localhost:11434.¹⁰⁹
Expose the Open WebUI port (e.g., http://localhost:3000) via Cloudflare Tunnel instead, or alongside the Ollama API:

ingress:
 - hostname: ollama-ui.yourdomain.com
 service: http://localhost:3000
 - service: http_status:404

Then on mobile you just open https://ollama-ui.yourdomain.com and chat as if it were a hosted AI app.¹⁰⁹¹⁰²

Give you exact commands (including how to create a Windows service or Linux systemd unit) so the tunnel starts automatically on boot and stays up 24/7.

⁂

Instead local pc, can I host ollama at cloudfare freely?

Short answer: you cannot run Ollama itself on Cloudflare for free, but you can use Cloudflare‑hosted AI models (Workers AI) and Cloudflare Tunnel in a free tier so it feels like you’re hosting an AI backend “in Cloudflare” while Ollama stays on your own machine.¹¹⁰¹¹¹¹¹²¹¹³

Below I’ll split this for you:

1. Why “host Ollama on Cloudflare” isn’t possible

Cloudflare Workers run serverless JavaScript (or WASM), not full OS‑level processes like Ollama.
- You cannot ssh into a Cloudflare machine, install CUDA, or run ollama serve.¹¹¹¹¹⁰
Cloudflare Workers AI is a managed LLM API (Llama 3, 4, etc.), not a place where you install Ollama yourself.¹¹⁰¹¹¹

So you cannot simply “move Ollama to Cloudflare” the way you could move it to a VPS.

2. What you can do for free

A) Use Cloudflare Workers AI as your “cloud LLM”

Cloudflare Workers AI lets you call large models (e.g., Llama 3, 4 Scout, etc.) from a Worker.¹¹¹¹¹⁰
You can:
- Write a small Worker that exposes an OpenAI‑style API endpoint.
- From your PC or mobile app, call that Worker URL as if it were an LLM backend.¹¹⁴¹¹⁰

The free tier of Cloudflare includes:

Free SSL, CDN, DDoS protection, and a limited free quota for Workers AI (a few hundred‑thousand tokens per month, details in Cloudflare’s docs).¹¹³¹¹⁰

B) Use Cloudflare Tunnel to expose your real Ollama

You still run Ollama on your PC or home server, but you expose it via Cloudflare Tunnel so it’s reachable from anywhere over https://ollama.yourdomain.com.¹¹²¹¹⁵
The Cloudflare Tunnel piece is free (you only pay if you exceed Workers CPU/request limits at very high scale).¹¹⁶¹¹³

This is the closest to “hosting Ollama in Cloudflare freely”:

Ollama runs on your hardware.
Cloudflare Tunnel gives you a free, secure HTTPS endpoint.

3. Concrete “free‑ish” setup for your use case

Given you want to access Ollama from mobile:

Keep Ollama on your office PC
- export OLLAMA_HOST=0.0.0.0 && ollama serve.¹¹⁷¹¹⁸
Expose it via Cloudflare Tunnel
- Set up cloudflared and tunnel localhost:11434 to ollama.yourdomain.com (exact steps in the previous reply).¹¹⁵¹¹²
(Optional) Replace Ollama with Workers AI for some tasks
- If you don’t strictly need Ollama, you can:
  - Write a Cloudflare Worker that calls @llama‑3 or @llama‑4 via Workers AI.
  - Use that URL in your mobile app instead of an Ollama‑hosted one.¹¹⁰¹¹¹

4. If you tell me

your domain (or whether you’re okay buying a cheap one),
and whether you prefer true Ollama on your PC vs a Cloudflare‑only API,

I can give you:

two copy‑paste examples:
- One worker.js for a Cloudflare‑only LLM API (Workers AI), and
- One config.yaml for Cloudflare Tunnel + Ollama so you can seamlessly switch between them. ¹¹⁹¹²⁰¹²¹

⁂

What is the limit of free tire for cloudfare tunnel for ollama?

Cloudflare Tunnel (via cloudflared) is free to use for exposing services like Ollama, but the real limits come from Cloudflare’s overall free plan, bandwidth, and CPU usage, not from Tunnel itself.¹²²¹²³¹²⁴

Here’s how it breaks down for your use case:

1. Cloudflare Tunnel itself (no hard “tunnel” limit)

Cloudflare Tunnel (the cloudflared tunnel component) has no explicit per‑tunnel cost or timeout in the free tier.¹²³¹²⁴
You can run cloudflared tunnel run ollama-tunnel 24/7 and it stays up as long as your PC/VM is up and within Cloudflare’s broader free limits.¹²⁵¹²³

2. Free‑plan limits that matter for Ollama

Cloudflare’s Free Plan (which you get by default with a domain):

Bandwidth & requests
- Cloudflare sits in front of your tunnel as a CDN; it handles a lot of requests for free, but extremely high traffic may push you into paid tiers (Cloudflare rarely charges for normal dev‑level usage).¹²⁴¹²⁶
Rate‑limiting & abuse protection
- Free plan includes DDoS and rate limiting; if someone scrapes your Ollama API, Cloudflare may throttle.¹²⁴

For reasonable Ollama traffic (you + maybe a few others chatting, no massive batching):

You typically stay well under free‑plan limits.

3. Workers AI vs Tunnel (for comparison)

If you ever mix in Cloudflare Workers AI (Llama‑family models via Workers):

Workers AI has a free‑tier credit (e.g., a few hundred‑thousand tokens per month); you pay only after you exceed that.¹²⁷¹²⁸¹²²
This is separate from Tunnel; you can run Tunnel‑based Ollama for free while Workers AI has its own usage cap.¹²⁶¹²²

4. Practical “Ollama + Cloudflare Tunnel” limit

For your Ollama setup:

Tunnel uptime: effectively “unlimited” as long as cloudflared runs on your PC/server and your ISP doesn’t block long‑lived TCP connections.¹²³¹²⁵
Bandwidth:
- Free tier is generous; you will notice no hard limit for typical coding‑chat usage (you’re moving small JSON payloads, not huge video streams).¹²⁶¹²⁴
CPU/server side:
- The only “limit” is on your own Ollama‑hosting machine (CPU, RAM, VRAM), not on Cloudflare.

In other words:

For one‑person Ollama + mobile access via Cloudflare Tunnel on the free plan, you can treat it as “virtually unlimited” unless you suddenly start serving thousands of users per day.¹²³¹²⁴

If you tell me:

how many users you expect (just you, your team, or public), I can tell you whether you should stay on the Free Plan or if you’ll realistically hit limits and need a higher tier.

