Curated Thoughts boundary

Type: agentic-system-analyses/types/agentic-system-boundary.md

Boundary and evidence

This run covers the Curated Thoughts repository's shipped memory system, not merely its retrieval API. Its intended use is to turn vault files and agent deposits into retained knowledge that people review and agents later recall. The target class is memory/knowledge/context-engineering system. Its model-dependent librarian, embedding queries and agent-facing context tools make it in scope. The boundary kind is whole-system: the desktop caller, headless operators, both MCP servers, maintenance, persistence and retrieval belong to the target. They are not excluded hosts simply because they call shared libraries. SRC-1, INTENT.md, src-tauri/src/main.rs, src-tauri/src/pipeline/mod.rs, tools/src/bin/ct.rs and src-tauri/src/tool_dispatch.rs establish these responsibilities.

The target's own statement of purpose is:

You put raw material in as ordinary files. A component called the Active Librarian reads those files and turns them into short, trustworthy facts (the "wisdom layer"). Any agent, even a weak one, can then recall those facts. The goal is that good structure does the work, so agents don't need to be clever. --- INTENT.md

Initial tracing follows the desktop binary's GUI/MCP dispatch, GUI startup opening the configured database and starting its pipeline, successful ingest calling the librarian, synthesis calling a model and validating proposal JSON, and proposal review calling the shared commit layer. It also follows ct dispatch through its shared read/write modules and the separate read-only MCP server's recall implementation. These paths cover production, admission, persistence and later consumption of memory; inclusion does not assert that all paths have identical policies. SRC-1, src-tauri/src/main.rs, src-tauri/src/lib.rs, src-tauri/src/pipeline/mod.rs, src-tauri/src/librarian/synthesis.rs, src-tauri/src/proposals_api.rs, tools/src/bin/ct.rs, tools/src/cmds.rs, tools/src/queries.rs, tools/src/cli_common.rs, tools/src/bin/curated_thoughts_mcp.rs.

One executable connection from ingest to model-dependent maintenance is:

if let Err(e) = crate::librarian::generate_summary( &mut conn, &path, crate::librarian::active_generation_model( crate::setup::recommended_model(), ) .as_str(), false, ) { --- src-tauri/src/pipeline/mod.rs

The scope includes shipped generation prompts, integration instructions and maintenance scheduling, not only Rust storage code. The frontend supplies model and embedding callbacks to its external wiki engine; startup also installs deletion-triggered healing and periodic pruning. The shipped skill instructs agents to recall before coding and deposit after non-trivial work. These are source-native instructions, not evidence that an external agent followed them. SRC-1, src/lib/wiki.ts, src/App.tsx, .skills/curated-thoughts/skill.md.

const cleanupHeal = startAutoHeal(); const cleanupMaintenance = startAutoMaintenance(); --- src/App.tsx

The frozen evidence is SRC-1 at full commit 433ebbd48c2ffd0e871e34a1505dcdef1a78c8fd, applicable through the run cutoff 2026-11-08. The acquired source object is preserved unchanged. The checkout's HEAD matches that pin and its status was clean when inspected. The overall tier is code-grounded, with documentation used for declared intent and evaluator documentation. No target program, test, example, model endpoint or service was executed. This boundary establishes an inspectable operational scope, not successful deployment, benchmark reproduction or measured downstream benefit.

Functional coverage

The tracked tree was enumerated at the reviewed commit. Every top-level tracked file or directory is classified below. Included means inside the functional evidence boundary, not exhaustively read. Initial inspection reached the material entry points and loop connections described above; deeper implementation and evaluation inspection remains for the analysts. Unread available files are not access gaps.

All functional coverage above refers to SRC-1. No material shipped memory responsibility was deliberately excluded. Inspection is sufficient to establish the boundary and initial entry-point connections, not a full behavioral audit of every included file.

External dependencies and prevented conclusions

The separate curated-thoughts-integrations repository is outside the frozen source set. The target explicitly identifies CTI as a separate consumer of ct wisdom match; this run includes the command's implementation and shipped local skill, but not CTI session-start, per-turn or mid-task delivery, harness caching, deduplication or installation behavior. It cannot infer those properties from this repository's README. SRC-1, INTENT.md, README.md, tools/src/bin/ct.rs, tools/src/queries.rs.

The npm wiki/ontology packages and Rust dependencies are external implementations. Their pinned declarations and this repository's adapters/callbacks are included; their package source is not acquired. That gap prevents independently verifying internal external-engine algorithms or assigning them to Curated Thoughts without qualifying the adapter boundary. Local Ollama/sidecar inference, external OpenAI-compatible generation/embedding endpoints, optional Jev classification, remote cloud bridge services and a PostgreSQL replica are deployment dependencies, not additional frozen sources. The run can inspect the local configuration/call/persistence contracts but cannot establish provider quality, hosted service behavior, end-to-end privacy or remote durability. SRC-1, package.json, src-tauri/Cargo.toml, src/lib/wiki.ts, src-tauri/src/cloud_bridge/mod.rs, README.md.

No actual user's vault, persistent brain or agent execution trace was supplied. Repository fixtures, benchmark snapshots and calibration artifacts remain available within SRC-1; they must be examined under their stated conditions before supporting empirical claims. Frozen-vector evaluator source explicitly distinguishes code-path regression from model quality:

//! Frozen-vector regression guard for the ct wisdom match floor (issue #265). //! Recomputes hit@2 / FP through wisdom_match_with_floor from the vectors //! the calibration runs froze. Guards the CODE PATH, not the model. --- src-tauri/tests/wisdom_gate_bench.rs

Source register