{"id":"rsi-wiki","relativePath":"rsi-wiki.md","title":"AI-RSI compounding wiki","markdown":"# AI-RSI compounding wiki\n\nThis project uses a compounding RSI workflow: each cycle improves itself.\n\n## Architecture\n\n- **Raw sources**: project artifacts and truth anchors (`docs/`, `tests/`, `pnpm` output, PRs).\n- **RSI wiki**: lightweight operational notes in this repo (`CLAUDE.md`, `docs/roadmap.md`, `README.md`, close-out log).\n- **Schema**: this section in `CLAUDE.md` that constrains update cadence, evidence, and scope rules.\n\n## Purpose\n\n- Stop re-deriving process from memory at every cycle.\n- Preserve what changed, why it changed, and what was learned.\n- Keep improvements compounding rather than starting over each iteration.\n\n## Documentation as one source of truth\n\nFor AI + automation stability, we treat `docs/` Markdown as the canonical source:\n- **Structured sections:** keep heading hierarchy and tables to make extraction deterministic.\n- **Metadata + IDs:** include file IDs, update timestamps, checksums, and anchor IDs so agents can anchor evidence to exact versions.\n- **Canonical links + stable anchors:** publish through `/docs` (human) and `/api/docs/manifest` + `/api/docs/content` (machine), avoiding drift between markdown and surfaces.\n- **Single truth rule:** any process or behavior change in docs should update only `docs/...`; `/docs` and APIs re-render automatically and must be used by AI and agents as the live contract.\n\n## Iteration pattern (copy/paste ritual)\n\n1. **Ingest**  \n   Run checks, collect artifacts, and summarize outcomes.\n2. **Update**  \n   Make the required RSI updates (`CLAUDE.md`, `docs/roadmap.md`, `README.md`) in one pass.\n3. **Record**  \n   Add one close-out row with evidence references.\n4. **Refactor process**  \n   Add next-cycle hypothesis and known failure modes to keep learning actionable.\n5. **Compounding feedback**\n   Add technical suggestions for the next cycle in a prioritized, implementation-ready form so process quality compounds from each pass.\n\n## Evidence anchors to keep\n\n- `pnpm test`, `pnpm lint`, `pnpm build`\n- `pnpm session:closeout`\n- smoke/probe command outputs when available\n- updated docs + close-out log rows\n\n## Health checks (lints for the workflow)\n\n- Are evidence links still valid and attached?\n- Are recurring defects being carried into the next cycle?\n- Are scope-expansion items deferred until checks are clean?\n- Are updates done in all three canonical docs every iteration?\n\n## Required close-out suggestion packet (minimum)\n\nEach iteration should include all of the following in the close-out note:\n\n1. **Measured impact summary**\n   - What changed at the file/module level (new files, moved responsibilities, refactor boundaries).\n   - What behavior was preserved or intentionally changed (state transitions, error paths, API/queue contracts, route invariants).\n2. **Technical findings**\n   - Any failure mode observed, and how the change altered its likelihood or blast radius.\n   - Any subtle coupling discovered across services, tests, or config (imports, shared env vars, retry windows, queue transitions, etc.).\n3. **Evidence-backed recommendation set**\n   - At least 3 ranked follow-up suggestions for compounding RSI in the next cycle:\n     - [ ] **P0/P1 reliability hardening** (e.g., edge condition handling, recovery behavior).\n     - [ ] **P0/P1 test quality improvement** (missing assertions, fuzz/contract gaps, negative-path coverage).\n     - [ ] **P1/P2 maintainability hardening** (decomposition, ownership boundaries, commentable invariants).\n   - Each suggestion must include owner + concrete measurable acceptance criterion (file names + command/test) so it can be executed with minimal ambiguity.\n4. **Next-cycle hypothesis**\n   - One falsifiable hypothesis with a confidence score and the exact test/smoke command that validates or rejects it.\n\nKeep this packet in `CLAUDE.md` updates as short bullets so evidence is actionable, not just declarative.\n\n## Minimal working contract\n\n- One measurable hypothesis per cycle.\n- Expand scope only after prior cycle is clean.\n- Pause RSI expansion if required checks/evidence are missing.\n- Keep evidence lightweight and append-only where practical.\n","sections":[{"level":2,"heading":"Architecture","anchor":"architecture"},{"level":2,"heading":"Purpose","anchor":"purpose"},{"level":2,"heading":"Documentation as one source of truth","anchor":"documentation-as-one-source-of-truth"},{"level":2,"heading":"Iteration pattern (copy/paste ritual)","anchor":"iteration-pattern-copy-paste-ritual"},{"level":2,"heading":"Evidence anchors to keep","anchor":"evidence-anchors-to-keep"},{"level":2,"heading":"Health checks (lints for the workflow)","anchor":"health-checks-lints-for-the-workflow"},{"level":2,"heading":"Required close-out suggestion packet (minimum)","anchor":"required-close-out-suggestion-packet-minimum"},{"level":2,"heading":"Minimal working contract","anchor":"minimal-working-contract"}],"html":"<h1 id=\"ai-rsi-compounding-wiki\">AI-RSI compounding wiki</h1>\n<p>This project uses a compounding RSI workflow: each cycle improves itself.