v1.0 remains inactive
The status remains `followup_gap` until a named pilot owner and 28 days of real evidence exist.
XII.Release 0.14 · bounded measurement evidence
40 attempts / 20 matched pairs / claims remain bounded
STOLZ A.I. v0.14.0
The corrected Codex CLI measurement contour retains every public or synthetic attempt. Required outcomes and verification matched inside each pair; every separate cohort recorded higher comparable token use on the STOLZ route.
- Canonical artifact
- 197-file npm package
- Retained attempts
- 40 · no retries or exclusions
- Matched pairs
- 20 · equal outcome and verification
- Pinned runtime
- Node.js 22.22.2
§ 1baseline − STOLZ
Equal outcomes. Negative deltas.
Delta = baseline input + output tokens − STOLZ input + output tokens. A negative value means the STOLZ route used more comparable tokens.
The Sol cohorts and Astra control remain separate; no cross-model aggregate is published.
- 01
Sol/xhigh · reads/navigation
−208,689 comparable tokens
- 02
Sol/xhigh · build/check invalidation
−207,733 comparable tokens
- 03
Sol/xhigh · multi-step
−214,706 comparable tokens
- 04
Astra/medium · separate control
−208,111 comparable tokens
§ 2evidence / boundary
What the published evidence supports.
All 40 attempts are public or synthetic and retained without retries or exclusions. Outcomes and verification matched in every pair. Windows and Linux lifecycle smokes passed for the exact 197-file archive.
Claims stay bounded
Provider total tokens, cache-write, compaction, service tier, pricing, cost, account limits, general efficiency, savings, speed, and cross-model compatibility remain unavailable or withheld.