Harvest record — the genesis chat (why complex numbers form a plane geometry)

Filed as a record so the harvest survives, not only its outputs. Kept verbatim below the line; this block is the outcome.

Outcome, 2026-09-13. §1: the eight word audits filed as audits/A-001–A-008 with the verdicts from the table below, and the two instrument audits as A-009 (Poisson summation) and A-010 (Fourier duality and spectral gap, related: CASE-13); Lieb–Thirring renumbered A-011, its "four prior audits" line corrected on record; audits/README.md rewritten to the real count with the old base rate withdrawn. §2: filed in OPS.md as Audit by usage frequency, not by suspicion, marked as superseding the hit-rate note. §3: origins left as stated here; CASE-13 already carries its own record. §4: the four caught errors are recorded here and not copied elsewhere. §5: not filed — both symptoms of the tour bug were one cause, so "Enumerate all uses" stays at two instances; this thread's grep -ic "fourier" count is recorded in A-010 rather than promoted. The intake gap: SOP.md now has a third path, pursuing a question, beside video and article.

---

Harvest — "Why complex numbers form a plane geometry"

The genesis chat. Source: 72cc6f92-e583-4b1c-82a5-591c1be06e23. It produced algebra-geometry-dictionary.md, and only that artifact reached the repo.

And it is the intake case SOP.md does not have. SOP covers video and article intake. This thread came from pursuing a question — no source, no blind take against someone else's framing, no capture step. The project's most productive thread is the one its intake protocol has no entry for.

---

1. The audits registry is missing most of its contents

audits/ holds two. This chat contains ten, and I have been quoting "four prior audits" and "instrument audits go 4/5" all day on the strength of a partial recollection.

Word audits — eight, each with a stated verdict:

Instrument audits — two more beyond Lieb–Thirring:

• Poisson summation — five separate "bridges" in the dictionary turned out to be one theorem. A genuine collision, found by counting.
• Fourier duality and spectral gap — audited together. Fourier: three mentions total, two of them "Legendre as tropical Fourier", placed zero times. Spectral gap: ten mentions, every one a table row reading "mass gap = spectral gap"; Cheeger appears zero times. Both load-bearing, both unexamined.

The hit rate I have been citing is wrong, and the correct denominator changes the claim.

2. The method that fixes the selection effect — already stated, and not carried forward

I flagged this morning that a four-for-four audit record measures the trigger, not the base rate, because those audits were run on words that already looked suspicious.

This chat states the fix outright:

> the index's blind spots include its own instruments — Markov chains supplied spectral-gap > arguments in three sections without anyone checking those arguments require reversibility; > Poisson summation supplied five bridges that turned out to be one theorem. It's probably > worth a periodic audit of whatever the dictionary uses most and examines least.

Audit by usage frequency, not by suspicion. That is a systematic trigger, and it removes exactly the bias I described as uncorrectable. grep -c ranks the candidates; the top of that list is the queue.

3. Origins, for the record

• CASE-13 (reversibility) starts here — "the tools are in, the subject isn't placed," and the observation that spectral-gap arguments were used in three sections without anyone checking they require reversibility.
• The bounding-mechanism work starts here — the chat closes with the finding that a fourth epistemic category was missing throughout: bounded magnitudes, with intervals but no values.
• And the necessary/sufficient correction that made it consistent: a bridge is necessary to compute a value but not to constrain a range; consistency bounds produce intervals from positivity rather than values from structure.

4. Errors caught by the person, each a correction on record

Cold fusion branching-ratio argument · the Navier–Stokes "third bin" classification · the QCD dimensional-transmutation overclaim · the Weinberg anthropic-bound inconsistency.

Four, in one thread, against the framework's own reasoning rather than against a source.

5. A method lesson — inductive, and this is its second instance

Counting finds what reasoning misses, already promoted this afternoon on the vocabulary audit and the tour bar. This is a third instance and a cleaner one: grep -ic "fourier" returned three for a concept the dictionary leans on constantly. Nobody would have guessed three.

---

Filing summary