</p>\n<h2 id=\"architecture\">Architecture</h2>\n<ul><li><strong>Raw sources</strong>: project artifacts and truth anchors (`docs/`, `tests/`, `pnpm` output, PRs).</li><li><strong>RSI wiki</strong>: lightweight operational notes in this repo (`CLAUDE.md`, `docs/roadmap.md`, `README.md`, close-out log).</li><li><strong>Schema</strong>: this section in `CLAUDE.md` that constrains update cadence, evidence, and scope rules.</li></ul>\n<h2 id=\"purpose\">Purpose</h2>\n<ul><li>Stop re-deriving process from memory at every cycle.</li><li>Preserve what changed, why it changed, and what was learned.</li><li>Keep improvements compounding rather than starting over each iteration.</li></ul>\n<h2 id=\"documentation-as-one-source-of-truth\">Documentation as one source of truth</h2>\n<p>For AI + automation stability, we treat `docs/` Markdown as the canonical source:</p>\n<ul><li><strong>Structured sections:</strong> keep heading hierarchy and tables to make extraction deterministic.</li><li><strong>Metadata + IDs:</strong> include file IDs, update timestamps, checksums, and anchor IDs so agents can anchor evidence to exact versions.</li><li><strong>Canonical links + stable anchors:</strong> publish through `/docs` (human) and `/api/docs/manifest` + `/api/docs/content` (machine), avoiding drift between markdown and surfaces.</li><li><strong>Single truth rule:</strong> any process or behavior change in docs should update only `docs/...`; `/docs` and APIs re-render automatically and must be used by AI and agents as the live contract.</li></ul>\n<h2 id=\"iteration-pattern-copy-paste-ritual\">Iteration pattern (copy/paste ritual)</h2>\n<ol><li><strong>Ingest</strong>  </li></ol>\n<p>   Run checks, collect artifacts, and summarize outcomes.</p>\n<ol><li><strong>Update</strong>  </li></ol>\n<p>   Make the required RSI updates (`CLAUDE.md`, `docs/roadmap.md`, `README.md`) in one pass.</p>\n<ol><li><strong>Record</strong>  </li></ol>\n<p>   Add one close-out row with evidence references.</p>\n<ol><li><strong>Refactor process</strong>  </li></ol>\n<p>   Add next-cycle hypothesis and known failure modes to keep learning actionable.</p>\n<ol><li><strong>Compounding feedback</strong></li></ol>\n<p>   Add technical suggestions for the next cycle in a prioritized, implementation-ready form so process quality compounds from each pass.</p>\n<h2 id=\"evidence-anchors-to-keep\">Evidence anchors to keep</h2>\n<ul><li>`pnpm test`, `pnpm lint`, `pnpm build`</li><li>`pnpm session:closeout`</li><li>smoke/probe command outputs when available</li><li>updated docs + close-out log rows</li></ul>\n<h2 id=\"health-checks-lints-for-the-workflow\">Health checks (lints for the workflow)</h2>\n<ul><li>Are evidence links still valid and attached?</li><li>Are recurring defects being carried into the next cycle?</li><li>Are scope-expansion items deferred until checks are clean?</li><li>Are updates done in all three canonical docs every iteration?</li></ul>\n<h2 id=\"required-close-out-suggestion-packet-minimum\">Required close-out suggestion packet (minimum)</h2>\n<p>Each iteration should include all of the following in the close-out note:</p>\n<ol><li><strong>Measured impact summary</strong></li></ol>\n<ol><li><strong>Technical findings</strong></li></ol>\n<ol><li><strong>Evidence-backed recommendation set</strong></li></ol>\n<ol><li><strong>Next-cycle hypothesis</strong></li></ol>\n<ul><li>What changed at the file/module level (new files, moved responsibilities, refactor boundaries).</li><li>What behavior was preserved or intentionally changed (state transitions, error paths, API/queue contracts, route invariants).</li><li>Any failure mode observed, and how the change altered its likelihood or blast radius.</li><li>Any subtle coupling discovered across services, tests, or config (imports, shared env vars, retry windows, queue transitions, etc.).</li><li>At least 3 ranked follow-up suggestions for compounding RSI in the next cycle:</li><li>[ ] <strong>P0/P1 reliability hardening</strong> (e.g., edge condition handling, recovery behavior).</li><li>[ ] <strong>P0/P1 test quality improvement</strong> (missing assertions, fuzz/contract gaps, negative-path coverage).</li><li>[ ] <strong>P1/P2 maintainability hardening</strong> (decomposition, ownership boundaries, commentable invariants).</li><li>Each suggestion must include owner + concrete measurable acceptance criterion (file names + command/test) so it can be executed with minimal ambiguity.</li><li>One falsifiable hypothesis with a confidence score and the exact test/smoke command that validates or rejects it.</li></ul>\n<p>Keep this packet in `CLAUDE.md` updates as short bullets so evidence is actionable, not just declarative.</p>\n<h2 id=\"minimal-working-contract\">Minimal working contract</h2>\n<ul><li>One measurable hypothesis per cycle.</li><li>Expand scope only after prior cycle is clean.</li><li>Pause RSI expansion if required checks/evidence are missing.</li><li>Keep evidence lightweight and append-only where practical.</li></ul>","updatedAt":"2018-10-20T01:46:40.000Z","checksum":"b64914fe6f200161bbba35b6a4e770d4559818937df58c5c1b6fe1edc2181553","checksumPrefix":"b64914fe6f20","anchorCount":8,"lineCount":79,"rawUrl":"/api/docs/content?path=rsi-wiki.md","htmlUrl":"/docs?doc=rsi-wiki.md","apiUrl":"/api/docs/content?path=rsi-wiki.md"}