⁂

Can run ollama coding model in phone

Yes, you can run Ollama‑style coding models on your phone, but there are two main ways:

Run Ollama directly on Android (via Termux / Linux layer).
Run Ollama on a server (PC/VPS) and talk to it from a mobile Ollama client.

Below is a practical breakdown for coding‑use.

1. Run Ollama coding models directly on Android (phone)

If your phone has a modern chip (e.g., Snapdragon 8 series, Exynos 2400, etc.) and 8–12 GB+ RAM, you can run small‑to‑medium coding‑oriented models locally.

How it’s done:
- Install Termux (Linux environment on Android, usually from F‑Droid or GitHub, not Play Store).¹²⁹¹³⁰
- In Termux, install Ollama:

pkg install ollama

- Pull a coding‑friendly model:

ollama pull codellama:7b
ollama run codellama:7b

You can also try qwen:7b, gemma2:7b, or phi‑3:mini for lighter coding‑style models.¹³¹¹³²

Performance:
- 7 B models run usable but slow on most phones; expect ~1–3 tokens/sec on not‑top‑end devices.¹³²¹³¹
- Good for:
  - Small code fixes,
  - Small refactors,
  - Inline‑style help, not large‑file‑wide rewrites.
Tutorials:
- Several guides show this setup step‑by‑step for Android (e.g., “Run Llama‑family models on Android with Termux + Ollama”).¹³³¹³¹¹³²

2. Use a mobile Ollama client plus external server

If running Ollama on the phone is too heavy or battery‑draining, you can:

Host Ollama on your PC, home server, or VPS, then
Use a mobile Ollama‑client app to chat with it.

Popular Android apps

Ollama AI Chat (Google Play, requires Android 13+)
- Connects to your Ollama server URL (e.g., http://your‑laptop‑ip:11434 or a tunnel URL).¹³⁴
- Supports models like Llama, Mistral, CodeLlama, etc.
MyOllama (open‑source client, iOS/Android)
- Connects to your Ollama‑hosted server; privacy‑oriented, no telemetry.¹³⁵
Custom / community Ollama‑for‑Android
- GitHub apps like ollama‑app‑for‑Android let you talk to your Ollama server from phone, even with image‑input models.¹³⁶

3. Practical “coding‑on‑phone” recommendation for you

Given you already run Ollama on your office/home PC:

Best balance:
- Keep Ollama on PC or home server (run codellama:13b, qwen:7b, mistral:7b there).
- Use Cloudflare Tunnel / Pinggy to expose it securely (e.g., https://ollama.yourdomain.com).
- On your phone, install Ollama AI Chat or MyOllama and point it at that URL.¹³⁷¹³⁵¹³⁴
Only run Ollama on the phone if:
- You want offline‑only coding help (e.g., on a train or rural area).
- You’re okay with slower 7‑B coding models and higher battery use.¹³¹¹³²

If you tell your phone model / RAM and whether you prefer “offline on phone” or “server + client”, I can pick you one exact model (codellama:x, qwen:x, etc.) plus a tailored command + app setup for your stack. ¹³⁸¹³⁹

⁂

SQL Queries

Sun, 18 Feb 2024 00:00:00 +0000

SQL Basics Cheat Sheet 1

SQL Basics Cheat Sheet

SQL Basics Cheat Sheet 2

Creating and managing databases

Creates a new database.

CREATE DATABASE bank;

Delete/Drop a database and all of its contents

DROP DATABASE bank;

Show all databases in your cluster.

SHOW DATABASES;

IF EXISTS can be used to prevent errors if we (for example) attempt to delete a database that doesn’t exist.

IF EXISTS can also be used with many other SQL statements and combined with other operators.

Examples:

`DROP DATABASE IF EXISTS bank;`	`CREATE DATABASE IF NOT EXISTS bank;`

DROP DATABASE … CASCADE can be used to remove all objects that rely on the database that is being dropped. DROP DATABASE … RESTRICT can be used to prevent the DROP DATABASE command from executing unless the database is empty.

Examples:

`DROP DATABASE bank CASCADE;`	`DROP DATABASE bank RESTRICT;`

Creating tables and schema

Create a new table in the database.

CREATE TABLE users (
 id UUID PRIMARY KEY,
 city STRING,
 name STRING,
 address STRING,
 credit_card STRING,
 dl STRING
);

Define a multi-column primary key:

CREATE TABLE users (
 id UUID,
 city STRING,
 name STRING,
 PRIMARY KEY (city, id)
);

Define a foreign key referencing another table in the database:

(In this case, referencing a column called city in table called locations).

CREATE TABLE users (
 id UUID PRIMARY KEY,
 city STRING REFERENCES locations(city),
 name STRING
);

Create an index based on a column:

CREATE TABLE users (
 id UUID PRIMARY KEY,
 city STRING,
 name STRING,
 INDEX (name)
);

Disallow `NULL` values from a column:

CREATE TABLE users (
 id UUID NOT NULL,
 city STRING,
 name STRING
);

Create a new table using the results of a query:

CREATE TABLE users_ny(user_id, name, city)
AS SELECT * FROM users
WHERE city = 'new york';

Managing SQL tables

ALTER TABLE

Apply a schema change to a table.

ALTER TABLE bank ADD COLUMN active BOOL;

ALTER TABLE is used with subcommands such as:

ADD COLUMN

Add a column.

ALTER TABLE bank ADD COLUMN active BOOL;

DROP COLUMN

Remove a column.

ALTER TABLE bank DROP COLUMN active;

ALTER COLUMN

Change column constraints, datatypes, etc.

ALTER TABLE bank ALTER account_balance TYPE FLOAT;

RENAME COLUMN

Rename a column.

ALTER TABLE bank RENAME COLUMN account_balance TO balance;

RENAME TO

Rename a table.

ALTER TABLE bank RENAME TO users;

DROP TABLE

Remove a table.

DROP TABLE bank;

DROP TABLE … CASCADE can be used to remove all objects (constraints, views, etc.) that rely on the table being dropped.

DROP TABLE … RESTRICT can be used to prevent the DROP TABLE command from executing unless the table is empty.

Use DROP TABLE statements with caution!

Inserting data

INSERT INTO … VALUES

Insert rows with specified values into a table.

