Rediscovery inventory — the five-chat mining pass

Rediscovery inventory — the five-chat mining pass

*Written 2026-09-16, filed 2026-09-21. The five-day gap is the point: it was written into Claude's memory store rather than the repo, where no repo check could see it, and the notes/asked.md row mining-5 correctly reported it missing while it existed elsewhere. See OPS.md, an answer can be filed to the wrong store.*

Kept separate, never merged. Per-chat inventories are not pooled — that rule is what keeps the rediscovery detector honest, and it is the same rule notes/rediscovery-inventory-continued-fractions.md and N-022 carry.

The evidence ledger for the selection-effect probe is [[N-021]], not this file. Where the two touch, N-021 owns the tally and this note defers to it; this is the per-chat raw material.

---

Cross-chat: the selection-effect pattern

One compatibility condition run until it fails yields a finite exceptional list.

• Surfaced blind in four chats: division algebras, exceptional lists, Niemeier lattices (archive), exceptional Lie groups (E8 chat).
• By the cross-chat two-instance rule, four blind rediscoveries make it a central structural object rather than a house preference.
• ~~Proposes elevating it to a top-level exemplar that the division-algebra ladder, exceptional Lie groups, exceptional lists, and Platonic/E-series point at as instances.~~ SUPERSEDED — kept as history, not as a live proposal. The two independent verdicts below concluded pattern, not mechanism, and [[N-021]] holds that position: the probe stays a pattern-with-one-proof until a second instance is proved through the same congruence, and the verdicts are evidence for staying at pattern, not for elevating. The elevation question remains open in N-021, where it is owned.
• E8-chat instance (item 9 / A5-D3): exceptional groups exist because the octonions stop at 8; the exceptional series terminates because the coincidences run out. Filed once, cross-referenced, not refiled.
• Two independent pattern-vs-mechanism verdicts. Wallis/NS: "four conditions, one silhouette — pattern not mechanism." E8 chat, independently (item 5): "exceptional structures rhyme, but rhyme is not identity." Folded as a second instance strengthening the probe, not a new finding.
• Verified evidence state — four genuinely-blind cross-chat instances, two this-chat instances (Niven, tops), one negative control (Chen–Liu), two independent verdicts. Any instance that turns out to be a tracked theme with back-references is dropped from the count. Reconciled against [[N-021]] on filing: the counts agree — four cross-chat plus two Wallis/NS is N-021's "six blind instances", same negative control, same two verdicts. (A wording ambiguity flagged here on filing -- N-021 headlining "six blind instances" while a later line read "four is the count of blind instances" -- was resolved in N-021 on 2026-09-21. The counts were never in dispute; the sentence was.)*

This chat (Wallis/NS): instances and negative control

• Niven's theorem — rationality of 2 sin(fπ) run over all rational fractions yields the finite list {0, 1/6, 1/2, 5/6, 1}; found blind from a wrong guess ({1/2}). A clean selection-effect instance.
• Integrable heavy tops — the Painlevé condition run over inertia/mass configurations yields exactly three (Euler, Lagrange, Kovalevskaya). The Painlevé test is literally run the compatibility condition until it fails. A clean instance.
• Chen–Liu squares involution — the negative control. Same silhouette as Franklin (perimeter-minimisation characterising fixed points) but no shared mechanism, so it fails to fit. Filed as the counterexample the probe needs: a pattern with no filed false positive is untested against its own.

E8/Lie chat: internal rediscoveries and self-policing

Negative-control profile — the chat mined itself honestly, mostly self-policed, very little to file. The job here was confirming its restraint.

• Item 5: the E8-is-everything retraction was stated locally and recurred 4–5 times because it was never promoted to a stated principle.
• The meta-point the chat surfaced: a correction stated locally recurs; only one promoted to a stated principle stops recurring. That is the mechanism of compression loss, and why unmarked retractions come back.
• Item 7: the Monster/Moonshine → physics leap is bucket 3, left in chat, not filed; the chat self-flagged it as its least-disciplined moment — the negative-control profile working at source.
• Items 1–4, 8: standard Lie theory (simple roots as basis, Weyl group as reflection group, Dynkin classification, E8 self-duality) — note-candidates at most, check before minting, none to the spine.
• Bucket 3, do not file: E8-is-everything (A1, A6, D5); consciousness overreach (A9); theory-of-everything overreach (D5).
• Real mathematics to check against what is filed: the root system as the group's own reflection combinatorics, with the Weyl group as its finite shadow (A2); Langlands as root systems on both sides, harmonic analysis controlling arithmetic (B7 — connect to [[EX-11]]); gauge symmetry as redundancy rather than a thing (B3 — check against the symplectic-has-no-local-data content).

Compression-loss finding: recorded evidence

The H = xp / Berry–Keating self-instance, from the Wallis/NS chat. Stated once, in the Berry–Keating turn, as the generator of the spiral transform; never promoted to a principle; and returned seven sections later as if new. Filed as recorded evidence that the mechanism — local corrections recur, only promoted ones stop — is real: the chat that diagnosed the failure mode in its own mining pass exhibits it. Diagnosis and disease in one document.

See also

• [[N-021]] — the selection-effect evidence ledger; owns the tally and the elevation question.
• notes/rediscovery-inventory-continued-fractions.md, N-022 — the other per-chat inventories, deliberately not merged with this one.