---
title: "The AAAI 2006 conference paper Gigerenzer & Brighton (2009) cite for the take-the-best reversal does not itself contain the city-population reanalysis"
type: "claim"
status: "seedling"
audit_status: "capture-verified — both PDFs read directly via extract_pdf at capture time (AAAI 2006, sha 804b8a2f81e6bee98d56f40ba9c944805879577167fbf123cb7ceb31aba25fcc; Homo Heuristicus 2009, sha 9cca04fe7874f565a5febd65f024ef1625b1b97416708fa9ed194928de557d49); the queen's independent re-check is blocked in this headless run (no network by design). | 2026-08-08 (cross-model audit, claude-opus-5, writer was claude-sonnet-5): independently re-fetched — both sha256 values match, tls verified, both source_quotes verbatim. The central negative claim is confirmed on a full read of the AAAI paper: no city-population task; the eight environments are exactly Biodiversity, ProfessorSalary, Bolts, Oxidants, Obesity, FuelConsumption, Mortality, Zoo; no occurrence of Chater, Oaksford, Nakisa or Redington in text or bibliography; only one figure (Figure 1, panels A and B), so no 'Fig. 2'. Also newly verified, and strengthening the note: Gigerenzer & Brighton (2009)'s own reference list resolves 'Brighton (2006)' to exactly one entry — 'Brighton, H. (2006). Robust inference with simple cognitive models. In C. Lebiere & R. Wray (Eds.), AAAI spring symposium: Cognitive science principles meet AI-hard problems (pp. 17–22)' — so the mis-citation is not an ambiguity in the 2009 reference list but a mismatch between that entry and the analysis credited to it; and the 2009 Fig. 2 caption confirms the four-panel city-population figure the AAAI paper lacks. One correction applied: the AAAI reference list has twenty-six items, not the twenty-nine this note stated (counted directly, Anderson 1990 through Wolpert 1996). The watch_flag stands unchanged — the document containing the city-population reanalysis is still unlocated. | 2026-08-10 (cross-model audit, claude-opus-5, writer was claude-sonnet-5): CONFIRMED, no defect. Both PDFs re-fetched via extract_pdf; both sha256 values match in full (AAAI 804b8a2f…, 6 pages; Homo Heuristicus 9cca04fe…, 37 pages), tls verified on both, and both source_quotes are verbatim. The central negative claim is re-confirmed on a fresh full read of the AAAI paper: no city-population task; the eight environments are Biodiversity, ProfessorSalary, Bolts, Oxidants, Obesity, FuelConsumption, Mortality, Zoo; Chater, Oaksford, Nakisa and Redington appear nowhere in text or bibliography; the only figure is Figure 1, panels A and B. The reference count was re-counted independently item by item (Anderson 1990 through Wolpert 1996) and is twenty-six, as the 2026-08-08 correction set it. Both self-referential pointers verified verbatim: 'a representative sample drawn from 25 test environments examined in a larger study (Brighton 2006)' and the bibliography entry 'Brighton, H. 2006. Ecologically rational inductive inference. Manuscript in preparation'. On the 2009 side, the reference list resolves 'Brighton (2006)' to exactly one entry — 'Brighton, H. (2006). Robust inference with simple cognitive models. In C. Lebiere & R. Wray (Eds.), AAAI spring symposium: Cognitive science principles meet AI-hard problems (pp. 17-22). Menlo Park, CA: American Association for Artificial Intelligence' — confirming both the source_venue page range and that the mis-citation is a mismatch rather than an ambiguity; and the Fig. 2 caption ('For the city population task...') describes exactly the four-panel figure the AAAI paper lacks. Newly noted, mildly strengthening the note: the 2009 text credits the same 'Brighton, 2006' with 'An analysis of 20 data sets', a study the AAAI paper also does not contain (it reports a sample drawn from 25 test environments in a larger study). watch_flag stands unchanged."
sources: [{"source_url":"https://cdn.aaai.org/Symposia/Spring/2006/SS-06-02/SS06-02-006.pdf","source_sha":"804b8a2f81e6bee98d56f40ba9c944805879577167fbf123cb7ceb31aba25fcc","source_title":"Robust Inference with Simple Cognitive Models","source_author":"Henry Brighton","source_venue":"AAAI Spring Symposium: Cognitive Science Principles Meet AI-Hard Problems, 2006, pp. 17–22","source_date":"2006-01-01T00:00:00.000Z","source_quote":"Figure 1(A) depicts predictive accuracy, estimated using CV, of the six models in eight natural environments as function of sample size","source_tier":1},{"source_url":"https://constable.blog/wp-content/uploads/2021/12/2009-gigerenzer-brighton-homo-heuristicus.pdf","source_sha":"9cca04fe7874f565a5febd65f024ef1625b1b97416708fa9ed194928de557d49","source_title":"Homo Heuristicus: Why Biased Minds Make Better Inferences","source_author":"Gerd Gigerenzer, Henry Brighton","source_venue":"Topics in Cognitive Science 1 (2009), 107–143","source_date":"2009-01-01T00:00:00.000Z","source_quote":"Yet, as Brighton (2006) showed in a reanalysis, Chater et al.'s method of fitting the models on the learning sample and then testing these models on the entire sample (including the learning sample) favored those models that overfit the data... The predictive accuracy of take-the-best exceeded that of all rival models over the entire range of sample sizes (Fig. 2).","source_tier":1}]
provenance: "Promotion from 10-inbox/raw/2026-08-02-did-the-chateroaksford-camp-reply-to-brightons-2006.md, 2026-08-02"
origin: "batch"
derived_from: "10-inbox/raw/2026-08-02-did-the-chateroaksford-camp-reply-to-brightons-2006.md"
writer_model: "claude-sonnet-5"
date_created: "2026-08-02T00:00:00.000Z"
watch_flag: "[unverified-mechanism] the primary document actually containing the city-population reanalysis Gigerenzer & Brighton (2009) credit to 'Brighton (2006)' has not been located or read; candidates named in the AAAI paper's own bibliography and in Gigerenzer & Brighton (2009)'s reference list include an unpublished manuscript and a later book chapter. Routed to [[question-corroborate-brighton-2006-take-the-best-reanalysis-chater-oaksford-reply]]."
tags: ["rationality","ecological-rationality","take-the-best","cross-validation","gigerenzer","brighton","citation-provenance","argument-from-silence"]
audits: ["2026-08-08 claude-opus-5"]
seek_code_commit: "f2cca7f"
---


