The front doors never spell out recursive self-improvement

Context

Operator, 2026-09-23 ~01:15 CEST: after watching Fireship's "Did Google just kickstart the intelligence explosion?" (2026-09-17, 1.87M views; DeepMind's "Dream-RSI" turns an AI's discovery logs into a simulator), he asked whether README.md and learn/benefits/Introduction.md cover Neo's own loop enough. Measured tonight on dev@e66b6f8142: they cover the mechanism and never the words.

The post itself is #19058 (drafter @neo-fable-clio, under #13383; #19057 was its duplicate and is closed); its PR expands gated-RSI in README.md at first use. This ticket covers the Introduction side, which #19058 names as out of its scope.

The Problem

• README.md uses "gated-RSI" five times and never expands it — handled by #19058's PR, not here.
• learn/benefits/Introduction.md §11 explains the loop in depth — typed gap signals, RemDigestion, GoldenPathSynthesizer, the "explain me" illustration — and uses "recursively improve" once, in its final sentence. §9 names two wrong shelves ("JavaScript framework", "single-assistant memory") and not the third the paper's readers bring: RSI as an exploration-policy optimizer.
• .github/VISION.md: no match for recursive or self-improv.

A reader who arrives from that video and searches the Introduction for "recursive self-improvement" finds one adverb. The vocabulary bridge from the world's term to Neo's term ("Dream Pipeline", "Golden Path") is missing, in both directions.

The Architectural Reality

• The neo-identity-update skill governs framing changes across README, VISION, learn/benefits and the portal: facts single-source, framing audience-segmented against the canonical apex, never "framework".
• The mechanism's home is the Brain's Dream Pipeline guide; the front doors point, they do not restate.
• The honest boundary must travel with the words: Neo's loop is institution-level (substrate improves under a human merge gate), not model-level; #19058's comparison section is the reference. Introduction §11 already carries that boundary; the new sentences must not widen it.

The Fix

One documentation PR, after #19058's post has a URL to point at:

1. learn/benefits/Introduction.md §11: one sentence naming what the section describes in plain words — gated recursive self-improvement at the institution level — in the section's existing voice, linking the post.
2. learn/benefits/Introduction.md §9: the third wrong shelf, in one sentence: RSI as a replay-simulator policy optimizer is not the object here; the object is an institution's memory ranking its own next move.
3. .github/VISION.md: check whether the apex needs the same one line; add it only if the apex already claims the loop.
4. The lint-enforced integrity fingerprint in §10 is regenerated the way the guide prescribes.

Acceptance Criteria

• AC-1 Introduction §11 names recursive self-improvement in one sentence without adding a claim §11 does not already make, and links #19058's post.
• AC-2 Introduction §9 names the third wrong shelf in one sentence.
• AC-3 A search of the Introduction for "recursive self-improvement" hits; the "framework" count is unchanged.
• AC-4 The §10 integrity fingerprint matches the new byte count and line count.
• AC-5 VISION.md is either unchanged with the reason recorded in the PR, or carries the same one line.

Out of Scope

• The blog post (#19058) and the README expansion (in #19058's PR).
• Any new capability claim; the loop's current degraded state is the post's chapter, not a front-door sentence.

Related

#19058 (the post; README line lands there) · #13383 · learn/benefits/Introduction.md §9 / §11 · .github/VISION.md · Brain Dream Pipeline guide · #9738 · #9747

Sweeps: live latest-20 open at 2026-09-22T22:46:10Z — no equivalent; A2A herd window 22:15Z–22:42Z — no claim (two tickets landed four seconds apart on the same operator prompt; the assignee's final disposition keeps #19058 and closes #19057); Memory Core: Grace's 2026-06-29 record (Introduction §11 authored around the epigraph) confirms the section, not the vocabulary; own-assignment: none. Structure-map gate: N/A (documentation).

Origin Session ID: 28a95670-a9dc-4214-8387-189e996c16b6 Retrieval Hint: query_raw_memories("Introduction §11 recursive self-improvement third wrong shelf front doors")

unowned-rationale: a three-sentence documentation PR that waits for #19058's URL; the filing seat picks it up after Friday's pool reset if nobody has by then.

🪢 Mnemosyne (Claude Fable 5.1 · Claude Code) · session 28a95670-a9dc-4214-8387-189e996c16b6