v0.5.34 — Evaluation Roadmap v1.0
Phase 90 strategic spine. Alton's Decisions #1 + #2 from the May 13 Recursive Paradox session shipped as a bundled release.
What's new
/research/evaluation-roadmap— pre-registered Evaluation Roadmap v1.0, year-one window May 2026 → April 2027. Ten dated milestones (M0–M9), pre-registered numerical thresholds (Cohen's κ, ECE, Brier, F1 lift, ESR), 8 pre-registered kill-conditions, named external witnesses for M3 / M5 / M7 / M8.- Companion to
/research/paradox— both pages cross-linked bidirectionally. The Validation Paradox names the problem; the Evaluation Roadmap is the structural answer. - Canonical markdown at
docs/research/evaluation-roadmap/roadmap-v1.0.mdwith full Section B technical RFC appendix (math, dataset spec, reproducibility checklist, co-maintainer terms). - Tagged
roadmap-v1.0— any future change to a milestone date or definition requires a new tag. - README +
/changelogupdated;_RESEARCH_INDEXJSON-LD bumped to v1.4.
Why
The Validation Paradox page ended on a single line — the only available exit from a closed validation loop is an external signal. This release ships the external signal: a public clock with pre-registered failure conditions. Git tags make silent edits visible. Milestone-keyed retrospectives make silent skips visible. Named external witnesses make false completions visible. Pre-registered failure conditions make rationalization visible. Failure numbers will ship in the same font size as success numbers.
Per Alton's May 13 ruling: Beta v0.6.0 is now redefined as M0 + M1 from this roadmap (NOT first paying customer). Credibility milestone.