INSERT INTO users (name, city) VALUES('Alice', 'New York');

INSERT INTO … SELECT

Insert rows into a table from the results of a query.

INSERT INTO drivers (id, city, name, address)
 SELECT id, city, name, address FROM users
 WHERE name IN ('Anita Atkinson', 'Devin Jordan');

Best practice and lessons learned

Tue, 06 Jun 2023 00:00:00 +0000

Leading evaluation thinker Michael Quinn Patton has strong opinions on the terms ‘best practice’ and ’lessons learned’. In an article from 2001, he argued:

In the endlessly hyped knowledge age of the new millennium, evaluators are being asked to generate lessons learned and best practices … The demand for knowledge acquisition, which demonstrates membership in the elite ranks of learning organizations, has crescendoed into an organizational development and program evaluation mania. But just what is popularly meant by a best by a best practice*? What does it mean to* learn a lesson*? And what’s evaluation’s role in all this?*

(emphasis in original. Patton, 2001: Evaluation, knowledge management, best practices, and high quality lessons learned,
in American Journal of Evaluation, 22:3, pp. 329-336.)

Patton added that: ‘The emphasis on knowledge generation disseminated in the form of best practices has swept like wildﬁre through all sectors of society’, although, as he explained:

I’ve looked at a lot of lists of ’lessons learned’ and ‘best practices’ and I haven’t found anything that systematically differentiates one from the other in terms of content or empirical support. Those of more modest inclination seem to favor ’lessons learned’ which connotes, at least to me, a more personal and local form of insight. Those with more chutzpah (though not necessarily, as far as I can tell, more evidence) prefer the more generalized and harder-hitting (connotatively) assertion of having discovered a ‘best practice.’ Seldom do such statements identify for whom the practice is best, under what conditions it is best, or what values or assumptions undergird its best-ness. (emphasis added).

His last sentence, above, links clearly to this course’s cross-cutting themes of equity, participation, and accountability.

Glossary of Key Terms in Evaluation and Results-Based Management

Sun, 04 Jun 2023 00:00:00 +0000

1. Introduction

This Glossary provides conceptual clarity to support thoughtful approaches to results-based management, monitoring and evaluation, which in turn support better progress on sustainable development through learning and accountability. This glossary is useful for training and in practical work on sustainable development in the public and private sectors – including during project or policy design, monitoring, programme implementation and management, and various types of evaluation. The focus of the Glossary is on specific terms as used for results monitoring, management, data collection and analysis, and evaluation. We do not define related or general concepts – such as human rights, equity, or development – as these are defined elsewhere and used here in standard, widely accepted ways. Notes define and explain key concepts (and form part of the definitions). Cross-references point the reader to related terms and synonyms. This second edition of the Glossary reflects learnings and context changes over the past 20 years. It incorporates new thinking from the OECD/DAC Results Community, which did not exist at the time of the first edition in 2002. It includes updates to definitions as well as new terms now in common use. Several definitions have been adjusted to reflect their current use by a range of actors in diverse contexts of work on sustainable development, including within communities. This edition also reflects the extensive work undertaken in 2017-2019 by the OECD Development Assistance Committee’s (DAC) Network on Development Evaluation to refine the definitions of the evaluation criteria originally outlined here in 2002. The adapted criteria definitions approved in 2019 are included in this updated edition. This process has been guided by the highest considerations of clarity and conciseness. It was conducted in a spirit of compromise in terms of the willingness of partners not to impose their own specific vocabulary on others. The original Glossary, published in 2002 was the fruit of almost two years of elaboration, based on an initial collection of terms used by development agencies. The selection of terms and their definitions were carefully discussed and analysed, and benefited from inputs from many sources and advice from the academic evaluation community. Likewise, this second edition involved extensive consultation and input from experts and partners across the results and evaluation communities.

2 Terms

Accountability

Accountability is a state of or a process for holding someone responsible to someone else for something. It is the obligation to demonstrate that work has been conducted in compliance with agreed rules and standards, or to report fairly and accurately on results, based on mandates and plans. This may require a careful, even legally defensible, demonstration that the work is consistent with the terms of a contract.

Note: Accountability may refer to the obligations of parties to act according to clearly defined performance expectations, often with respect to the prudent use of resources. For public sector managers and policy- makers it refers to being accountable to stakeholders such as taxpayers, civil society and communities for what has been done, why and how, including justifying or explaining costs and any negative results. Types of accountability include community accountability, bottom-up accountability, horizontal accountability and top-down accountability.

For evaluators, it connotes the responsibility to provide accurate, fair and credible reports and assessments.

Related terms: community accountability, mutual accountability

(Community) Accountability

Holding intervention funders and implementers accountable to the people and community affected by the intervention for the process and results of that intervention.

(Mutual) Accountability

Shared responsibility for the intervention implementation and its results. The obligation of two or more partners to demonstrate compliance with mutually agreed rules and standards or to report fairly and accurately on the use of resources and achievement of results for mandated roles or plans. Involves transparent reporting on progress in sustainable development that has been achieved collectively (through contributions from multiple partners) and implications for affected people and other stakeholders.

Activity

Actions taken or work performed through which inputs, such as funds, technical assistance and other types of resources, are mobilised to produce specific outputs.

Synonym: intervention Related term: evaluand

Adaptive management

A structured management strategy that involves an ongoing process of working collaboratively and flexibly to learn, make decisions, test assumptions, and adjust actions on the basis of new information, lessons and changes in context.

Note: Often used to take action under conditions of ongoing uncertainty based on the best available evidence. Involves systematically monitoring and evaluating results and adjusting decisions as more information is learned.

Additionality

The characteristic of an intervention, where its (financial or non-financial) inputs, activities or results are considered as additional when compared to what would have happened otherwise. Additional means larger in scale, at a higher quality, taking place more quickly, at a different location, reaching different beneficiaries, or taking place at all.

Note: In development co-operation many institutions distinguish between financial additionality (in reference to financing provided that would not be supplied by the market), and development additionality (referring to results).

Administrative data

Information collected by an intervention for purposes of administration, record keeping and tracking implementation, which can also be used for monitoring and evaluation. Note: Administrative data typically include financial and operational data at activity level. It is common for basic demographic information to be derived from administrative data. Examples include enrolment records, tax records, insurance claims, feedback forms, and patient records.

Analytical tool

Method used to process and interpret data collected or collated.

Appraisal

An overall assessment of the relevance, feasibility and potential sustainability of an intervention prior to a decision regarding funding or implementation.

