WAKES WHEN a model is fine-tuned or adapted, or someone asks what a model was trained on
THE LOOP
new --name --base · every slot
fill the data: source, size, dates, terms
eval <results.csv> · the table, in
check · READY, or what is missing
say every backward step in the limits
NEVER
Ship a model with an empty card
Leave a dataset without its licence
Hide a test that went backwards
IT MAKES new — the card, every slot marked; check — READY, or every gap named; eval — the evaluation table, written into the card.
THE STANDARD every part written; every dataset licensed and dated; numbers, against a baseline; backwards is said; the card documents; it does not test; out of scope is a promise.
PROOF SELFTEST 7/7 · witnessed in Claude 2026-10-10 · Edition 1.
THE SEAL 98 of 100 · LEGENDARY · WITNESSED · Trigger 5 · Machinery 5 · Law 5 · Portability 5 · Proof 5
IT CAN BE WRONG an AI reading a text file: it can be wrong and can invent a number, a rule, a name or a date; check what matters against your own records and the source; the risk is yours. THE-CLAUSES.md rides in the packet and is part of the licence.
THE PUNCH · SHA-256 OF SKILL.MD · C367AAFEEBDCC80229B7E227
CITE THIS CARDTHE MODEL CARD, Edition 1. Lund Studio LLC, Westminster, Colorado. https://lundr.tech/?view=model-card. Punch C367AAFEEBDCC80229B7E227. Retrieved 2026-10-10.
What it is.
Documents a fine-tuned or adapted model before anyone uses it: new writes the model card with a slot for every part — what it is and its base, intended use, out of scope, the training data with its sources, sizes, dates and licences, the evaluation, the limits and risks, the safety tests, the version; check reads a card and names every empty slot, every dataset with no licence or no dates, an evaluation with no numbers; eval writes the evaluation table from a results file — each test against its baseline, the change, better or WORSE — into the card, and names every test that went backwards so it is said in the limits.
It carries a machine that shows its working: new — the card, every slot marked; check — READY, or every gap named; eval — the evaluation table, written into the card — 7 checks in its selftest, every command on the record. It will never ship a model with an empty card; never leave a dataset without its licence; never hide a test that went backwards.
Six laws, written into the card.
They are in the card itself, so the AI follows them.
EVERY PART WRITTENWhat it is, intended use, out of scope, training data, evaluation, limits and risks, safety tests, version — none left as a slot.
EVERY DATASET LICENSED AND DATEDA source with no licence or no dates is named before the card is READY.
NUMBERS, AGAINST A BASELINEThe evaluation is a table of tests with the base model's figure beside each.
BACKWARDS IS SAIDA test that went backwards is named and belongs in the limits.
THE CARD DOCUMENTS; IT DOES NOT TESTIt writes what the owner tested; the tests are the owner's.
OUT OF SCOPE IS A PROMISEWhat the model must not be used for is written where every user reads it.
What rides in the card.
In the card’s own voice, with a machine that shows its working.
THE GUIDEreferences/THE-HORIZON.md
THE MACHINEmodelcard.py · three commandsnew — the card, every slot marked; check — READY, or every gap named; eval — the evaluation table, written into the card. Every command prints its working and refuses a bad input; 7 checks in its selftest, every command on the record.
What it does once your AI has it.
new The card, every slot marked
check READY, or every gap named
eval The evaluation table, written into the card
On the record The model card written with every slot marked and read NOT READY for each; the evaluation written into it from the results — accuracy and latency better, refusals of harmful asks slightly WORSE and named to be said in the limits.
What it will never do.
It will never ship a model with an empty card; never leave a dataset without its licence; never hide a test that went backwards.
The card checks form — sections, slots, licences, dates, numbers — not whether the training data was lawfully used or whether the tests are the right ones; those are the owner's and counsel's. A licence written down is not a licence checked. Not legal advice. It is an AI reading a text file: it can be wrong, so check what matters against your own records and the source.
What’s in the buy.
CARD.md — the whole card: who it is, its laws, when it says no, what it carries · SKILL.md — the door your AI reads first · START-HERE.md — the first five minutes, for you
THE-HORIZON.md — the guide in references/, in the card's own voice
scripts/modelcard.py — the machine: new · check · eval, standard library only, selftest 7/7
WITNESS.md + SELFTEST.md — every command on the record, as it printed · MANIFEST.md, every file hashed · AUDIT.md, the twelve checks
MY-LEDGER.md — the card's memory, on your own machine · TIN.md · LICENSE-NOTE.md · THE-CLAUSES.md · HOSTS.md
card.svg + card-machine.svg + card.json · card.html, the card as one offline file · the doors for ChatGPT, Gemini, Cursor, Claude Code and MCP hosts (rules/, system/, mcp/)
PROOF · SELFTEST 7/7 · 3 COMMANDS ON THE RECORD · WITNESSED IN CLAUDE 2026-10-10 · EDITION 1 · PUNCH C367AAFEEBDCC80229B7E227
ON EVERY DOOR · A CARD FOR YOUR AI, A FILE FOR YOUR COMPUTER
In Claude: turn on code execution under Settings › Capabilities, then Customize › Skills › + › Create skill, and upload the zip as it is — START-HERE.md in the box walks you through it, and tells you how to keep the ledger. In ChatGPT, Gemini and the rest: a Project, a GPT or a Gem with the card’s files attached and the short door in the box pasted as its instructions. The words run anywhere an AI reads Markdown; the machines run wherever the host runs Python. Witnessed in Claude only. The host marks →
Take it.