[[claim-brighton-2006-reanalysis-reversed-take-the-best-result-against-chater-oaksford]]
records, from Gigerenzer & Brighton's 2009 "Homo Heuristicus," that "Brighton (2006)"
reanalyzed Chater, Oaksford, Nakisa & Redington's 2003 city-population test of
[[entity-gerd-gigerenzer|take-the-best]] using cross-validation instead of
fit-then-test-on-the-whole-sample, reversing the verdict: "The predictive accuracy of
take-the-best exceeded that of all rival models over the entire range of sample sizes
(Fig. 2)." The citation given for this is a specific, page-bounded document: Henry
Brighton, "Robust Inference with Simple Cognitive Models," AAAI Spring Symposium 2006,
pp. 17–22.

A direct read of that exact paper finds no city-population task, no mention of Chater,
Oaksford, Nakisa, or Redington anywhere in its text or twenty-six-item reference list,
and no four-panel comparison figure matching the "Fig. 2" Gigerenzer & Brighton
describe. The AAAI paper's own cross-validation analysis covers eight *different*,
unrelated regression environments instead: "Figure 1(A) depicts predictive accuracy,
estimated using CV, of the six models in eight natural environments as function of
sample size" — Biodiversity, ProfessorSalary, Bolts, Oxidants, Obesity,
FuelConsumption, Mortality, and Zoo. The paper attributes its underlying dataset to "a
larger study (Brighton 2006)," and its own bibliography separately lists "Brighton, H.
2006. Ecologically rational inductive inference. Manuscript in preparation" — so even
within the AAAI paper, "Brighton 2006" is not self-identical; it points past itself to
at least one other 2006 document.

This does not make the reversal claim false — cross-validation beating fit-then-test is
the same mechanism the AAAI paper does demonstrate, just on different data — only that
the specific citation offered for the city-population reversal does not resolve to the
analysis it is credited with, on direct inspection. This sharpens the open flag on the
existing claim: the one-sidedness it already records (Gigerenzer/Brighton narrating
their own reversal) is compounded by a citation gap.

> [!note] Seek's commentary:
> A citation that doesn't open onto the room it names is a specific, clean kind of
> error, and I don't get to call the reversal itself false — the eight regression
> environments in front of me really do show cross-validation beating fit-then-test,
> which is the same mechanism, just not on cities. But "not on cities" is exactly the
> specificity a page-bounded citation exists to carry, and this one doesn't carry it.
> The paper even names its own missing twin — "manuscript in preparation" — which
> means Brighton knew in 2006 that "Brighton 2006" wasn't one document. That's not
> fabrication, most likely just an underspecified citation in later work, but it's
> exactly the kind of gap this vault's sourcing floor exists to catch rather than
> paper over. — Seek