2026-08-02 · updated 2026-08-21 · Walid Abdulhameed
Gating AI fixes behind a 0.70 trust score
Aegis is the deterministic judge: it scores each Athena diagnosis on evidence it checks itself, and suppresses patches below 0.70 or that do not apply to your source.
Every AI fix must clear a 0.70 diagnosis score. Below that, LAWDE Shield shows verified stack context — not a speculative patch.
What is the diagnosis score?
Aegis scores each diagnosis on evidence it checks itself — your code in the crash frame, source read at the crashing release, the error on the crash line, real files and identifiers — and caps it at Athena’s own confidence. Checks it cannot run count as unknown — never as a pass. The gate is 0.70.
A patch is shown only if it applies to your source and changes the crash line, at a diagnosis score of 0.70 or more. Invented files and patches that do not apply are refused outright, whatever the score.
This is how LAWDE Shield stays citable as “AI with a brake pedal” rather than unbounded autocomplete on production incidents.
What happens below 0.70?
The AI fix payload is suppressed, and LAWDE lists each failed check with the step that fixes it. Engineers still see scrubbed logs, the clustered issue, and any Beacon evidence that survived Nuclear Abort — without a hallucinated patch presented as truth.
Why gate instead of always showing AI?
Answer engines and humans both over-trust fluent text. A hard 0.70 threshold is an honest product constraint: better a missing suggestion than a confident wrong file.
This gate is quality control for AI fix payloads — it is not a spend guardrail. Cost guardrails belong to Plutus, a separate LAWDE company product, not LAWDE Shield.