# IRF for NGOs: take it to your own AI

Implementation Readiness Framework for NGOs, published in association with the Nordic Blockchain Association. Pack version 1.0.0.

## What this is

A folder of plain files. You put them in the AI assistant your organisation already uses. The assistant then asks you about a problem you have, or reads a meeting transcript and asks only what is missing, and writes a one-page decision brief on whether blockchain fits that problem.

Often the answer is no. When it is, the brief says why and what to try instead.

## What it does

- It asks about you, about your problems, about which of 145 uses of blockchain might fit, and about whether your organisation is ready.
- It takes answers from a transcript or notes if you have them. It checks each guess with you before the guess counts.
- A script does the arithmetic, so the same answers give the same result each time.
- It writes the decision brief. After a yes it can also write a tender that any supplier can answer.

## What it does not do

- It does not run by itself. It needs an AI assistant that you already use and trust.
- It sends nothing anywhere. The files contain no address to send to and no tracking. The script reads two folders on your machine and prints to the screen.
- It keeps nothing. There is no account, no database and no copy of your answers on our side.
- It does not recommend a product or a company.
- It is not a legal, security or procurement review.
- It can be wrong. Your assistant can mishear a transcript or pick the wrong option. That is why it confirms every guess and why the numbers come from the script. Read the brief before you pass it on.

## Where your words go

You choose where your words go. The pack sends nothing. What you type or paste goes to the AI assistant you set the pack up in, under that assistant's terms and your organisation's settings. If the questions touch something sensitive, such as identity, land or protection cases, check with your data protection lead what may go into that assistant before you start. The pack marks the sensitive questions and lets you skip them.

## Set it up

The wording differs between products. In each one, create a project or a custom assistant, add the files from this folder, and paste the instruction.

1. Create a project in your assistant. In Claude and in ChatGPT Enterprise the word is project. Gemini, Copilot in M365 and others use their own word, such as Gem, agent or notebook. Pick the one that lets you add files and standing instructions.
2. Upload this folder: `QUESTIONS.md`, `USE-CASES.md`, `SCORING.md`, `TENDER-QUESTIONS.md`, `templates/`, `scorer/` and `data/`. If your product limits file size, upload the markdown files and the templates first. The catalogue in `data/` is large and only the script needs it.
3. Paste the whole of `INSTRUCTIONS.md` into the project's instructions. Then start a conversation, and paste or attach your meeting transcript if you have one.

If your assistant cannot run code, it says that the ranking is approximate. You can then run the script yourself on a laptop. It needs Python 3 or Node 18, and no other package:

```
python3 scorer/scorer.py --answers answers.json --data data > result.json
```

## Have your security team look at it first

The folder holds text, JSON and two scripts, `scorer/scorer.py` and `scorer/scorer.js`. Neither script opens a network connection or writes anything except the output you ask for. Your security team can read both in full. To check that what you hold is what was published:

```
sha256sum -c SHA256SUMS --ignore-missing
```

Run it in the folder that holds the zip file and `SHA256SUMS`. The checksum of the zip is also on the download page, and `SHA256SUMS` lists every file inside the pack. `VERSION.txt` names the source commit.

## What is in the folder

| File | What it is |
|---|---|
| `INSTRUCTIONS.md` | The instruction for the AI assistant. |
| `QUESTIONS.md` | Every question it may ask, in order, with the exact answers each one accepts. |
| `USE-CASES.md` | The 145 use cases in plain words, with their fit gates and examples. |
| `SCORING.md` | How the numbers and verdicts are made, written out. |
| `TENDER-QUESTIONS.md` | The questions that fill a tender, per use case. |
| `templates/` | The decision brief and the tender, as forms, and two example files. |
| `scorer/` | The script that does the arithmetic, in Python and in Node. |
| `data/` | The data the script and the questions come from. |
| `feedback-note.md` | Five prompts for an optional note back to us. |
| `LICENCE.md` | The two licences and the attribution line. |
| `VERSION.txt` | The version, the source commit and a hash of the contents. |

## Feedback

A note back is optional. `feedback-note.md` has the five prompts and the addresses. It asks for no organisation details.

## Contact

contact@nordicblockchain.com