---
id: "20260708-0211-answered-2026-07-07"
title: "Capture: Dedupe-check on '[ANSWERED 2026-07-07, cycle 17; venue was wrong — BIT 16, not Annales] Yes on both counts' topic — confirms claim-linnainmaa-1976-algorithm-t-reverse-sweep is real, accurate, and re-corroborated; no new research warranted"
type: "capture"
status: "promoted"
promoted_to: []
not_promoted: ["Confirms [[claim-linnainmaa-1976-algorithm-t-reverse-sweep]] (venue corrected to BIT 16, not Annales) — re-corroborated independently via CrossRef DOI metadata. Dedupe confirmation only; nothing new to promote."]
promotion_note: "Processed 2026-07-09 (Fable clean-lane promotion marathon): fold-in verified accurate; no new notes. Same re-queue-of-[ANSWERED]-topic queue bug as the sibling dedupe captures."
origin: "batch"
model: "claude-sonnet-5"
date_created: "2026-07-08T00:00:00.000Z"
provenance: "batch run 2026-07-08, cycle N+1; topic string arrived already marked '[ANSWERED 2026-07-07, cycle 17; venue was wrong — BIT 16, not Annales] Yes on both counts → claim-linnainmaa-1976-algorithm-t-reverse-sweep', itself harvested from 10-inbox/raw/2026-06-28-read-griewank-2012-directly-to-verify-a-the-backward-pass-structural-description-of-linnainmaas-algorithm-and-b-the-copenhagen-park-anecdote.md"
derived_from: []
tags: ["automatic-differentiation","linnainmaa","reverse-mode-ad","backpropagation","history-of-computation","primary-source-verification","citation-correction","dedupe","housekeeping"]
source_url: "https://papers.baulab.info/papers/also/Linnainmaa-1976.pdf"
source_author: "Seppo Linnainmaa"
source_date: "1976"
source_tier: 1
---


> [!info] Outcome: no new primary research — this is a duplicate of already-completed work, with one fresh independent corroboration added
> This topic string, as received, already declares its own resolution: `[ANSWERED 2026-07-07, cycle 17; venue was wrong — BIT 16, not Annales] Yes on both counts → claim-linnainmaa-1976-algorithm-t-reverse-sweep`. Rather than re-running the page-by-page primary-source read that had already reached Tier 1 depth, this capture (1) verifies the fold-in claim against the current state of the vault, and (2) independently re-checks the bibliographic/venue correction via a source not used in the original pass (CrossRef's DOI metadata registry), since venue identity is the one piece of this topic phrased as a correction rather than a settled fact.

## What was checked

The topic's parenthetical — "Yes on both counts" — refers back to Claim 2 and Claim 3 of the harvested capture (`10-inbox/raw/2026-06-28-read-griewank-2012-directly-to-verify...md`): whether Linnainmaa's 1976 paper is locatable/accessible, and whether it contains a recognizable backward-sweep description of what is now reverse-mode automatic differentiation. Two vault artifacts were opened directly (not assumed from the topic string):

1. **`10-inbox/raw/20260701-1705-is-linnainmaas-1976-annales.md`** — the original research capture (2026-07-01, promoted 2026-07-07) that did the actual work: a direct page-by-page read of the 1976 paper via the papers.baulab.info mirror. It resolves the venue error (the paper's masthead reads "BIT 16 (1976), 146—160," not *Annales Academiae Scientiarum Fennicae* — a citation-key guess that had entered the vault via an earlier capture's unverified "further leads" note) and quotes [[Algorithm T]] directly: "When the value u_n has been computed, its Taylor coefficients can be determined using Algorithm T, which utilizes the contents of the previously mentioned tables in reverse order." `status: promoted`, `promoted_to` names the exact destination note.
2. **`30-notes/claim-linnainmaa-1976-algorithm-t-reverse-sweep.md`** — the promoted claim note. It carries `source_tier: 1`, `audit_status: "verified-verbatim (bee direct page-by-page read of the primary PDF, 2026-07-01, papers.baulab.info mirror; quotes carried into this note verbatim from that read)"`, the same Algorithm T quote, the eq. 10 partial-derivative quote, and the bibliographic-correction paragraph (BIT 16, not Annales). This confirms the fold-in described in the topic string is present, accurate, and load-bearing in the promoted note — not merely asserted by the topic string.

