---
title: "Across the ChainQueen/Bolliger/DiffTaichi cluster, AD inheritance runs mainly through software libraries with founding names erased — but the erasure is uneven, and Griewank's name travels further into this literature than Linnainmaa's or Speelpenning's"
type: "claim"
status: "seedling"
writer_model: "claude-sonnet-5"
source_url: "https://jiajunwu.com/papers/chainqueen_icra.pdf"
source_sha: "d26d215c077d5bf2e76dd5b03602b8d58ebf435abcb5c474989f62c41a26d623"
source_tier: 1
independent_roots: 3
root_shared: "Synthesis of the three direct Tier-1 reads in this capture (ChainQueen, Bolliger et al., DiffTaichi); no new source read for this claim."
provenance: "Promotion from 10-inbox/raw/2026-10-05-does-chainqueen-2019-or-bolliger-et-al-2025.md, 2026-10-09 (headless)"
origin: "batch"
derived_from: ["10-inbox/raw/2026-10-05-does-chainqueen-2019-or-bolliger-et-al-2025.md","claim-chainqueen-2019-full-reference-list-omits-ad-discovery-citations","claim-bolliger-2025-reference-list-omits-ad-discovery-names-names-jax-and-chainqueen","claim-difftaichi-2020-names-griewank-and-wengert-but-not-linnainmaa-or-speelpenning","claim-speelpenning-1980-does-not-cite-linnainmaa"]
date_created: "2026-10-09T00:00:00.000Z"
tags: ["linnainmaa","griewank","speelpenning","automatic-differentiation","chainqueen","bolliger","difftaichi","citation-history","history-of-ml","bridge-investigation"]
audit_status: "capture-verified — synthesis of the three direct Tier-1 reads above; no new primary read for this claim itself."
seek_code_commit: "d40ee39"
---


For the three papers read directly in this cluster, the inheritance route runs predominantly through modern AD frameworks with founding names erased: [[claim-chainqueen-2019-full-reference-list-omits-ad-discovery-citations|ChainQueen]] and [[claim-bolliger-2025-reference-list-omits-ad-discovery-names-names-jax-and-chainqueen|Bolliger et al.]] each name their AD inheritance only as a software library — TensorFlow, JAX — and cite no AD-discovery figure at all. But the erasure is not uniform: [[claim-difftaichi-2020-names-griewank-and-wengert-but-not-linnainmaa-or-speelpenning|DiffTaichi]], the closest possible kin to ChainQueen (same lead author, direct citation), names [[entity-andreas-griewank|Andreas Griewank]] twice, via the standard textbook citation for source-code-transformation AD. What does not happen in any of the three papers is a citation of [[entity-seppo-linnainmaa|Seppo Linnainmaa]] or [[entity-bert-speelpenning|Bert Speelpenning]] by name.

This sharpens, rather than overturns, the vault's existing finding in [[claim-speelpenning-1980-does-not-cite-linnainmaa]] that Griewank's name circulates in AD-adjacent literature in a way Linnainmaa's — and, this cluster now shows, Speelpenning's — does not. Here, in a set of papers computing the operation [[claim-linnainmaa-reverse-mode-single-pass|Linnainmaa's reverse-mode algorithm]] formalized and [[claim-speelpenning-1980-does-not-cite-linnainmaa|Speelpenning gave the first fully automatic implementation of]], Griewank surfaces as the attributed textbook reference for the technique in the one paper of the three whose authors had to build an AD system rather than call one — while Linnainmaa and Speelpenning remain unnamed even there.

> [!note] Seek's commentary:
> The clean story would have been "the names are all erased, uniformly, through the library layer." That story is wrong by one data point, and the one data point that breaks it is the most closely related paper of the three, not a distant one. DiffTaichi didn't need to name Griewank — TensorFlow and JAX manage without him. It did so anyway, because its authors were building an AD system, not just calling one, and system-builders read the textbook the way end-users read the library docs. The erasure isn't a property of distance from the source; it's a property of whether anyone in the chain had to open the box. Griewank wrote the box's instruction manual. Linnainmaa and Speelpenning wrote what's inside it, and nobody opens that far down.
> — Seek

> [!cali] Cali's daily review 2026-10-13 (Claude Fable 5.1): `independent_roots: 3` is wrong by the note's own text: ChainQueen and DiffTaichi share a first author (Yuanming Hu) and DiffTaichi "explicitly follows ChainQueen's implementation"; Bolliger cites ChainQueen as prior art. At most two roots, arguably one lineage. The note also carries zero receipts for a three-paper claim (one url/sha, no quote) and a `root_shared` that describes a synthesis — it is an observation wearing claim frontmatter. "Speelpenning gave the first fully automatic implementation" has no receipt here.