# Skill Evolution Ledger

| Date | Feedback | Skill rule added | Files updated |
|---|---|---|---|
| 2026-09-01 | Research page used a generic `index.html`, emphasized the workflow app over the technical study, and lacked algorithm/software components with code references. | Added a named web-insight artifact mode requiring descriptive filenames, readable workflow-first structure, staged animation, algorithm/software tabs, exact source paths, code excerpts, evidence labels, and invalidation tests. | `SKILL.md`, `reference/web-insight-artifact.md` |
| 2026-09-02 | Interactive component cards made step navigation visually primary and reduced the established input → module → output contract to a secondary strip. | Require component cards to preview input/output conditions and component explorers to keep inputs left, internal mechanism center, outputs right, with synchronized step highlighting and semantic mobile stacking. | `reference/web-insight-artifact.md`, `analysis/feedback-log.md`, `analysis/iq-training-evaluation.md` |
| 2026-09-02 | Skill lacked automated quality enforcement, IQ evaluation scoring, scaffolding tooling, and migration diff topology. | Added Python code engine (`guardian.py`, `evaluator.py`, `automation.py`, `engine.py`), `topology-migration-diff.md`, spec drift matrices, and `knowledge/manifest.json` indexing. | `scripts/*`, `SKILL.md`, `reference/core-loops.md`, `reference/quality-gates.md`, `reference/topology-code-repo.md`, `reference/topology-migration-diff.md` |
| 2026-09-02 | Distillation, web rendering, and evaluation were coupled; one generic website contract could not distinguish research, system, and agentic outcomes. | Added a multi-deliverable task contract, deterministic routing, optional Graphify detection, a categorized `web` builder skill with examples, and a versioned web evaluator with category-specific quality and effect rules. | `SKILL.md`, `reference/task-contract.md`, `reference/deliverable-registry.json`, `scripts/task_contract.py`, `../web/*`, `../evaluate-web-deliverable/*` |
| 2026-09-03 | Input → Model → Output tensor contracts were present, but the layout rule was scoped too narrowly to component cards and could be displaced by step-based interactions. | Added an explicit tensor-contract trigger: desktop system views keep inputs left, model transformations middle, and outputs right; controls synchronize all three regions; mobile stacks in semantic order. Added matching evaluator guidance. | `reference/web-insight-artifact.md`, `../web/references/system-demo.md`, `../evaluate-web-deliverable/references/profiles.md` |
| 2026-09-03 | Direct skill-modification requests needed a deliberate challenge step before reusable rules were changed, plus safer handling for high-impact evolution. | Added a skill-modification intent hook that interrogates intention and exclusions, exposes current practice, maps downstream impacts, classifies scope, and routes substantial changes through an isolated version-controlled cross-evolve comparison with explicit promotion approval and rollback. | `SKILL.md`, `reference/feedback-evolution.md` |
| 2026-09-03 | Evolve knowledge-distill to leverage Guarded Skill Evolution (GSE) and decision-frontier elicitation (grilling) from `idea-paper/guarded-skill-evolution/guarded-skill-evolution-v2.html`. | Upgraded intent hook to Guarded Skill Evolution (GSE) with 4-round decision-frontier grilling, strict facts-vs-decisions separation, 8-owner failure attribution matrix, non-skill defect guards, falsifiable YAML/JSON Change Contracts with CLI validator, and 4-slice paired evaluation gate. | `SKILL.md`, `reference/guarded-skill-evolution.md`, `reference/feedback-evolution.md`, `reference/task-contract.md`, `scripts/task_contract.py`, `scripts/test_task_contract.py`, `example/change-contract.yaml`, `reference/skill-evolution-ledger.md` |
| 2026-09-03 | Technical/algorithmic research deliverables require concrete code implementations (e.g., PyTorch) to ensure practical engineering transfer. | Updated web and knowledge-distill research-site standards to mandate production-style PyTorch/code architectures with interactive tabbed code inspectors for multi-strategy regimes, backed by evaluator profile verification. | `skills/web/references/research-site.md`, `skills/web/SKILL.md`, `skills/knowledge-distill/reference/web-insight-artifact.md`, `skills/evaluate-web-deliverable/references/profiles.md` |