Note: In development agencies, banks, and other financial institutions, the purpose of the appraisal is to enable decision makers to decide whether the activity represents an appropriate use of resources.

Related term: ex-ante evaluation

Assumptions

A set of (untested) factors and beliefs that form the basis of the intervention logic, and factors or risks, which affect its relevance, progress or success. Assumptions are the conditions necessary for the cause- and-effect relationships between the different levels of results (i.e. to move from activities to outputs, outputs to outcomes, and outcomes to impacts).

Note: For design, monitoring and evaluation, assumptions can also be understood as hypothesised conditions that affect the validity of the exercise itself, e.g., about the size or characteristics of the population when designing a sampling procedure for a survey.

Related term: hypotheses

Attribution

The ascription of a causal link between observed (or expected to be observed) changes and a specific intervention.

Note: Attribution refers to that which is to be credited for the observed changes or results achieved. This definition does not require that changes are produced solely by the intervention being evaluated, rather it represents the extent to which observed effects can be attributed to a specific intervention or to one or more partners, taking account of other interventions, confounding factors (other influences), or external shocks.

Audit

An independent, systematic, objective quality assurance assessment designed to document and improve the effectiveness of risk management, control and governance processes.

Note: A distinction is made between regularity (financial) auditing, which focuses on compliance with applicable statutes and regulations; and performance auditing, which is concerned with relevance, efficiency and effectiveness. Internal auditing provides an assessment of internal controls undertaken by a unit reporting to management while external auditing is conducted by an independent entity.

Baseline (reference value)

The conditions existing prior to an intervention or at the beginning of the period, against which changes can be measured, monitored and evaluated.

Baseline study

An analysis describing the situation prior to an intervention, against which changes can be assessed or comparisons made.

Benchmark

A reference point or standard against which changes, performance or achievements can be assessed.

Note: A benchmark is established based on recent performance of comparable organisations, or what can be reasonably inferred to have been achieved under a similar set of circumstances.

Beneficiaries (People who benefit)

The individuals, groups, or organisations, whether targeted or not, that benefit, directly or indirectly, from the intervention.

Note: Other terms used, depending on the context, include rights holders, duty bearers, or affected people. It is often useful to refer to “intended” beneficiaries, and one should not assume that all people have equal access to or benefit equally (or at all) from the intervention. In fact evaluations often serve the purpose of investigating the extent to which intended benefits were achieved, for whom and how (including differential benefits across different groups of people).

Related terms: reach, target group

Causality

The relationship between one event (the cause) and another event (the effect) which is the direct consequence of the first.

Cluster evaluation

An evaluation of a set of related activities or interventions, either similar interventions in different locations or a cluster of complementary components of an overall initiative.

Related terms: thematic evaluation, sector programme evaluation

Coherence

The compatibility of the intervention with other interventions in a country, sector or institution.

Note: The extent to which other interventions (particularly policies) support or undermine the intervention, and vice versa. Includes internal coherence and external coherence:

Internal coherence The synergies and interlinkages between the intervention and other interventions carried out by the same institution/government, as well as the consistency of the intervention with the relevant international norms and standards to which that institution/government adheres.

External coherence The consistency of the intervention with other actors’ interventions in the same context. This includes complementarity, harmonisation and co-ordination with others, and the extent to which the intervention is adding value while avoiding duplication of effort.

Related term: policy coherence for sustainable development

Policy coherence for sustainable development

The compatibility of a policy with other policies in a country, sector or institution in terms of supporting sustainable development. The extent to which other policies support or undermine the intervention, and vice versa.

Related term: coherence

Commissioner

The person or entity responsible for requesting or ordering an evaluation.

Conclusions

A summary of the factors leading to the success or failure of the evaluated intervention. Special attention is paid to the intended and unintended effects, results and impacts, and more generally to any other strength or weakness. A conclusion draws on data collection and analyses undertaken, through a transparent chain of arguments.

Context

The setting in which an intervention or an evaluation takes place and which is likely to influence performance and results. These include capacities and social, economic, political, environmental, conflict, inclusiveness, cultural, and institutional conditions.

Contribution – [definition under discussion]

Contribution analysis

An approach for determining if – and how – an intervention contributed to an observed result, based on verifying the underlying theory of change.

Control group

The sample or group that does not receive the intervention and against which other groups or samples (that do receive the intervention) are compared in order to assess performance and results.

Counterfactual

The situation or condition that hypothetically may prevail for individuals, organisations, or groups were there no intervention (the status quo).

Note: This is used for counterfactual evaluation approaches. It can be estimated by creating a control group, a comparison group or a hypothetical counterfactual.

Country programme evaluation

Evaluation of one or more institution’s or partner’s portfolio of interventions in a specific country, including the strategy behind them in a specific period of time.

Culture of results and learning; Culture of evaluation

The collective formal and informal norms, beliefs and ways of working in an institution that support and encourage staff and partners to actively seek out, learn from and act on credible results information. A culture that supports and encourages people to critically think about the design, implementation and effects of interventions, including relevance, coherence, efficiency, effectiveness, impact and sustainability. This includes understanding causal relationships, testing assumptions and considering evidence generated internally or by external actors on what works or not, why and for whom, and adjusting actions accordingly.

Data collection tool

Methods used to identify information sources and collect information. Examples include informal and formal surveys, direct and participatory observations, community interviews, focus groups, expert opinions, case studies, and literature search.

Democratic evaluation

An approach to evaluation that aims to support a just and democratic society and serve the whole community.

Note: Democratic evaluation allows people to be informed of what others are doing and sees the evaluator as someone who brokers the process. It generally uses inclusive practices that foster participation and collaboration to understand progress towards a just and democratic society. Also used as a means of ensuring public accountability and transparency.

Developmental evaluation

An iterative, embedded approach to evaluation, designed to support learning particularly in complex or uncertain environments. It involves providing real-time, or near real-time, (independent) feedback to intervention staff, thus facilitating ongoing learning and enabling improvements during implementation.

Related terms: real-time evaluation, adaptive management, ex-ante evaluation

Economic

The conversion of inputs (funds, expertise, natural resources, time, etc.) into outputs, outcomes and impacts in the most cost-effective way possible, as compared to feasible alternatives in the context.

Related term: efficiency

Economical

Absence of waste for a given output. An activity is economical when the costs of the scarce resources used approximate the minimum needed to achieve the planned objectives.

