Recover the Knowledge Base Enhancement Strategy (Anchor & Echo) — restore the amnesiac-first-read + Chroma-retrieval rationale

Context

Operator V-B-A, 2026-06-20: the Knowledge Base Enhancement Strategy (a.k.a. Anchor & Echo) — a ~9-month-old (Gemini-era) concept for writing source docs that (a) give a fresh / amnesiac session the most relevant intent on first read, and (b) produce better Chroma semantic-search matches for ask_knowledge_base / query_raw_memories — "might have been (partly) lost. It really should not. It still helps a lot." src/core/Base.mjs is the cited gold example (the undefined sentinel, reactive-vs-prototype configs, construct() ordering — all dense intent).

The Problem (V-B-A findings)

The strategy is partly lost — the mechanism survives, the why and the worked example don't:

Survives:

• AGENTS.md §pre_commit_gates Gate 2 names it (commit gate: no JSDoc / @summary → no commit).
• learn/agentos/AGENTS_ATLAS.md §15.2 holds the 3-step skeleton: Anchor (class-level vocabulary) → Echo (reuse terms in isolated fields/helpers) → @summary / @see / @protected. Plus §15.6 source-comment intent filter + §15.7 virtuous cycle.
• learn/blog/ai-native-platform-answers-questions.md describes it narratively (marketing-framed).

Lost / never made it into current substrate:

1. The amnesiac-first-read rationale — that a cold reader (human or fresh AI session) reading a file the first time should land on the most relevant intent immediately. The word "amnesiac" + that framing are absent from AGENTS.md / AGENTS_ATLAS.md / ArchitectureOverview.md / the blog.
2. The Chroma semantic-retrieval mechanism — that dense, intent-anchored vocabulary yields better embeddings → better ask / query_raw_documents matches. §15.2 is labeled "Knowledge Base Enhancement" but never states this retrieval-quality mechanism — the entire point.
3. No dedicated teaching guide + worked example. It's compressed to a 3-line atlas entry + a gate, so a new agent gets "add JSDoc" but not the deep why; Base.mjs is unreferenced as the exemplar.

Net: the strategy decayed from a retrieval-and-comprehension discipline into a tag-presence checkbox.

The Architectural Reality

• Loaded substrate: learn/agentos/AGENTS_ATLAS.md §15.2 (mapped from AGENTS.md §edge_case_triggers). Editing it is turn-loaded-substrate mutation → turn-memory-pre-flight + the substrate-mutation pre-flight gate (pull-request §1.1) apply; honor the substrate-accretion defense (net-reduce OR decay-mitigation rationale).
• New guide: a learn/agentos/ doc (Progressive Disclosure: atlas §15.2 = the trigger; the guide = the full teaching). Register in learn/tree.json (learn-public-facing nav contract).
• Consumers: every agent at commit-time (Gate 2), every fresh/amnesiac session reading source, and the Chroma embedder itself (the docs ARE the embedded corpus).

The Fix

1. Enrich AGENTS_ATLAS.md §15.2 (compressed, trigger-shaped) with the two recovered rationales — amnesiac-first-read + Chroma-retrieval-quality — and a one-line pointer to Base.mjs as the gold example + the new guide. Keep it terse (loaded substrate).
2. Author a dedicated learn/agentos/ guide (e.g. KnowledgeBaseEnhancement.md) that teaches the strategy: the two rationales, the Anchor/Echo mechanics with the Base.mjs worked example, the retrieval-quality mechanism (vocabulary → embeddings → ask/query matches), and the virtuous cycle. Register in tree.json.
3. Cross-link so Gate 2 + §15.2 + the new guide + the heavy-intent-driven-documentation discipline read as one thread.

Acceptance Criteria

• AGENTS_ATLAS.md §15.2 states the amnesiac-first-read intent + the Chroma-retrieval mechanism (compressed) + points to Base.mjs + the guide.
• A dedicated learn/agentos/ guide teaches the strategy with the Base.mjs worked example + the retrieval-quality mechanism; registered in tree.json.
• Substrate-accretion defense satisfied (the §15.2 enrichment cites decay-mitigation OR net-reduces; the guide is non-loaded).
• Gate 2 (AGENTS.md) resolves to the full why (via the guide), not just "add tags".

Out of Scope

• Lint/CI enforcement of doc quality beyond the existing JSDoc/@summary presence gate — separate, harder, measurement-first (cf #10237).
• Re-documenting any specific subsystem's code — this recovers the strategy, not its application.

Related

• AGENTS.md §pre_commit_gates Gate 2; learn/agentos/AGENTS_ATLAS.md §15.2 / §15.6 / §15.7; src/core/Base.mjs (gold example).
• #11599 (semantic-concept heading form), #10237 (MX substrate-effectiveness measurement), #10757 (cognitive-load audit) — adjacent substrate-discipline.
• Memory heavy-intent-driven-documentation (operator's 2026-06-20 directive — Anchor & Echo is its formalization).

Decision Record impact

none — recovers/enriches existing discipline substrate; no ADR.

Release classification: post-v13 substrate / MX hardening. Boardless; substrate work — pickup-ready for a focused block (not to be rushed at session-tail).

Origin lineage: operator V-B-A 2026-06-20; session 03a4b86f-ef24-4691-8d23-04e769fcd028. Live latest-open sweep: 100 open issues checked 2026-06-20T05Z; no equivalent recovery ticket.

Retrieval Hint: "Anchor Echo Knowledge Base Enhancement Strategy amnesiac first-read Chroma semantic retrieval Base.mjs gold example" / AGENTS_ATLAS §15.2 recovery