Changelog

0.1.0 — 2026-07-21

• Created AI-ROS Bootstrap Kit.
• Added initial Constitution, REP integration, prompts, templates, mobile setup, and shortcut specifications.

0.1.1 — 2026-07-22

• Imported and normalized 50 source files from input-documents/.
• Moved handbook, course, research program, and research relay documents into canonical repository locations.
• Promoted the bootstrap README, roadmap, state, governance, prompts, templates, and mobile docs into the repository structure.
• Archived the derived Chapter 1 ZIP package and preserved import audit records.
• Removed the duplicate REP source copy from the bootstrap kit after hash verification.

Unreleased

• Added eleven proposed ROS backlog missions covering RFR-001 through RFR-010 plus the NX-005 audit/no-audit counterfactual, with dependencies, evidence, success criteria, stop conditions, and generated registry entries.
• Added a versioned ROS historical-attribution schema, eight-record pre-install work backfill, human reconstruction index, accepted compatibility decision, validator/test coverage, and a single authoritative current-state policy.
• Renamed 45 Markdown-formatted .txt research, semantic-control architecture, and Time Entry documents to .md, repaired repository references, and made the canonical state-constrained corpus discoverable by Research Publisher.
• Extended the RFR-009 repository-integrity validator with metadata coverage by artifact class, declared relationship resolution, generated/dependency-tree exclusions, three new calibration probes, and a schema 1.1 inventory; recorded the 131-document baseline without enforcing a premature legacy migration.
• Reorganized the tracked state-machine-documents/ corpus into the canonical research, architecture, and Time Entry project areas; preserved the complete 01–12 prompt/report series, separated its synthesis and experiments, and consolidated two SHA-256-identical duplicate files.
• Added navigation indexes for state-constrained research, semantic-control architecture, and Time Entry, plus a repository migration decision and validation record.
• Added a Node 24 research-publisher dependency, reproducible npm scripts, and a GitHub Actions workflow that builds and deploys the research site to GitHub Pages on pushes to main or manual dispatch; non-site fixtures, templates, and colliding legacy split documents are excluded from the published corpus.
• Executed the AI Research Mission Generator and added a current state-of-field REP, knowledge-gap analysis, research roadmap, and priority matrix.
• Selected long-horizon agent evaluation and verification as the highest-value research mission.
• Superseded the generator at prompts/AI-Research-Mission.md with a ready-to-run empirical research prompt.
• Recorded the execution and supersession decision under docs/repository/.
• Started the highest-priority research mission and completed Evaluation Research Cycle 001.
• Added the experimental research/evaluation/ workspace with a charter, candidate task suite, evaluation specification, evidence/hypothesis/experiment registries, failure taxonomy, immutable cycle report, and next-agent handoff.
• Recorded the decision to audit task integrity before running capability baselines.
• Completed provisional ET-004 and ET-014 task-integrity pilots with seven versioned calibration probes.
• Added a dependency-free evaluation CLI that validates contracts, calibrates deterministic graders, and exports blind disposable fixtures.
• Added current results, a minimum-sufficient-evaluation decision framework, threat model, system architecture, teaching guide, and evaluation roadmap v2.
• Revised the program roadmap so containment, telemetry, pause, and rollback precede adversarial or long-horizon baselines.
• Recorded Cycles 002 and 003, including negative results, unresolved hypotheses, and the independent-review blocker.
• Canonicalized root README, current-state, and roadmap paths by removing identical case-only Git entries while preserving their history.
• Added a repository fitness check for case-colliding tracked paths.
• Added the canonical research/frontier/ analysis with ten evidence-traceable RFRs, document-level frontiers, repository health metrics, and machine-readable index/dependency graph.
• Recorded the frontier scope and semantic-canonicalization decision under docs/repository/.
• Advanced RFR-009 to Validation with a dependency-free stable-identifier, explicit local-link, and frontier graph/index integrity validator.
• Added four seeded integrity tests, integrated the checks into evalctl.py repo-audit, and preserved the first machine-readable inventory under data/validation/.
• Completed the repository-wide non-human experimental evidence review.
• Added a seven-item master inventory, machine-readable experiment matrix, lineage/dependency graph, comparability and quality analysis, contradiction registry, failure taxonomy, cumulative findings, and ranked research backlog under research/analysis/.
• Added the canonical comparative-review REP and repository structural decision while preserving historical evaluation records unchanged.
• Recorded that current evidence contains zero independent experimental replications and zero agent capability trials; the 7/7 result is an in-sample regression check over dependent designer probes.
• Added an executable next mission for independent pilot review and a frozen grader challenge on blinded unseen outcomes.