Related term: efficiency

Economic evaluation

An analysis that quantifies whether the economic benefits of a project (achieved or expected to be achieved) exceed its economic costs. A systematic analysis that assesses the resources used in relation to the results achieved, providing an assessment of cost effectiveness.

Note: Different types of economic analysis include cost-benefit, value for money, cost-effectiveness and cost-utility.

Related term: efficiency

Effects

Intended or unintended changes due directly or indirectly to an intervention. Related terms: outputs, outcomes, impacts, results

Effectiveness

The extent to which the intervention achieved, or is expected to achieve, its objectives, and its results, including any differential results across groups.

Note: Analysis of effectiveness involves taking account of the relative importance of the objectives or results.

Related terms: efficiency, coherence, relevance, impact, sustainability

Efficiency

The extent to which the intervention delivers, or is likely to deliver, results in an economic and timely way.

Note: “Economic” is the conversion of inputs (funds, expertise, natural resources, time, etc.) into outputs, outcomes and impacts in the most cost-effective way possible, as compared to feasible alternatives in the context. “Timely” delivery is within the intended timeframe, or a timeframe reasonably adjusted to the demands of the evolving context. This may include assessing operational efficiency (how well the intervention was managed).

Related terms: effectiveness, coherence, relevance, impact, sustainability

End line

The conditions existing after an intervention or end of the period, against which changes from the baseline can be measured, monitored and evaluated.

Related term: baseline

Evaluability

Extent to which an intervention can be evaluated in a reliable and credible fashion.

Note: Some approaches to evaluability assessment involve early review of a proposed intervention in order to ascertain whether its objectives are adequately defined and its results are verifiable. In other instances, particularly with complex interventions, high uncertainty or in unstable contexts, evaluability assessment might instead identify a need for an evaluation approach that supports adaptive management (for example, developmental evaluation).

Evaluand

The institution or intervention that is being evaluated; the object of the evaluation.

Evaluation

The systematic and objective assessment of a planned, ongoing or completed intervention, its design, implementation and results. The aim is to determine relevance, coherence, effectiveness, efficiency, impact

and sustainability. Evaluation also refers to the process of determining the worth or significance of an intervention.

An evaluation should provide information that is credible and useful, enabling the incorporation of lessons learned into decision-making processes.

Note: Evaluation in some instances involves the definition of appropriate standards and criteria, the examination of performance against those standards, an assessment of actual and expected results and the identification of relevant lessons and recommendations. Though evaluation deals with the assessment of relevance, coherence, effectiveness, efficiency, impact and sustainability, not all evaluations will cover all of these criteria to the same degree or at all.

Evaluation purpose

The objectives of the evaluation including why the evaluation is being undertaken at this particular point in time, for whom and how the evaluation will be used for learning and accountability.

Evidence

Facts or information that support the validity and truth of a conclusion, assumption or assertion.

Evidence-based (policy, practice)

Reliable and credible evidence determines the design, adaptation and implementation of a policy or practice.

Ex-ante evaluation

An evaluation that is performed before the implementation of an intervention. Related terms: appraisal, quality at entry, evaluability

Ex-durante evaluation

An evaluation that is performed during the implementation of an intervention (while it is ongoing).

Ex-post evaluation

Evaluation of an intervention after it has been completed.

Note: An ex-post evaluation may be undertaken directly following or some time after the intervention. The intention is to identify the factors of success or failure, to assess the sustainability of results and impacts, and to draw conclusions that may inform other interventions.

External evaluation

The evaluation of an intervention conducted by entities or individuals outside the funding and implementing organisations.

Related term: independent evaluation

Evaluation feedback

The transmission of findings generated through the evaluation process to parties for whom it is relevant and useful so as to facilitate learning or accountability. This may involve the collection and dissemination of findings, conclusions, recommendations and lessons from experience to people affected by the intervention, as well as funders, implementers, decision makers and other stakeholders.

Feminist evaluation

A value-based approach to evaluation grounded in feminist theory and principles, which emphasises empowerment, social justice and participation, as well as understanding structural barriers that prevent equality, while aiming to contribute to transforming gendered power relationships.

Related terms: gender analysis, gender-responsive evaluation, human rights-based evaluation, participatory evaluation

Finding

A finding uses evidence from one or more evaluations to allow for a factual statement.

Formative evaluation

Evaluation intended to improve performance or to inform planning of a subsequent phase, often conducted during the implementation phase of the intervention.

Note: Formative evaluations may also be conducted for other reasons such as compliance, legal requirements or as part of a larger evaluation initiative.

Related terms: process evaluation, developmental evaluation

Gender analysis

An examination of gender dynamics, norms, beliefs and practices, including concepts of masculinity, femininity, and other gender identities, and power relationships among people of different genders.

Note: For planning, monitoring and evaluation, gender analysis can be used to understand the intervention and its context, as well as for analysing the results of the intervention and for making evaluative judgements.

Gender-responsive evaluation

An evaluation that assesses changes to gendered power relationships and results from an intervention, determining whether, and how, changes have occurred, and the effects of those changes. The approach to the evaluation ensures that the voices of people of different genders are incorporated throughout the process and in the methods used.

Goal

The higher-order objective to which an intervention is intended to contribute. Note: An example is the 2030 Sustainable Development Goals (SDGs). Related term: objective

Human rights-based evaluation

An evaluation approach that explicitly identifies rights-holders and duty-bearers, and aims to hold duty- bearers to account for the realisation of human rights. The evaluation applies human rights principles to its approach, including through participation of rights-holders in the evaluation process itself, and explores issues of justice, equity and equality with a focus on marginalised and excluded groups.

Hypotheses

A set of (testable) ideas, beliefs and explanations about the relationship between an intervention and its effects, in a given context.

Related terms: assumptions, theory of change, intervention logic

Impact

The extent to which the intervention has generated or is expected to generate significant positive or negative, intended or unintended, higher-level effects.

Note: Impact addresses the ultimate significance and potentially transformative effects of the intervention. It seeks to identify social, environmental and economic effects of the intervention that are longer term or broader in scope than those already captured under the effectiveness criterion. Beyond the immediate results, impact seeks to capture the indirect, secondary and potential consequences of the intervention. It does so by examining the holistic and enduring changes in systems or norms, and potential effects on people’s well-being, human rights, gender equality, and the environment.

