← Work
Public package

Dailies

Trust-aware release decisions for AI changes.

Year2026
RoleProduct · Engineering
StackTypeScript · AI Evaluation · CLI
The problem

AI releases can pass ordinary software tests while their behavioral evidence is incomplete, untrusted, or scoped to the wrong population. A binary gate then forces infrastructure failure and product failure into the same answer.

The system

Dailies runs or coordinates evaluation over declared evidence scopes, preserves trust and incompleteness, applies customer-owned policy, and emits an auditable promote, block, or inconclusive decision.

Try it
npx dailies@latest --config dailies.config.json
System properties
  1. 01Tri-state decisions prevent missing or unverifiable evidence from becoming an accidental approval.
  2. 02Every decision is bound to an exact evidence scope, input identity, and visible trust class.
  3. 03Customer policy keeps mandatory, blocking, advisory, and explicitly compensatory criteria distinct.
  4. 04Deterministic reports retain coverage, retries, evidence operations, and decision precedence for audit.