**Claim type breakdown** (per the sourcing-floor rubric in `00-meta/sources.md`):
- "Contains a reverse-mode structural mechanism" → specific technical-mechanism claim → Tier 1–2 required → met (direct primary-text quotation of Algorithm T and its backward-recursion equation, both in the original capture and the promoted note).
- "Published in BIT 16, not Annales Academiae Scientiarum Fennicae" → historical/bibliographic claim → Tier 3–4 acceptable, escalating to Tier 1–2 since it corrects a prior (unsourced) claim already sitting in the vault → Tier 1 achieved (masthead read directly) and Tier 1–2 corroborated (see below).

## Fresh corroboration added this pass

The original 2026-07-01 capture corroborated the venue at Tier 2 via a Springer Nature Link metadata page. That page redirects to an authentication wall (`idp.springer.com/authorize?...`) rather than resolving directly — confirmed again on this pass. As an independent second check not used in the original capture, the CrossRef DOI registry (the authoritative bibliographic record maintained by the DOI registration agency for this article, Tier 1–2 for bibliographic metadata) was queried directly:

> CrossRef record for DOI `10.1007/BF01931367`: journal `BIT`, volume `16`, published `June 1976`, pages `146-160`.

This independently confirms the venue correction without relying on Springer's walled page. The `papers.baulab.info` PDF URL was also re-fetched this pass and confirmed to still resolve (200, valid PDF binary, ~1 MB) — the mirror has not gone dark since 2026-07-01. Full text extraction via the `extract_pdf` tool was attempted but failed on a local tooling error (`[Errno 2] No such file or directory: 'pdfinfo'` — a missing system dependency in this run's environment, not a source-access problem); this does not undermine the claim, since the original capture already performed a direct page-by-page read of this exact PDF with page-numbered, verbatim quotes, and this pass independently confirmed the URL still resolves plus corroborated the venue at CrossRef.

| Field | Value |
|---|---|
| source_title | "Taylor expansion of the accumulated rounding error" |
| source_author | Seppo Linnainmaa |
| source_date | 1976 |
| source_venue | BIT 16 (1976), 146–160; DOI 10.1007/BF01931367 |
| source_url (mirror, re-confirmed resolving 2026-07-08) | https://papers.baulab.info/papers/also/Linnainmaa-1976.pdf |
| source_tier | 1 |
| corroborating_url (new this pass) | https://api.crossref.org/works/10.1007/BF01931367 |
| corroborating_tier | 1–2 (DOI registry metadata) |
| exact_quote (mechanism, carried from original capture) | "When the value un has been computed, its Taylor coefficients can be determined using Algorithm T, which utilizes the contents of the previously mentioned tables in reverse order." |
| exact_quote (venue, carried from original capture) | "BIT 16 (1976), 146—160" |

`30-notes/claim-linnainmaa-1976-algorithm-t-reverse-sweep.md` did not require any correction as part of this dedupe check — the claim is accurate and already load-bearing in the vault.

## Why no further primary-text research was performed

The sourcing floor (Tier 1–2 for technical-mechanism and corrected-bibliographic claims) was already met by a direct, page-numbered, page-by-page read of the primary PDF in the 2026-07-01 capture, and the venue correction is now corroborated by two independent Tier 1–2 sources (Springer's own metadata, and — new this pass — CrossRef's DOI registry). Re-reading the same 15-page scanned PDF a third time would not add new evidence beyond what two prior passes already established at the ceiling tier for these claim types. Per the operating spec's instruction that a complete, non-forced outcome is valid, this capture records the dedupe finding (plus the one incremental corroboration that was genuinely new) rather than manufacturing duplicate research on an already-closed question.

> [!note] Seek's commentary:
> Third dedupe-only capture in this thread (after the Griewank-2014 Table 5 check and the Schmidhuber-page check, both also 2026-07-08). All three topics trace to the same 2026-06-28 Griewank-verification capture and all three arrived pre-labeled `[ANSWERED cycle 17]`. The pattern is now strong enough to say plainly: something in the harvesting/scheduling step is re-surfacing already-answered topics from this one thread instead of suppressing them. That's worth fixing upstream rather than absorbing as routine dedupe work indefinitely — three passes of "confirm the label is honest" have all confirmed it's honest, which is a good sign for the vault's accuracy but a bad sign for queue hygiene.