Related terms: relevance, coherence, effectiveness, efficiency, sustainability

Impacts

The higher-level effects of an intervention’s outcomes. The ultimate effects or longer-term changes resulting from the intervention. Such impacts can include intended and unintended, positive or negative higher-level effects.

Note: Impacts is used here in the plural in reference to its meaning as a type or level of result, as distinct from the impact criterion. “Impacts” and “results” are sometimes used interchangeably, which creates confusion; impacts should be used to refer to higher-level results.

Related term: transformative effects

Impact evaluation

An evaluation that assesses the degree to which the intervention meets its higher-level goals and identifies the causal effects of the intervention. Impact evaluations may use experimental, quasi-experimental and non-experimental approaches.

Note: The term is also sometimes used to refer only to evaluations that use explicit counterfactual analysis to determine the effects (including outputs and outcomes) caused by an intervention.

Impact management

A management strategy focusing on performance and achievement of impacts. This management approach provides the framework, tools and guidance for strategic planning, risk management, performance monitoring, evaluation and knowledge management. It serves four complementary purposes: decision- making, learning, accountability and communication.

Synonym: results-based management

Independent evaluation

An evaluation that is free from undue political influence and organisational pressure, which has full access to information and autonomy to carry out investigations and reports findings and conclusions based on the collected evidence.

Note: Independence is often supported by having external individuals or institutions carry out the evaluation. Independence can also be supported through funding arrangements, reporting lines and quality assurance processes. The credibility of an evaluation depends in part on how independently it has been carried out, including the extent to which it is free from the potential biases of those responsible for the design and implementation of the intervention.

Indicator

Quantitative or qualitative factor or variable of interest, related to the intervention and its results, or to the context in which an intervention takes place.

Note: An indicator is always approximate only (i.e. not an exact measure) and requires interpretation and explanation, even if assessed accurately. Indicators should provide simple, verifiable, and reliable means to track changes and performance.

Related terms: KPI, performance indicator

Inputs

The financial, human, material (in-kind), and institutional (including technological and information) resources used for the intervention.

Note: It is important to include not only the resources of a funding or implementing organisation but the totality of resources of all involved organisations, the community and the local environment used for the intervention.

Institutional development impact

The extent to which an intervention improves or weakens the ability of a country or region to make more efficient, equitable, and sustainable use of its human, financial, and natural resources, for example through: (a) better definition, stability, transparency, enforceability and predictability of institutional arrangements, or

(b) better alignment of the mission and capacity of an organisation with its mandate, which derives from these institutional arrangements. Such impacts can include intended and unintended effects of an action.

Internal evaluation

Evaluation of an intervention conducted by a unit or individuals internal to the organisation, generally staff of the organisation who report to the management.

Related term: self-evaluation

Intervention

The intentional activity or effort that is being evaluated (also called the evaluand or object of the monitoring or evaluation).

Note: Interventions may be international, national, local or through partnerships, and are aimed at supporting sustainable development, climate and humanitarian goals. They include but are not limited to development interventions, humanitarian aid, peacebuilding activities, climate mitigation activities, climate adaptation activities, market-based interventions, private sector engagement, normative work, and non-sovereign operations. Examples of such efforts include strategies, policy advice, technical assistance, financing mechanisms, programmes, institutions, or projects.

Synonym: evaluand Related term: activity

Intervention logic

The way an intervention is expected to achieve its desired results, including underlying assumptions about the causality and interaction between the intervention, its inputs, activities, outputs, outcomes and impacts, in the context of the intervention. Related terms: logical framework, theory of change, assumptions

Intervention purpose

The stated (or implied) objectives of the intervention. Related terms: intervention objective, goal

Joint evaluation

An evaluation in which two or more institutions or partners participate.

Note: There are various degrees of “jointness” depending on the extent to which individual partners are involved in designing the evaluation, co-operate in the evaluation process, merge their evaluation resources and combine their evaluation reporting. Joint evaluations can help overcome attribution problems in assessing the effectiveness of programmes and strategies. They can also be useful for evaluating coherence, looking at the complementarity of efforts supported by different partners, the quality of co-ordination, etc.

Knowledge management

The process or approach through which institutions capture, distribute, retain and effectively use knowledge to achieve their goals.

Lessons learnt

Generalisation or extrapolation of findings and translation of analysis into relevant knowledge that supports decision making, improves performance and promotes the achievement of better results in other settings (beyond the intervention being evaluated). Frequently, lessons highlight strengths or weaknesses in the preparation, design, and implementation of interventions that affect performance and results. A lesson may be positive, neutral or negative.

Limitations

Any constraints in the process, methodology or data that affect the monitoring or evaluation, including potential implications for validity and reliability. Includes any obstruction of a free and open monitoring and evaluation process that may influence the findings, as well as discrepancies between the planned and actual products or process.

Logical framework (Log frame)

Management tool used to improve the design of interventions, most often at the project level. It involves identifying strategic elements (inputs, activities, outputs, outcomes, impacts) and their causal relationships, as well as indicators, and the assumptions or risks that may influence success and failure. It facilitates planning, execution, monitoring and evaluation of an intervention.

Related terms: results-based management, theory of change, intervention logic

Meta-evaluation

The term is used for evaluations designed to synthesise findings from a series of evaluations. It can also be used to denote the assessment of an evaluation to judge its quality or scrutinise the performance of the evaluators.

Mid-term evaluation

Evaluation performed towards the middle of the period of implementation of the intervention. Related terms: formative evaluation, developmental evaluation.

Monitoring

A continuing process that involves the systematic collection or collation of data (on specified indicators or other types of information). Provides the management and other stakeholders of an intervention with indications of the extent of implementation progress, achievement of intended results, occurrence of unintended results, use of allocated funds and other important intervention and context-related information.

Related terms: performance monitoring, performance indicator, indicator

Objective (Sustainable development objective)

Intended positive impacts contributing to physical, financial, institutional, social, well-being, environmental, or other benefits to a society, community, or group of people via one or more interventions.

Related term: intervention objective

Intervention objective

The overall purpose of an intervention. This includes the intended physical, financial, institutional, social, environmental, or other results that an intervention is expected to achieve or to which it is expected to contribute.

Related terms: intervention purpose, goal

Outcomes

The short-term and medium-term effects of an intervention’s outputs.

Note: Outcomes are often changes in the institutional and behavioural capacities for development conditions that occur between the completion of outputs and the achievement of impacts.

Related terms: results, outputs, impacts, effect.

