Open evaluation grids for AI projects
An examination to put an AI project through before you launch it. It shows what the file does not say, and leaves a written decision you can defend in the room — in 45 minutes, on your own case.
Read the frameworks (open access)
Published, timestamped and refutable — the testing protocol is public. · CC BY-SA 4.0 · Zenodo DOI · OSF preregistered · INPI priority (e-Soleau) · ORCID
“Compliance applies to a project already chosen. It does not say whether it should have been chosen — that is what these grids are for, before committing the money, the teams, the reputation.”
One file, five ways in — depending on what you have to decide.
The review leaves a written record: the grid filled in, the warning signals, the reasoned decision. That record serves again — it documents your corporate social responsibility (CSR) and feeds your European sustainability report (CSRD) or your fundamental rights impact assessment, required by the AI Act, the European law on AI.
A reusable building block — neither a certification nor a label.
One reading answers one question: does this project hold? What BRS and HUMICS add is a common language — the same criteria, the same words, the same rules. That is what makes forty readings comparable: which projects hold, which ones we stop, and why, across countries and teams.
The output of a reading is not a score: it is the record of what was examined — including what raised no issue at all. That is the part an auditor, a regulator or an investor asks for two years later.
R2.BRS
penalty if the team falters
not triggered
R1.HUMICS.P2
a zero caps the pillar
not triggered
J1
viability / alignment gap
TRIGGERED
GARDE_FOU.ASCENDANT
the committee departs upward
TRIGGERED
A real excerpt from a reading record: the findings are reproduced word for word; the italic labels are added here for readability. The rules applied are published in the frameworks, and the record carries their exact version.
The frameworks are free: your teams can pick them up without asking anyone. At this scale the work lies elsewhere — training your rapporteurs to score with a fact behind every rating, calibrating the weightings for your sector, and reviewing a sample so your internal readings hold up in front of a third party.
Discuss it for your organisation →
You write directly to the author of the frameworks. A first conversation, no commitment.
Business Readiness Score
Is this AI venture commercially viable?
Technology & Product · Market & Positioning · Business Model · Team & Governance · Execution & Roadmap
0–5 score (BRS-A adaptive / BRS-N neutral) · Quick Scan or Full Review · anchored in observables, N/S rule for missing data.
Human alignment framework
Is this AI deployment humanly aligned?
Technological Ethics · Human & Social Impact · Responsible Innovation · Environment & Sustainability · Financial Sustainability
Reading tiers · HUMICS-A / HUMICS-N.
The picture to decide
Four BRS tiers × four HUMICS tiers: sixteen crossings, five decision classes. Each cell keeps its professional term — and its plain-language translation.
ALIGNMENT (HUMICS)
The matrix is a deliberation framework: it structures the discussion, it does not replace judgment. The committee decides, not the grid: it may depart from the indicated class, provided it gives reasons — the matrix then makes that choice traceable and defensible. The sixteen cells and the joint rules appear in the document “The viable + aligned matrix” (open access).
45 minutes, free. It orients, it does not decide.
On file, in committee: every pillar checked against observables.
The two readings meet in the matrix: one picture to decide.
At every step, the decision remains yours.
Try it
BRS and HUMICS are two open grids for examining an AI project: can it exist? should it exist this way? This mini-scan gives a preview of them — six markers, one minute, on an idea or a case you know. It orients, it does not decide.
Someone would pay today for what the project promises.
The business model is sketched: what it costs, what it brings in.
The team can ship a usable first version within six months.
The people affected would know an AI takes part in the decision.
A human can contest and correct an output of the system.
The deployment would remain defensible if the press covered it tomorrow.
0 / 6 — answer all six markers to place the project in the matrix.
Move to the next step
A mini-scan orients. A full review (2–3 h) runs the entire grid through a double reading, locates the warning signals on YOUR case, and produces a documented, defensible recommendation. It starts with a first 45-minute examination, free of charge.
Book 45 minutes, freeThe case. In 2018, Reuters revealed that Amazon had abandoned an AI recruiting tool: it penalized résumés containing the word “women’s”.
What HUMICS would have seen, as early as 2014-2015 : no bias audit, no monitoring of discrimination in production, no right to an explanation. The safeguard was missing — spottable before the scandal, not after.
The case. A credit card whose limits were set by an algorithm. The New York regulator concluded there was no unlawful discrimination under credit law.
What the grids would have seen, as early as 2019 : HUMICS — a decision impossible for the customer to explain or contest. BRS — unproven product economics, carried by a bank new to consumer credit.
Neither a certification, nor a label, nor one more standard. Deliberation frameworks: open to criticism and improvement — that is why everything is published. And nothing that decides for you: you decide. The tool widens what you look at before deciding, and makes defensible what you decided.
Partly, yes — as a brick. Examining an AI project with the grids produces a traceable deliberation (grid, warning signals, reasoned decision) reusable for your AI governance and sustainability reporting: it feeds a fundamental-rights impact assessment (AI Act) and the governance / impact side of CSRD reporting, and overlaps several ISO 26000 subjects. BRS/HUMICS certifies no compliance and covers only your AI projects — an input, not an attestation.
Humanity, Intelligence, Collaboration, Synergy: the name states the ambition of the framework — to assess whether an AI deployment aligns with what holds a society together, not merely its compliance.
No. These are open reference frameworks; there is no certifying body.
They are different layers, and they add up. The European AI Act sets the legal obligations; ISO/IEC 42001 organises AI management within the organisation; ISO/IEC 42005 (2025) describes how to document a system’s impacts — and asks each organisation to set its own thresholds and scales, providing none. BRS and HUMICS are one such framework: sixty sub-criteria, three named anchors each, contestable point by point. None replaces another.
No. The matrix proposes a reading; the decision stays with the committee, which may depart from it with stated reasons. The tool then makes that choice traceable and defensible — it is a deliberation framework, not an automaton.
They make no such claim. The validation agenda is falsifiable and preregistered on OSF; refutations are invited.
Yes, under the CC BY-SA 4.0 license (attribution + share-alike). For embedded commercial use (proprietary tools, white-label) or a partnership, a dedicated license is available — get in touch. “BRS Business Readiness Score” and “HUMICS” are the subject of a trademark filing.
Anchored observables and an N/S rule for missing data; two readings, -A (adaptive) and -N (neutral).
The protocol is open on OSF (osf.io/avhw8); get in touch for feedback or co-authorship.
Everything is public: the grids, the matrix, the casebook, the test protocol. Refutations will be published as readily as confirmations. A substantive critique to raise? Write to me: david.desroses@brs-humics.com
A real case to put through the grid, or a method to roll out across your teams? Let’s talk.
Book 45 minutes, freeDavid Desroses, independent researcher — L&L AI Agency
answers from the published documents