One reproducible number. Verify it yourself.
The Codebase Assurance Index is an open, reproducible 0–100 standard for the condition of a codebase. The algorithm, the lenses and the rubric are public, and the reference scorer is open source. You don't have to trust us — verify any number yourself.
Same commit + frozen rubric → the same number, every run. That invariant is the whole point of this site.
A sample delivery — dated, signed, verifiable. Same commit + frozen rubric reproduces this exact card.
Producers push. The registry holds. Consumers read.
Two applications close the loop around this standard — neither of them owns the number.
Watchdog produces
Watchdog scans a codebase, computes the CAI under the published rubric, and pushes a signed evidence package to the registry.
Produce a survey →cai holds & verifies
The registry stores signed CAI-delivery packages — dated, content-hashed, Ed25519-signed. Anyone with a package can check it against the standard.
The registry →Assay consumes
Assay turns evidence from the registry into decision reports — due diligence, procurement verification, consequences reads.
Commission a decision →- Watchdog producesscans the codebase, computes the CAI under the published rubric, signs the evidence package
- The registry holds & verifiesdated, content-hashed, Ed25519-signed deliveries — checkable by anyone against the standard
- Assay consumesturns registry evidence into decision reports — due diligence, procurement, consequences
Neither app owns the number — the standard does. That separation is what makes a shared survey worth trusting.
Open, reproducible, and versioned — by construction.
Open verdict
The algorithm, the ten lenses and the rubric are published here. The reference scorer is open source — the verdict is not a private formula.
How the CAI is computed →Reproducibility invariant
Same commit + frozen rubric → the same number. Run the open scorer over a survey's evidence and you get the same CAI — or you've found a discrepancy.
Verify a survey →Versioned & contestable
Any change that can move a score bumps the rubric version, and a scoring change not in the published spec fails CI. Every number stays re-derivable from a rule you can read.
Rubric versions →Check the work — that's what the standard is for.
Reproduce a survey from its evidence, or read exactly how the index is computed.
Producers → watchdog.canine.dev · Consumers → assay.canine.dev