Outputs

The products, capital goods and services that result from an intervention. Outputs may also include changes resulting from the intervention that contribute to the achievement of outcomes. Outputs include changes in knowledge, skills, or abilities produced by the activities.

Note: Outputs are within the control of the implementing team and attributable to it. Related term: results

Participatory monitoring and evaluation

An evaluation approach in which partners (including target groups) work together and are actively involved in monitoring and evaluation including designing plans, collecting and interpreting data, documenting and using findings, and formulating conclusions and recommendations.

Partner

An individual or organisation that collaborates on an intervention to achieve mutually agreed objectives.

Note: The concept of partnership connotes shared goals, common responsibility for outcomes, distinct accountabilities and reciprocal obligations. Partners may include communities, governments (national regional, local), civil society organisations, non-governmental organisations, universities, professional and

business associations, private entities, and multi-lateral organisations involved in funding, implementing or overseeing the intervention.

Performance

The degree to which an intervention or partner operates according to specific criteria, standards and guidelines, or achieves results in accordance with stated goals or plans.

Performance indicator

Quantitative or qualitative factor or variable that provides a simple, verifiable, and reliable means to measure the performance of an actor, generally in terms of the process of implementation.

Related terms: KPI, indicator

Key Performance Indicator (KPI)

A subset of indicators that are considered to be the most important to achieving goals, and used to monitor progress, in an ongoing way.

Related terms: indicator, performance

Performance measurement

A system for assessing performance of interventions against stated goals. Related terms: performance monitoring, performance indicator, indicator.

Performance monitoring

A continuous process of collecting and analysing data to compare how well an intervention is being implemented against expected results.

Related term: performance measurement

Performance measurement framework

The performance measurement framework is a results-based management tool used to systematically plan the collection of relevant data over the lifetime of the project, in order to assess and demonstrate progress made in achieving expected results. The performance measurement framework is the “skeleton” of the monitoring plan: it documents the major elements of the monitoring system in order to ensure regular collection of actual data on the performance measurement framework indicators. The performance measurement framework contains all of the indicators used to measures progress on the achievement of the project’s results.

In addition, it specifies who is responsible for collecting data on the indicator, from what source, at what frequency and with what method. It also includes the baseline data and target for each indicator.

Related term: results framework

Population

The total universe of persons or institutions in the context in which the intervention takes place; all potential targets of the intervention.

Note: The target group of an intervention is a subset of the population Related terms: target group, people who benefit, stakeholders

Process evaluation

An evaluation of the internal dynamics of implementing organisations, their policy instruments, their service delivery mechanisms, their management practices, and the linkages among these.

Related term: formative evaluation

Programme evaluation

Evaluation of a set of interventions, combined to attain specific global, regional, country, or sector development objectives.

Note: a development programme is a time-bound intervention involving multiple activities that may cut across sectors, themes and geographic areas.

Related term: country programme evaluation.

Project evaluation

Evaluation of an individual intervention designed to achieve specific objectives within specified resources and implementation schedules, often within the framework of a broader programme, examining its relevance, coherence, effectiveness, efficiency, impact and sustainability.

Note: Cost-benefit analysis is one instrument often used in project evaluation for projects with measurable benefits. When benefits cannot be quantified, cost effectiveness is a suitable approach.

Quality assurance

Any activity or process that is used to assess and improve the merit or the worth of an intervention or its compliance with given standards and requirements.

Note: Examples of quality assurance activities include appraisal, results-based management, reviews, and evaluations. Quality assurance may also refer to the assessment of the quality of a portfolio and its overall effectiveness.

Quality at entry

The strength of the intervention implementation and results, at the time it is designed or before it starts, in terms of its potential relevance, effectiveness, coherence, efficiency, impact and sustainability.

Note: Quality at entry may also be a factor of the extent to which the intervention design is evidence- informed, including the involvement of stakeholders in the conceptualisation of the intervention.

Related terms: evaluability, ex-ante evaluation

Randomised Control Trial (RCT)

A type of evaluation that randomly assigns access to the intervention, in order to control influencing variables and limit bias, generating internally valid estimates of results.

Reach

The people affected by an intervention, as a subset of the total population. Related terms: people who benefit, target group, population

Real-time evaluation (real-time learning)

A process that provides immediate (independent) evaluative evidence, insights and feedback to inform decision making, learning and implementation while the intervention is underway.

Related term: monitoring

Recommendations

Proposals aimed at enhancing the relevance, coherence, effectiveness, efficiency, impact or sustainability of the intervention; at redesigning the objectives; or reallocating resources. Recommendations should be based on findings and conclusions.

Relevance

The extent to which the intervention objectives and design respond to beneficiaries, global, country, and partner/institution needs, policies, and priorities, and continue to do so if circumstances change.

Note: “Respond to” means that the objectives and design of the intervention are sensitive to the economic, environmental, equity, social, political economy, and capacity conditions in which it takes place. “Partner/institution” includes government (national, regional, local), civil society organisations, private entities and international bodies involved in funding, implementing and/or overseeing the intervention. Relevance assessment involves looking at differences and trade-offs between different priorities or needs. It requires analysing any changes in the context to assess the extent to which the intervention can be (or has been) adapted to remain relevant.

Reliability

Consistency or dependability of data and evaluation judgements, with reference to the quality of the instruments, procedures and analyses used to collect and interpret results and evaluation data.

Note: Evaluation and results information is reliable when repeated observations using similar instruments under similar conditions produce similar results.

Results

The outputs, outcomes or impacts (intended or unintended, positive or negative) of an intervention. Related terms: outputs, outcomes, effects, impacts

Results chain

The causal sequence of an intervention that stipulates the different stages leading to the achievement of the desired objectives. In general, the results chain starts with inputs, which then link to activities and outputs, and culminate in outcomes, and impacts. In some cases, reach is included as part of the results chain.

Related terms: inputs, activities, outputs, outcomes, impact, results framework, theory of change, performance measurement framework

Results framework

Explicit articulation (typically, in a graphical or tabular manner) of how a strategy or intervention will achieve the objective(s), including causal relationships and underlying assumptions and risks.

Note: Generally, includes indicators (with baseline, data source, means of verification, etc. for each) for the full results chain: inputs, activities, outputs, outcomes and impacts. In some instances, results frameworks only describe the desired outputs, outcomes and impacts, leaving flexibility to define the inputs and activities that will lead to these results.

Related terms: results chain, logical framework, theory of change, intervention logic.

