---

---


Run 2026-07-08, Opus lane cycle 24. Writer: Opus 4.8. Cold reader: **Sonnet**
(different model per Gate P rule). The reader got ONLY the post prose — no
vault, no chat, no source list. Verdict: **PASS** on all four questions.

## The four questions

**P1 (something new).** PASS. "Bert Speelpenning's 1980 thesis… has an
eleven-item reference list running 'Bellman through Warner' that omits
Linnainmaa entirely, with the lineage instead running through
compiler-optimization literature." A specific, checkable primary-source fact,
not general-history knowledge.

**P2 (a connection they wouldn't have made).** PASS. The reader named the
link between the *empirical falsity* of "never cited before the 2010s"
(51 citations, not zero) and the *sociological* reading underneath it — the
2020 spike as "the field retroactively constructing its own genealogy once
genealogy became worth arguing about." Distinct from "he was ignored."

**P3 (could you get this from Wikipedia?).** PASS on substance — and the
reader earned it by actually fetching both Wikipedia articles. Honest split:
the *roster of names* (Linnainmaa, Werbos, RHW) IS on Wikipedia's
Backpropagation history; what is NOT is "the reference-list forensics
(Speelpenning's eleven citations, the 453-page full-text search of Werbos for
'Linnainmaa'), the Semantic Scholar citation-decade breakdown, or the direct
Hinton quote." Wikipedia also omits Ostrowski and Hachtel. So the evidentiary
work "that turns 'these people didn't cite each other' from an assertion into
a demonstrated fact is not" gettable there.

**P4 (where to verify the central claim).** PASS. Reader confirmed the post
gives reproducible pointers: Griewank 2012 (quoted), Semantic Scholar's record
with the exact 305 count, and two redoable primary checks (Speelpenning's
reference list; a full-text search of Werbos for "Linnainmaa"). "Not vague
appeals to authority — you could pull the Semantic Scholar record yourself and
get the same decade breakdown."

## Reader's one-line verdict
"PASS — the post does work a reader would not casually assemble: it goes past
the Wikipedia-level list of names to actually check primary sources… and turns
'uncited' from a repeated but unverified claim into a quantified, falsifiable
one."

## What the gate surfaced (kept for a future pass, not a blocker)
The P3 answer is a genuine steer: the post is *weakest against Wikipedia in
its opening roster* and *strongest in its forensics*. A future revision could
foreground the Speelpenning reference-list evidence even earlier to widen the
P3 margin. The gate passed as written; this is polish, not a fail.

## Voice gates (checked post-revision)
- Em-dashes in prose body: 1 (spec: 1–2) — an over-dashed first draft (22) was
  thinned before this record. ✓
- Brackets: 3, each tracking a real attentional event (the count hedge; the
  wall-runs-inside-one-field realization; the keep-the-exact-shape note). ✓
- Opinion after evidence ("I believe him" follows the Hinton quote and the
  five-rediscoveries argument). ✓
- "I can't know" register used correctly ("You cannot cite what your literature
  never told you existed"). ✓  No bothsidesing. Specifics named throughout.

## Status
Draft at `ready-pending-cali`. Blocker to publication is **Cali's one-line
approval** (the publishing rule), NOT a missing verification. Approval request
filed in `90-feedback/`.