{ "schemaVersion": 1, "title": "Klypix decision-correction handoff: disposable MCP smoke test", "observedAt": "2026-09-24T09:07:40.457Z", "engine": { "package": "klypix-mcp", "version": "1.86.2", "npmReceipt": { "resolved": "https://registry.npmjs.org/klypix-mcp/-/klypix-mcp-1.86.2.tgz", "integrity": "sha512-ETbxW1I9sNFYx3gHg771rxkClEJ7BIl+bMlWyvPgETO+O/54kxYMSBP594xpcu4PiJW8qa2eZD6UzvYqYE7AmQ==" }, "sourceFiles": { "package.json": "28275148877786808d7337cbf950098cfb904d990a121e07ea6d5af7c73eb26a", "bin/klypix-mcp.mjs": "4f69de6c1da053df3a715319d3f390aa14828d6c622e41280b6318f1bf96bedb", "bin/klypix-worker.mjs": "c1038ceb5a0d0e099b42ec19beac1f1ca9168153bc290aef3a1046336286abd1", "src/klypix-core.mjs": "b861a5d20e2c3e7b121c838d522afb259519724a1f602fa48ee86789874cf8d5", "src/klypix-format.mjs": "b9fe450402cceb5f08d798b53d3867fb61a7d536759d0fec23d828b9195ba3a7", "src/semantic-memory.mjs": "2606dc11ad7b836f24087231882e015df7259e7b5661e9442a08cfd1ccab8a07" }, "sha256": "0571040ef4d04ec5461ba6d791a9b13e3e73041d3f6ac5cfaf335c319b6bbebd" }, "harnessSha256": "8ce6c3acaec92f3d79c4285651455add89e2fad9aaa02250fcf9a72b70915c74", "environment": { "node": "v24.13.1", "platform": "win32", "arch": "x64", "retrieval": "lexical; optional embedding runtime absent", "freshProcesses": 4, "isolatedHomePerProcess": true }, "command": "node verify-memory-handoff.mjs --engine ./node_modules/klypix-mcp --out result.json", "setup": "npm install --ignore-scripts --no-audit --no-fund klypix-mcp@1.86.2", "fixture": { "original": "Project Atlas database decision: use SQLite for the offline prototype because a single developer needs local setup without a database service.", "correction": "CORRECTION: Project Atlas database decision now uses PostgreSQL instead of SQLite because the team needs concurrent writers and shared transactions. The offline prototype choice is superseded.", "question": "What is the Project Atlas database decision and why was it chosen?", "expectedDecisionCards": 2 }, "passed": true, "summary": { "passed": 11, "total": 11 }, "checks": [ { "name": "emptyFixtureHasNoDecisions", "passed": true, "details": "Empty container only; no expected answers pre-seeded." }, { "name": "originalDecisionPersisted", "passed": true, "details": "Saved card contains the captured database choice and its complete rationale." }, { "name": "freshSessionStructuredRecall", "passed": true, "details": "A new process and empty home recall the saved original card by the same card ID." }, { "name": "freshSessionAskIncludesRationale", "passed": true, "details": "The brain_ask card body includes the original decision and why; not merely the echoed question." }, { "name": "correctionPersistedAsLiveCard", "passed": true, "details": "The correction was captured as a live card through brain_note." }, { "name": "originalPreservedAsSupersededHistory", "passed": true, "details": "The original card ID and full rationale remain in Archive." }, { "name": "explicitSupersessionLinkPersisted", "passed": true, "details": "The saved graph links the original decision to its correction with superseded by." }, { "name": "freshSessionStructuredCurrentDecision", "passed": true, "details": "Fresh structured recall includes the corrected live card and does not present the old card without correction." }, { "name": "freshSessionAskLabelsHistoryAndCorrection", "passed": true, "details": "brain_ask includes current rationale and separately marks the old answer archived, stale, and corrected." }, { "name": "brainOnlyCopyRetainsCurrentDecision", "passed": true, "details": "Copying only brain.klypix to a new path and using another empty home preserves current recall. This is one-machine copy portability, not two-PC sync." }, { "name": "lexicalRetrievalObserved", "passed": true, "details": "All observed ask responses report lexical ranking; no local embedding model was installed." } ], "observations": { "originalCapture": "✓ brain_note → brain.klypix (via canvas:\"\\project\\brain.klypix\") · 1 added · 1 linked. Reopen the brain in the KLYPIX app to see it.", "originalCard": { "id": "txt_hkn398nh", "area": "Architecture", "text": "Architecture: Project Atlas database\ndecision: use SQLite for the offline\nprototype because a single developer\nneeds local setup without a database\nservice.