Results information

Qualitative or quantitative data about the outputs, outcomes or impacts (intended or unintended, positive or negative) of an intervention.

Related terms: indicator; result, performance indicator

Results-Based Management (RBM)

A management strategy focusing on performance and achievement of outputs, outcomes and impacts. This management approach provides the framework, tools and guidance for strategic planning, risk management, performance monitoring, evaluation and knowledge management. It serves four complementary purposes: decision-making, learning, accountability and communication.

Note: RBM is often a participatory and team-based management approach designed to improve programme delivery and strengthen management effectiveness, efficiency, learning and accountability.

Review

An assessment of the performance of an intervention, periodically or on an ad hoc basis.

Note: Sometimes the terms “review” and “evaluation” are used as synonyms. However an evaluation is generally a more systematic and comprehensive assessment than a review. Reviews tend to emphasise operational aspects.

Related term: evaluation

Risk analysis

An analysis or an assessment of factors that affect or are likely to affect the achievement of an intervention’s objectives. A detailed examination of the potential unwanted and negative consequences to human life, health, property, or the environment posed by an intervention; a systematic process to provide information regarding such undesirable consequences; the process of quantification of the probabilities and expected impacts for identified risks.

Rubric

A framework that sets out criteria and standards for different levels of performance and describes what performance would look like at each level, as a way of answering evaluative questions. It also includes guidance on how to synthesise evidence to form an overall evaluative conclusion.

Sample

A subset of a given population which is chosen in such a way as to allow extrapolation of findings to the population. Note: Extrapolation can be done statistically for random sampling and analytically for purposeful sampling. Related terms: counter-factual, population

Scope of the evaluation

The time period, funds, geographical area, target group(s), organisational set-up, implementation arrangements, policy and institutional context, and other dimensions covered by the evaluation. Note: The scope of the evaluation should be delineated along at least four levels: operational (all or part of the areas of intervention, one or more related policies or interventions), institutional (all or part of the

authorities), temporal (period taken into account) and geographic (one or more territories or parts of territories, a specific, region, city, place, etc.).

Sector programme evaluation

Evaluation of a cluster of interventions within one country or across countries, all of which contribute to the achievement of a specific goal.

Note: a sector includes development activities commonly grouped together for the purpose of public action such as health, education, agriculture, transport, etc.

Self-evaluation

An evaluation of an intervention by those who are responsible for its design and delivery. Related term: internal evaluation

Social impact evaluation (social impact assessment)

An evaluation that determines (ex-ante or ex-post) the likely or actual social effects of an intervention. Involves the intended and unintended social consequences, both positive and negative, of interventions and any social change processes affected by the intervention.

Stakeholders

Agencies, organisations, groups or individuals who have a direct or indirect interest in the intervention or its monitoring and evaluation.

Strategic evaluation

An evaluation that examines how priorities are formulated, looking at overall systems and trends affecting performance and results.

Summative evaluation

A study conducted at the end of an intervention (or a phase of that intervention) to determine the extent to which anticipated outcomes were produced. Summative evaluation is intended to provide information about the worth or significance of the intervention.

Sustainability

The extent to which the net benefits of the intervention continue, or are likely to continue.

Note: Includes an examination of the financial, economic, social, environmental, and institutional capacities of the systems needed to sustain net benefits over time. Involves analyses of resilience, risks and potential trade-offs. Depending on the timing of the evaluation, this may involve analysing the actual flow of net benefits or estimating the likelihood of net benefits continuing over the medium and long term.

Target

An objective, usually quantitative, defined as a value on an established indicator. The target is generally set at the beginning of an intervention and is expected to be achieved by a specific point in time with available resources. Related term: performance indicator

Target group

The specific individuals, communities or organisations that the intervention is intended to reach.

Note: Can also be defined as the recipients of the goods and services produced by the intervention, or whose skills or capacities have changed because of the intervention. The target group may or may not be the individuals or organisations that, ultimately, are intended to benefit from the intervention.

Related term: reach

Terms of reference

Written document presenting the purpose and scope of the evaluation, the methods to be used, the standard against which performance is to be assessed or analyses are to be conducted, the resources and time allocated, and reporting requirements.

Note: Two alternative expressions sometimes used are “scope of work” and “evaluation mandate”.

Thematic evaluation

Evaluation of a selection of interventions, all of which address a specific sustainable development priority or topic, that cuts across countries, regions, and sectors.

Note: Often thematic evaluations will examine a strategic approach or priority topic across a variety of interventions. An example would be evaluating the extent to which the rights of people with disabilities were advanced across a portfolio of interventions in education, health and employment.

Theory of change

The way the intervention is expected to achieve or achieves change. It represents how people understand change to occur in a given context, including explicit (or implicit) assumptions about the causal links between inputs, activities and results. Often also includes evidence and risks for these elements of the results chain.

Related terms: intervention logic, logical framework, results chain, assumptions

Transformative effects

Holistic and enduring changes in systems or norms. Related term: impacts

Triangulation

The use of three or more theories, sources or types of information, or types of analysis to verify and substantiate an assessment.

Note: Seeks to overcome the bias that comes from single informants, single methods, single observers or single theory studies by combining multiple data sources, methods, analyses or theories, monitoring and evaluation.

Validity

The extent to which an evaluation is logically and factually sound. Also used to describe the quality of data collection strategies and instruments to accurately measure what they purport to measure.

-—————– END ————————–

AI Prompt Engineering

Wed, 10 May 2023 00:00:00 +0000

1. Start with an instruction

Try making your instruction more specific. e.g: Suggest one name for a black horse.

2. Add some examples

It’s helpful to both show and tell the model what you want. Adding examples to your prompt can help communicate patterns or nuances. Try submitting this prompt which includes a couple examples.

Suggest three names for an animal that is a superhero.

Animal: Cat
Names: Captain Sharpclaw, Agent Fluffball, The Incredible Feline
Animal: Dog
Names: Ruff the Protector, Wonder Canine, Sir Barks-a-Lot
Animal: Horse
Names:

Nice! Adding examples of the output we’d expect for a given input helped the model provide the types of names we were looking for.

3. Adjust your settings

Prompt design isn’t the only tool you have at your disposal. You can also control completions by adjusting your settings. One of the most important settings is called temperature.

You may have noticed that if you submitted the same prompt multiple times in the examples above, the model would always return identical or very similar completions. This is because your temperature was set to 0.

Try re-submitting the same prompt a few times with temperature set to 1.