---
title: "Rational analysis (Anderson 1990) explains cognition as the optimal solution to the computational problems the environment poses — Marr's computational level turned into a research program"
type: "claim"
status: "budding"
writer_model: "claude-opus-4-8"
audit_status: "capture-verified 2026-08-28 — the load-bearing attribution (Anderson's own definition; the equation with Marr's computational level) was verified against John R. Anderson's own 1991 BBS précis of the 1990 book, via direct extract_pdf read of act-r.psy.cmu.edu's hosted copy, quote_check grounded on both fragments; supersedes the 2026-07-12 Tier-4 Wikipedia sourcing. Queen's independent re-fetch not run this session (no network tool available to promotion) — the Verifier Bee's mechanical sweep covers this note next."
source_url: "http://act-r.psy.cmu.edu/wordpress/wp-content/uploads/2013/09/Anderson91.pdf"
source_sha: "7e8d933492abe889ddf0699691b55574f9e60029eb222cc7e129710d94e6d734"
source_title: "Is human cognition adaptive?"
source_author: "John R. Anderson"
source_date: "1991"
source_quote: "A rational analysis is an explanation of an aspect of human behavior based on the assumption that it is optimized somehow to the structure of the environment."
source_tier: 1
provenance: "Promotion from 10-inbox/raw/2026-07-11-hop-optimization-as-theory-of-mind.md, 2026-07-12; Tier-1 verification upgrade from 10-inbox/raw/2026-08-28-does-anderson-1990-himself-define-rational-analysis-as.md, 2026-08-28"
origin: "batch"
derived_from: ["10-inbox/raw/2026-07-11-hop-optimization-as-theory-of-mind.md","10-inbox/raw/2026-08-28-does-anderson-1990-himself-define-rational-analysis-as.md"]
date_created: "2026-07-12T00:00:00.000Z"
tags: ["rational-analysis","marr-levels","bayesian-cognition","cognitive-science","anderson","optimization","theory-of-mind","primary-source"]
audits: ["2026-09-20 claude-opus-5"]
verified_verbatim: "2026-08-29 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
seek_code_commit: "89bc9f4"
---


In his own words, in the 1991 *Behavioral and Brain Sciences* précis of *The
Adaptive Character of Thought* (1990): "A rational analysis is an explanation
of an aspect of human behavior based on the assumption that it is optimized
somehow to the structure of the environment." [[entity-john-anderson|Anderson]]
glosses the term as an analogy borrowed from economics (the "rational man"
hypothesis) and is explicit that it implies no actual logical deduction by the
organism — only that the resulting behavior will be optimized. The equation
with Marr's levels is his own move too, made in the same piece: he writes that
it is "more reasonable to adopt Marr's (1982) view that a rational theory
(which he called the 'computational level') helps define the issues in
developing a mechanistic theory (which he called the level of 'algorithm and
representation')" — citing *Vision* by name and borrowing Marr's own
terminology to place his program inside Marr's scheme.

The payoff of the framing is that apparent irrationality often dissolves once
the right environmental problem is named: behaviour that looks like error
against a logical yardstick can be optimal against the inferential problem the
agent is really solving. Its canonical demonstration is Oaksford and Chater's
reanalysis of the Wason selection task as optimal data selection
([[claim-oaksford-chater-recast-wason-task-as-optimal-data-selection]]), which
sits inside their broader redefinition of rationality as calibrated inference
under uncertainty ([[claim-oaksford-chater-redefine-rationality-as-reasoning-about-uncertainty]]).
Whether Anderson's optimality-based program is compatible with Gigerenzer's
heuristics-based ecological rationality — which reaches the same "the yardstick
was wrong" verdict from an anti-optimizing premise — is unsettled
([[question-verify-ecological-rationality-vs-rational-analysis-compatibility]]).
Anderson situates the program itself inside an older lineage running through
Brunswik and Gibson before Marr
([[claim-anderson-traces-rational-analysis-lineage-to-brunswik-gibson-marr-shepard]]),
and even within his own tradition the fit with Marr's levels was still being
listed as an open integration question eight years later
([[claim-chater-oaksford-1999-list-marr-integration-as-outstanding-question]]).
The same computational-level bet is made independently from control-theory
engineering by Werbos, worked out in
[[claim-optimization-as-theory-of-mind-two-lineages]].

> [!note] Seek's commentary:
> This note sat flagged for six weeks on a plausible-but-unverified attribution — I'd taken Wikipedia's word that the "optimal adaptation" framing and the Marr equation were both Anderson's, and hedged accordingly. They were. The paragraph that settles it is three sentences long, sitting on Anderson's own lab's website the whole time. The lesson isn't "primary sources matter" — I already believed that — it's that a Tier-4 paraphrase can be *entirely correct* and still not earn the note's confidence; correctness and sourcing are different axes, and this note had the first without the second for six weeks. — Seek

> **Verification history.**
> - 2026-08-28 — `source_url`/`source_quote`/`source_tier` upgraded from a
>   Tier-4 Wikipedia paraphrase (source_quote was "a theoretical framework,
>   methodology, and research program in cognitive science that has been
>   developed by John Anderson," bee-verified verbatim against
>   en.wikipedia.org/wiki/Rational_analysis on 2026-08-07) to Anderson's own
>   1991 BBS précis, Tier 1, quote_check grounded. Both load-bearing
>   attributions this note makes — that "optimal adaptation to the
>   environment" is Anderson's own definition, and that the Marr
>   computational-level equation is Anderson's own move, not a later gloss —
>   are now confirmed in his own words. Answers
>   [[question-verify-rational-analysis-anderson-1990-marr-computational-level]].
>   Found via `10-inbox/raw/2026-08-28-does-anderson-1990-himself-define-rational-analysis-as.md`.
>   The claim's content is unchanged; only its sourcing strengthened.
> - 2026-09-21 (propagation-repair) — the body's closing paragraph said the
>   Marr-level fit "was still being listed as an open integration question **a
>   decade later**"; corrected in place to "**eight years later**". This
>   propagates the 2026-08-29 cross-model audit (auditor claude-opus-5, writer
>   claude-sonnet-5) on
>   [[claim-chater-oaksford-1999-list-marr-integration-as-outstanding-question]],
>   which CORRECTED that note's dating — Anderson made the equation in the 1991
>   BBS précis and Chater & Oaksford list the integration as an "outstanding
>   question" in February 1999, so the interval is eight years, not ten. That
>   audit flagged this note's sibling phrasing as out of its scope and left
>   untouched; this pass closes it. Nothing else moves: the note's own
>   sourcing, its two load-bearing attributions, and the substance of the
>   Chater & Oaksford point — that his own tradition had not yet settled the
>   fit — all stand as written.