Version boundary · July 2026
v1.0-candidate makes the reasoning inspectable. It does not make it validated.
This page separates artifacts that exist now from empirical claims that remain open. The distinction is part of the public model contract.
What exists now
Current evidence
- 40 public records and 83 paths with mandatory explicit causal steps.
- Twelve case-specific metric rationales and a base-to-final decision trail for every path.
- Stable path, step, and source identifiers in every machine-readable record.
- Explicit dominant and co-dominant semantics with no array-order fallback.
- One shared scoring engine for guided, expert, test, and documentation surfaces.
- Automated corpus, explanation-quality, scoring, route, accessibility-contract, and build checks.
What remains open
Open validation work
- Zero independent path reviews are recorded in the public corpus.
- Inter-rater agreement has not been measured.
- Construct validity against recovery burden, expert triage, or operational outcomes has not been established.
- Calibration and outcome-prediction claims are out of scope for this release.
- Independent scoring and validation studies remain planned open work.
This candidate label is not an accuracy claim. The rationales make the stored judgments understandable and challengeable; they do not independently validate the path, value, or band.
Release history
Public version record.
v1.0-candidate
Model-first corpus schema, explicit dominance, path-specific metric explanations, shared scoring engine, evidence-traceable public records, accessible custom controls, executable review workflow, and machine-readable outputs.