docs: Distinguish supplied messages from automatic extracted facts

[buildingjoshbetter]

Automatic transcript capture stores selected extracted facts, while benchmark and direct-import routes store supplied message content after preprocessing. The documentation blurred those routes and implied complete conversation coverage. Refs #744; stacked on #828.

Documentation scope

Clarify the input routes in README and architecture docs, including bulk replacement, filtered extraction input, limited diagnostic buffers/snapshots and the stored-memory scope of ordinary and Deep Search. Distinguish installed hook bounds from separate core helpers. LoCoMo resolves relative dates before import; supplied benchmark content is not a claim of byte-verbatim original history. Benchmark scores, tier tables, code samples and runtime behavior are unchanged.

Prepared publication head: e116a91254534e3c9e92cd079002cb5f06e282d1. Proposed remote refresh destination: fix/ingestion-coverage-744-v2. Intended base: draft #828 through fix/tier-public-publish-v1, exact prepared Unit C head 5873e5a4978f37bcd31772d6840717fb217df713.

The feature delta is exactly README.md and docs/architecture.md, 30 insertions/eight deletions. Production and tests are identical to the Unit C base. Both documents are byte-identical to previously reviewed docs head 51927b9a6495327d6b3379c1e591822943fe615d. All 13 new relative links target tracked public files, the coverage anchor resolves, and all 41 README table rows and ten fenced code blocks across both documents are preserved.

Inherited base evidence

The base contains the narrow #684 closed/I/O cached-handle correction. Waiting requires exact Engine/failed-handle proof and verified automatic-owner receipts, with current-handle identity rechecked around waiting and one shared 10,000 ms budget. Current-owned writers refuse before opening; manual/receiptless owners and serving/already-admitted callers retain refusal. The historical macOS and old #829 Windows refusal logs did not establish the actual owner identity; duplicate jobs passed. Separate synthetic actual-function controls reproduce the verified automatic-owner condition. Those runtime findings remain distinct from this two-document delta.

The base has 26 passing reconnect methods, three explicit GC-enabled assertions and 42 passing independent reconnect controls. All 101 boundary methods and 550 unittest assertion and fail call AST nodes remain preserved; the complete memory-only boundary run passes 101/101. Original failures remain preserved.

Original combined full source 5cd23db000fd2a7c98e93395be33cf8638641db1 failed ten MCP fixture tests: nine missing-time NameError traces and one parallel-wait assertion cascade. Its source, container and log remain preserved, log SHA256 2a706b32240afa808f582e1751beeea1a24302f1a6f86f3eb3eddec5fa85befd. The inherited MCP-only fixture repair retains all 11 test methods and 52 unittest/Mock assertion and fail call AST nodes, binds actual source-derived collaborators and strict unexpected-wait/import guards, and passes 11 safe methods plus 35 independent controls. This repair changes no production, model, retrieval or GC behavior.

Gate	Exact source	Passing tests / subtests	Runner elapsed	Full-log SHA256
Expanded focused	3a9360b1ba6bf210a1b54e05287117923fbbaae5	833 / 633	39.36 s	e1c203f112a3dedc0411874785f00a528c6bd3d34809657f1903df46da7ee0df
Combined full	3a9360b1ba6bf210a1b54e05287117923fbbaae5	3,735 / 3,223	412.643 s	a750f39645d1c866e5e68cddaa9ad49fe3616f3500be6180dadbbb4344dbeaed
Exact Unit C base full	5873e5a4978f37bcd31772d6840717fb217df713	3,531 / 2,396	411.415 s	13376c51a4174b4b2dc8fda569911d0e625a2cba9912e4bfe8d75bba3f0bd3bc

Both full gates exited 0 with 17 skips and three network deselections, no timeout/OOM and matching actual source/build/log hashes. They ran sequentially in offline installed-wheel images with eight CPUs, 16 GiB and read-only source/models/fixtures; native SQLite flags stayed enabled. No GC-disabling override was used; three selected regression assertions observed GC enabled at their checkpoints. Full-run runtime GC state was not recorded. All-test installed import origin remains unmeasured. These are inherited runtime/base gates, not a separate new heavy docs gate.

Receipts: .work/reports/native-integrations-focused-validation-v6.json, .work/reports/native-integrations-combined-full-validation-v2.json and .work/reports/tier-public-publication-validation-v4.json, with corresponding actual container-config receipts. New-head hosted CI remains pending. Validated private integration remains 7bd70e9fa4812e9925b86652dbe5c3a5c92bafe1; new private adoption is separate.

The documentation delta does not establish native host lifecycle/capture/model-visible context acceptance. Complete archive retrieval and durable source citations remain future work. The original 26.24 GB Mac memory fix remains unproven. Broader issue acceptance stays open. Draft only; no merge or release.