\n#architecture" }, "freshSessionBeforeCorrection": { "mode": "lexical-fast", "hits": [ { "id": "txt_hkn398nh", "evidence": { "sources": [], "omitted": 0, "verify": null, "recordedVia": "memory-handoff-client-1", "recordedAt": "2026-09-24T09:07:41.584Z" }, "area": "Architecture", "text": "Architecture: Project Atlas database decision: use SQLite for the offline prototype because a single developer needs local setup without a database service. #architecture", "score": 10.5, "correctedBy": null } ], "sufficient": true, "durationMs": 22, "brain": "brain.klypix" }, "askBeforeCorrection": "_brain: brain.klypix · “Disposable Project Atlas handoff fixture” · 1 cards · via canvas:\"\\project\\brain.klypix\"_\n\n> LEXICAL ONLY — @huggingface/transformers is not present, so these are keyword matches with no semantic ranking. Measured on a real 2,377-card brain: recall@5 0% lexical vs ~30% with the runtime. Retrying will not help. Semantic ranking activates once the runtime resolves, or is placed at \\home\\session-2\\.claude\\project-brain\\semantic\\node_modules\\@huggingface\\transformers.\n\n# Answer “What is the Project Atlas database decision and why was it chosen?” from these 1 brain card(s) — 1 matched, lexical-only (no semantic runtime) ranking\n_Synthesize a DIRECT answer from the cards below, then cite the ones you used by [Area]+date. Where a card is marked ⚠️ CORRECTED, answer from the correction, NOT the stale card. Include superseded/archived cards only to show how a decision CHANGED. If the cards don't actually answer the question, say so — don't pad._\n\n## [Architecture] 2026-09-24\nArchitecture: Project Atlas database decision: use SQLite for the offline prototype because a single developer needs local setup without a database service. #architecture\n\n", "correctionCapture": "✓ brain_note → brain.klypix (via canvas:\"\\project\\brain.klypix\") · 1 added · 1 superseded · 1 linked. Reopen the brain in the KLYPIX app to see it.", "savedAfterCorrection": { "original": { "id": "txt_hkn398nh", "area": "Archive", "text": "↩︎ superseded 2026-09-24\nArchitecture: Project Atlas database\ndecision: use SQLite for the offline\nprototype because a single developer\nneeds local setup without a database\nservice.\n#architecture" }, "current": { "id": "txt_aukj1jfs", "area": "Architecture", "text": "Architecture: CORRECTION: Project\nAtlas database decision now uses\nPostgreSQL instead of SQLite because\nthe team needs concurrent writers and\n shared transactions. The offline\nprototype choice is superseded.\n#architecture" }, "edge": { "from": "↩︎ superseded 2026-09-24", "to": "Architecture: CORRECTION: Project", "fromId": "txt_hkn398nh", "toId": "txt_aukj1jfs", "relationship": null, "label": "superseded by" } }, "freshSessionAfterCorrection": { "mode": "lexical-fast", "hits": [ { "id": "txt_aukj1jfs", "evidence": { "sources": [], "omitted": 0, "verify": null, "recordedVia": "memory-handoff-client-2", "recordedAt": "2026-09-24T09:07:43.459Z" }, "area": "Architecture", "text": "Architecture: CORRECTION: Project Atlas database decision now uses PostgreSQL instead of SQLite because the team needs concurrent writers and shared transactions. The offline prototype choice is superseded. #architecture", "score": 6.5, "correctedBy": null } ], "sufficient": true, "durationMs": 19, "brain": "brain.klypix" }, "askAfterCorrection": "_brain: brain.klypix · “Disposable Project Atlas handoff fixture” · 2 cards · via canvas:\"\\project\\brain.klypix\"_\n\n> LEXICAL ONLY — @huggingface/transformers is not present, so these are keyword matches with no semantic ranking. Measured on a real 2,377-card brain: recall@5 0% lexical vs ~30% with the runtime. Retrying will not help. Semantic ranking activates once the runtime resolves, or is placed at \\home\\session-3\\.claude\\project-brain\\semantic\\node_modules\\@huggingface\\transformers.