Research and policy centres
A recommendation before a funder
Who Sijal is for
Investment committees
A deal before approval
Consulting firms
A recommendation before delivery
NGOs
Priorities before donors
Universities
Curriculum and planning committees
Boards of directors
A decision before shareholders
What you receive
The decision
A written ruling on the question you raised, with its reasons and the evidence it rests on. Written to be put before your committee, funder or board, so a reader sees what was decided and why.
What it means for you: a decision you can defend when someone asks why.
Case file and evidence
The references and documents the case rests on. Every claim is tied to its source, so a reader can check it and see how far the source supports it.
What it means for you: material an independent reviewer can examine without having to trust us.
Rulings of the six judges
Six AI judges rule one at a time, none seeing another's ruling. Each ruling carries its reasons and references, and you see where they agreed and where they differed.
What it means for you: you see the disagreement itself, not only its outcome.
Fixed record
The decision record cannot be changed once it is finalized: the question, the sources, the rulings and the decision. Return months later and see how the decision was reached.
What it means for you: an auditable trail for every decision.
The Sijal Observatory
Measuring verdict repeatability and analytic quality on completed sessions.
Figures are illustrative, shown to preview the design. They are not results.
- Verdict repeatability · Sijal 18 of 20 runs
- Verdict repeatability · Direct model 14 of 20 runs
- Analytic standards met · Sijal 29 of 36
- Analytic standards met · Multi-agent model 21 of 36
| Question | Ruling Issues decided | Verdict repeatability 5 runs | Baseline concordance | Analytic standards compliance ICD 203 |
|---|---|---|---|---|
| Should a university adopt one AI-use policy? Sample | 2 of 3 issues Support 2, opposing 1 |
Sijal
5 of 5 runs identical
Direct model (baseline)
4 of 5 runs identical
|
Concordant |
Sijal
8 of 9
Multi-agent model (baseline)
5 of 9
|
| Should a policy report be released as a draft for public comment? Sample | 2 of 3 issues Support 1, opposing 2 |
Sijal
4 of 5 runs identical
Direct model (baseline)
5 of 5 runs identical
|
Differs |
Sijal
7 of 9
Multi-agent model (baseline)
6 of 9
|
| Should the new initiative be funded from reserves or a grant? Sample | 3 of 3 issues Support 3, opposing 0 |
Sijal
5 of 5 runs identical
Direct model (baseline)
3 of 5 runs identical
|
Concordant |
Sijal
8 of 9
Multi-agent model (baseline)
4 of 9
|
| Should the organisation expand into a new region this year? Sample | 2 of 3 issues Support 1, opposing 2 |
Sijal
4 of 5 runs identical
Direct model (baseline)
2 of 5 runs identical
|
Differs |
Sijal
6 of 9
Multi-agent model (baseline)
6 of 9
|
● run matching the modal verdict · ○ divergent run · each segment is one of nine standards met
Modal verdict: the ruling most runs returned.
Sample data. Replaced by real session measurements when available.
Automated application of published criteria, not an official evaluation. Verdict repeatability: an approach inspired by Kahneman, Sibony and Sunstein (Noise). Analytic standards: ICD 203 (US Office of the Director of National Intelligence).