The published standard · AS-1.1.0
The Assay Standard
Every artifact Assay produces is graded against this page — and this page is generated from the same code that grades. What you read here is what runs.
“The standard does not bend for our own marketing.”
The pass rule — exact
allEvery hard check must pass. Any failure blocks or triggers repair.
≥ 72Weighted craft mean across the six axes.
≥ 60No single craft axis may fall below the floor.
Failing drafts are repaired at most 2 times — and every draft’s report ships in the dossier, including the failures.
The Standard in practice
Hard checks — deterministic · 15 laws
01Claim coverageCLAIM_COVERAGEEvery rendered sentence resolves to a confirmed claim backed by existing evidence, and every number appears in that evidence. Unsupported prose cannot ship. If a sentence has no confirmed claim behind it, it does not ship.
02Evidence resolvesEVIDENCE_RESOLVESEvery referenced evidence item exists, and file-backed evidence has readable content. A dangling or unreadable source is a hard failure. Every claim must point at evidence that actually exists in the ledger.
03Link livenessLINK_LIVENESSEvery URL in the artifact is fetch-checked and must resolve live. A dead or unsafe link never passes. A dead link never earns “Linked” — we fetch every URL and check.
04No placeholdersPLACEHOLDER_TEXTNo [BRACKETS], "YOUR X HERE", TBD, lorem, or XXX placeholder text survives into any artifact. No [BRACKETS], no “TBD”, no lorem. A placeholder is an automatic failure.
05Date sanityDATE_SANITYNo future dates, each end is on/after its start, tenure math is correct, and overlaps are flagged — all evaluated in the candidate's timezone. Timelines must be possible: no future starts, no overlapping impossibilities.
06Cross-artifact consistencyXARTIFACT_CONSISTENCYEvery number tied to a claim renders identically wherever that claim appears across the resume, letter, and stories. The cover letter may not contradict the résumé. One dossier, one truth.
07Format lawFORMAT_LAWThe ATS variant is single-column with no tables/images/text-boxes, uses only approved section headings, and stays within 2 pages; the designed variant meets a 4.5:1 body-text contrast ratio. Headings an ATS actually recognizes — structure that machines can read.
08DOCX integrityDOCX_INTEGRITYThe generated .docx reopens in a parser and contains the same section headings as the ATS resume. The .docx must open, parse, and mirror the ATS structure.
09ATS parse-backATS_PARSE_BACKThe ATS PDF is re-parsed by Assay's deterministic parser and diffed field-by-field against the source profile; 100% of required fields must survive. Verified against Assay's deterministic parser and ATS format law — not a simulation of any specific vendor. We re-parse the rendered PDF and diff it field by field against your profile.
10Contact validityCONTACT_VALIDITYThe contact email is well-formed and every contact link is a syntactically valid URL. Contact details must be present and well-formed, or the document fails.
11PII hygienePII_HYGIENEA share-view artifact exposes only the fields the candidate approved; unapproved personal data must be redacted. Nothing you did not approve for exposure leaves the dossier.
12JD keyword coverage (informational)JD_COVERAGEWe report weighted must/nice keyword coverage against the job description. We report coverage; we do not stuff keywords. This check never fails. Reported honestly against the brief — coverage is informed, never stuffed.
13Interview answer integrityINTERVIEW_INTEGRITYA finalized interview answer must be STAR-complete and every factual number must agree with the confirmed ledger; contradictions block the answer.
14STAR completenessSTAR_COMPLETENESSStory-bank entries must make situation, task, action, and result legible; a result without the path to it is not a complete interview story.
15Rendered portfolio contrastPORTFOLIO_CONTRASTPortfolio pages are screenshot-sampled in Chromium; rendered body text must measure at least 4.5:1 against its actual background.
Craft axes — Claude critic, scored 0–100
| Axis | Weight | What it grades |
|---|---|---|
| Voice | 1 | Confident, human, and specific — not generic AI prose. |
| Specificity | 1.5 | Concrete actions, tools, and outcomes rather than vague responsibilities. |
| Quantification | 1.5 | Impact is measured where evidence supports it — never invented. |
| Positioning | 1 | The strongest, most relevant evidence leads. |
| Tailoring | 1.5 | The artifact answers this specific brief, not a generic role. |
| Evidence honesty | 2 | Every claim is framed at the strength its evidence earns — no overreach. |
Repair policy
When a draft fails, the Tribunal issues a repair brief — the exact findings, in plain language — and the Forge rewrites under the same claim gate. At most 2 repairs; then the dossier ships with its honest final verdict. Reports are never discarded: the first draft’s failure is part of the record.
source: packages/tribunal/src/standard.ts · rendered via renderStandardMarkdown · regenerated every build