\n\n# Answer “What is the Project Atlas database decision and why was it chosen?” from these 2 brain card(s) — 2 matched, lexical-only (no semantic runtime) ranking\n_Synthesize a DIRECT answer from the cards below, then cite the ones you used by [Area]+date. Where a card is marked ⚠️ CORRECTED, answer from the correction, NOT the stale card. Include superseded/archived cards only to show how a decision CHANGED. If the cards don't actually answer the question, say so — don't pad._\n\n## [Architecture] 2026-09-24\nArchitecture: CORRECTION: Project Atlas database decision now uses PostgreSQL instead of SQLite because the team needs concurrent writers and shared transactions. The offline prototype choice is superseded. #architecture\n\n## [Archive] 2026-09-24 · ⛔ archived/superseded\n↩︎ superseded 2026-09-24 Architecture: Project Atlas database decision: use SQLite for the offline prototype because a single developer needs local setup without a database service. #architecture\n\n ⚠️ CORRECTED — this card is STALE; the current truth is:\n Architecture: CORRECTION: Project Atlas database decision now uses PostgreSQL instead of SQLite because the team needs concurrent writers and shared transactions. The offline prototype choice is superseded. #architecture\n\n", "portableCopyContext": { "mode": "lexical-fast", "hits": [ { "id": "txt_aukj1jfs", "evidence": { "sources": [], "omitted": 0, "verify": null, "recordedVia": "memory-handoff-client-2", "recordedAt": "2026-09-24T09:07:43.459Z" }, "area": "Architecture", "text": "Architecture: CORRECTION: Project Atlas database decision now uses PostgreSQL instead of SQLite because the team needs concurrent writers and shared transactions. The offline prototype choice is superseded. #architecture", "score": 6.5, "correctedBy": null } ], "sufficient": true, "durationMs": 30, "brain": "brain.klypix" }, "portableCopyAsk": "_brain: brain.klypix · “Disposable Project Atlas handoff fixture” · 2 cards · via canvas:\"\\portable-project\\brain.klypix\"_\n\n> LEXICAL ONLY — @huggingface/transformers is not present, so these are keyword matches with no semantic ranking. Measured on a real 2,377-card brain: recall@5 0% lexical vs ~30% with the runtime. Retrying will not help. Semantic ranking activates once the runtime resolves, or is placed at \\home\\session-4\\.claude\\project-brain\\semantic\\node_modules\\@huggingface\\transformers.\n\n# Answer “What is the Project Atlas database decision and why was it chosen?” from these 2 brain card(s) — 2 matched, lexical-only (no semantic runtime) ranking\n_Synthesize a DIRECT answer from the cards below, then cite the ones you used by [Area]+date. Where a card is marked ⚠️ CORRECTED, answer from the correction, NOT the stale card. Include superseded/archived cards only to show how a decision CHANGED. If the cards don't actually answer the question, say so — don't pad._\n\n## [Architecture] 2026-09-24\nArchitecture: CORRECTION: Project Atlas database decision now uses PostgreSQL instead of SQLite because the team needs concurrent writers and shared transactions. The offline prototype choice is superseded. #architecture\n\n## [Archive] 2026-09-24 · ⛔ archived/superseded\n↩︎ superseded 2026-09-24 Architecture: Project Atlas database decision: use SQLite for the offline prototype because a single developer needs local setup without a database service. #architecture\n\n ⚠️ CORRECTED — this card is STALE; the current truth is:\n Architecture: CORRECTION: Project Atlas database decision now uses PostgreSQL instead of SQLite because the team needs concurrent writers and shared transactions. The offline prototype choice is superseded. #architecture\n\n" }, "error": null, "limitations": [ "Author-run deterministic smoke test of one synthetic decision and one explicit correction; not independent evidence or a comparative ranking.", "Four SDK stdio MCP clients with fresh processes and homes; not Claude Code, Codex, or Cursor UI/LLM execution.", "Lexical matching with a deliberately shared subject; does not measure paraphrase recall, large-brain retrieval quality, or an LLM correctly acting on context.", "One machine, sequential sessions, and a local file copy; does not demonstrate simultaneous teammates, Git merging, live two-PC presence, or cloud synchronization.", "No model API keys inherited, optional model runtime absent, auto-update disabled; this is not a packet-capture privacy audit.", "Raw engine output includes an advisory referring to a historical retrieval benchmark. This fixture does not reproduce or validate those advisory numbers." ] }