1,593tests
The repository test suite runs with zero warnings.
MeasuredProof
Inspect the evidence. Keep the caveats.
Tasaab’s diligence figures are generated from the current repository and designed to fail when they drift. A small, scoped result is reported as exactly that.
4statutory modules
BC forestry, Washington forestry, Phase I ESA, and land development.
Measured38government GIS endpoints
Registered endpoints were checked against live service metadata on August 17, 2026.
Published39scored stream-buffer cases
Twenty exact matches, fifteen professional-discretion cases, four code-version differences, and zero observed engine defects.
MeasuredSource: Tasaab Engine data room. Endpoint verification is dated, never permanent. The 39-case result covers stream buffers only and is not a claim of universal correctness.
Regression evidence
The useful proof is what broke.
Real public data exposed errors that plausible-looking demos would never reveal. Each correction became a regression test.
A federal records search was pointed at brownfields only.
The registered EPA layer would have missed RCRA generators, Superfund sites, NPDES facilities, and TRI reporters without raising an error. Endpoint verification caught and corrected the layer.
A missing wetland table suppressed valid stream rules.
The refusal check was jurisdiction-wide. It now keys on jurisdiction and feature kind, preserving known stream rules while correctly refusing unresolved wetlands.
Adoption date and effective date were not the same.
King County Ordinance 20024 was adopted in December 2025 but took effect in February 2026. The engine now selects the table in force on the report date.
Blind comparison
39 cases. Every disagreement adjudicated.
Stream-buffer cases were extracted from professional reports and scored against the transcription in force on each report’s date. Disagreement was investigated instead of collapsed into a single accuracy percentage.
Known limits
What the engine does not claim.
A credible compliance system makes its boundary visible before a professional relies on it.
- 01
Washington riparian-width tables are machine-read and remain unconfirmed by a person.
- 02
Several Washington rules have not been transcribed and therefore refuse rather than infer.
- 03
Proprietary ASTM E1527-21 tables are not reproduced; public 40 CFR Part 312 requirements are used where applicable.
- 04
King County wetland buffers require field facts that cannot be resolved from spatial data alone.
- 05
Authentication, tenancy, managed hosting, and white-label SSO are not part of the current engine.
- 06
No language model is wired by default; the adapter is optional and prose-only.
Evaluate before you licence
Ground-truth it against your own work.
Start with one known filing. Compare sources, thresholds, omissions, and the final unsigned pack before discussing deployment.