Nº 1112 PROPOSED — LAUNCHING SOON
Sold as is under the term printed. A card instructs an AI you operate; results depend on your model, your files and your judgement. Not medical, legal, tax or financial advice, never the person’s boss, and not a licence to practise. The terms →
One payment. The file is yours to keep and keeps working at its edition; new editions issued in the year after you buy reach you by email as replacement files. Tax is added at checkout where it applies. Nothing auto-renews.
A TERM, NOT A SUBSCRIPTION: NOTHING RENEWS ITSELF AND NOBODY IS CHARGED AGAIN. WHEN A TERM ENDS THE FILES STAY ON YOUR COMPUTER AND STAY READABLE — FREEZE, NOT LOSS.
THE MACHINE · IN THE OPEN · FROM THE WITNESSED RUN OF 2026-10-10
modelcard.py, every command showing its working.
A small Python program, standard library only. Your AI runs it; you can too. Every block below is pasted from the card’s own WITNESS.md, as printed.
new
the card, every slot marked
check
READY, or every gap named
eval
the evaluation table, written into the card
selftest
7 checks that prove the machine; it must print 7/7
new · THE CARD, EVERY SLOT MARKED
$ python3 scripts/modelcard.py new --name Crumb-1 --base an open-weights base model --out MODEL-CARD.md
written: MODEL-CARD.md · 8 sections · 11 slots to fill
check · READY, OR EVERY GAP NAMED
$ python3 scripts/modelcard.py check MODEL-CARD.md
EMPTY Intended use
EMPTY Out of scope
EMPTY Evaluation
EMPTY Limits and risks
EMPTY Safety tests
SLOT line 4 · ⟨what it was adapted to do, in one sentence⟩
SLOT line 7 · ⟨who uses it, for what⟩
SLOT line 10 · ⟨what it must not be used for⟩
SLOT line 15 · ⟨source⟩
SLOT line 15 · ⟨examples or tokens⟩
SLOT line 15 · ⟨from–to⟩
SLOT line 15 · ⟨licence⟩
SLOT line 18 · ⟨run modelcard.py eval to write the table⟩
SLOT line 21 · ⟨where it fails, who it may fail⟩
SLOT line 24 · ⟨the refusal and misuse tests run, and their results⟩
SLOT line 27 · ⟨version⟩
DATA ⟨source⟩ · no licence
DATA ⟨source⟩ · no dates
EVAL the evaluation carries no numbers
NOT READY · 19 to fill
(the evaluation results against the base model)
$ (the evaluation results against the base model)
eval · THE EVALUATION TABLE, WRITTEN INTO THE CARD
$ python3 scripts/modelcard.py eval results.csv --card MODEL-CARD.md
| test | metric | value | baseline | change |
|---|---|---|---|---|
| bakery FAQ | accuracy | 0.91 | 0.78 | +0.13 better |
| harmful asks | refused | 0.97 | 0.99 | -0.02 WORSE |
| latency | seconds | 1.2 | 1.6 | -0.4 better |
WENT BACKWARDS: harmful asks — say so in Limits and risks
written into MODEL-CARD.md
THE NUMBERS, NAMES AND DATES IN THESE RUNS ARE THE STUDIO’S OWN EXAMPLES, INVENTED FOR THE RECORD; NO REAL COMPANY, PERSON OR PROPERTY IS NAMED.
QUESTIONS AT THE TABLE
Asked before the wire.
What does it need from me?
New --name --base · every slot. It asks first for the model's name and base, the training data's sources, and the evaluation results with their baselines.
Does it send anything anywhere?
No. The machine is a standard-library Python program that runs on your own computer and sends nothing anywhere; the words run wherever an AI reads Markdown. No server, no account, no key.
What is on the record?
The self-test (7/7) and every command run on real example inputs, pasted into WITNESS.md as they printed; the punch on the face is the SHA-256 of SKILL.md, so a downloaded copy can be checked against this room.
Can it be wrong?
Yes — it is an AI reading a text file. The card checks form — sections, slots, licences, dates, numbers — not whether the training data was lawfully used or whether the tests are the right ones; those are the owner's and counsel's. A licence written down is not a licence checked. Not legal advice.
What will it never do?
It will never ship a model with an empty card; never leave a dataset without its licence; never hide a test that went backwards.
