How the readiness gate works
A verdict of Ready, Needs review, or Not readyisn't a separate AI judgment layered on top of the analysis — it's a deterministic, auditable function of the findings the AI passes actually produce. This page explains exactly what feeds the gate and what each verdict means, so a "Ready" result is something you can trust because you can see how it was reached, not because you're taking the AI's word for it.
Four AI passes, one deterministic gate
Every document you upload goes through four independent AI passes, each producing structured findings — not free-text opinions. A separate, plain-code function then applies fixed rules to those findings to compute the verdict. This split matters: the AI extracts facts (does this document say X, is this signature present, is this date consistent), and ordinary code — the same every time, for every document — decides what those facts mean for readiness. The AI never decides the verdict directly.
Regulatory findings
Checks the document against the specific regulation/standard you selected, citing only requirements that actually appear in that framework's rubric — see the knowledge base for the exact source text each check is grounded in. Each finding gets a severity tier (critical/major/minor/best-practice) and a status (met/gap/partial/not-applicable).
Document integrity
Structure (are expected sections present), dates (extracted, then checked for logical ordering by a separate deterministic function — not the AI itself), and signatures (visually confirmed present or blank from the actual rendered page images, not just whether text mentions a name).
QMS metadata
Owner, approver(s), effective date, periodic review, version, and change summary — the fields a real QMS platform would require before letting a document go live. Certain document types (CAPA, Deviation, Change Control, Training Record) get an additional type-specific check, e.g. a CAPA needs a linked source event.
Proofreading
Spelling, grammar, and formatting-consistency issues. Advisory only — proofreading findings never affect the gate verdict, and everything is flagged rather than filtered, since a human reviewer, not the AI, is the one who should decide what's a real issue versus intentional jargon.
Exactly what makes a verdict "Not ready"
Any one of the following is enough to block a document, listed exactly as the gate function checks them — nothing paraphrased or simplified from the actual logic:
- Any regulatory finding marked critical severity and gap status (unmet).
- Any expected signature block that's visually blank.
- Any missing document structure elements (e.g. a Purpose, Scope, or Procedure section that's absent).
- No document owner/author identified.
- No approver identified.
- No effective date stated.
- Neither a periodic review date nor a review frequency stated.
- A non-first version with no change summary describing what changed.
- An unmet type-specific requirement for the document's type (e.g. a CAPA with no linked source event).
What makes a verdict "Needs review" instead
When none of the blocking conditions above are present, but at least one of these is, the document is downgraded from "Ready" to "Needs review" rather than passed cleanly:
- Any regulatory finding marked major severity and gap status.
- A date-consistency issue — an illogical ordering, an unparseable date, or an inconsistent date format.
- No document ID/number identified.
A document with zero blocking conditions and zero of these warning conditions is verdicted Ready. Minor and best-practice findings, and proofreading issues, are always shown but never affect the verdict either way — they're informational, not gating.
Why this design, not just an AI opinion
Most AI compliance tools either let a model reason freely over regulation text (which can hallucinate citations or apply inconsistent judgment run to run) or produce a vague relevance score with no traceable reasoning. Regscribe grounds every regulatory finding against a maintained rubric document — the same one shown in the knowledge base — and then applies a fixed, code-level rule set to decide the verdict from those findings, not a second round of AI judgment. The same findings will always produce the same verdict. You can read exactly why a document was blocked, and you can read the actual gate function's logic yourself if you want to — it isn't a black box.
None of this makes Regscribe's output a guarantee of passing a real regulatory audit — AI-generated findings can still be wrong, and the gate can only reason about what the AI passes report to it. See our Terms of Service for the full advisory-only disclaimer, and a real, redacted sample report to see this in action.