In the product
The builders and the ledger
Metroliq is AI-native: the datasheet becomes a specification, the specification becomes a calibration form, the failure becomes an investigation and a CAPA, the old export becomes a mapped import — as drafts, on a ledger, for a qualified person to approve.
Nothing is signed by a model. Every AI output carries the model, its version, the prompt hash, the output, the citations, who approved it and whether it was edited. The certificate states when a specification was AI-assisted — and only when it was.
Datasheets and manuals into typed tolerances with provenance and verification tiers, plus a product library that grows as it is used.
Five-facet measurement forms drafted from the specification.
Studies proposed from published guides with the measurement equation.
Out-of-tolerance investigations grounded in reverse traceability; CAPA drafts; generic QMS record drafts.
Column mapping for spreadsheets and legacy exports, shown before loading.
Your old certificate PDF rebuilt as a template bound to live data.
Field mapping proposed for the ten-system catalog and REST/CSV.
Weekly work plans as narrative; management review drafts; training quizzes.
Answers from a maintained product guide plus a live snapshot of your workspace. Read-only by design.
It is labelled as AI-drafted, it is not in force until a person approves it, and the approval is a signature.
Feature, model and version, prompt and its SHA-256, output, citations, approver, edited flag. The assistant's answers are logged the same way.
When a specification came from a manufacturer datasheet with AI help, the certificate states it. When the origin is not the manufacturer, the flag is cleared and nothing is claimed.
Uncertainty, TUR, guard bands, PFA, intervals and verdicts are computed by deterministic code with its own test suite. The AI writes; it does not calculate.
Drafting runs on Anthropic Claude models through a gateway service, with the model and version recorded on every ledger entry so a draft can always be traced to what produced it. Prompts include only the record in context — the datasheet, the failure, the export — never another tenant's data.
The assistant is read-only. It explains and points; it never changes a record. The product guide it answers from is maintained by us; the live numbers come from your workspace at the moment you ask.
If your policy does not allow AI drafting for a class of record, the builders are optional per workspace — the deterministic engine, the signatures and the audit trail do not depend on them.
The tour runs the working platform, not a mockup. Pilot on your own instruments with your own imported history — we set it up with you, and fixes ship in hours and days, not release cycles.