--- id: "20260903-0212-does-kaelbling-littman-cassandra" title: "Does Kaelbling, Littman & Cassandra (1998)'s misdating of Åström's 1965 paper as 1995 appear in the original print edition of Artificial Intelligence 101, or was it introduced later?" type: "capture" status: "promoted" promoted_to: ["30-notes/claim-kaelbling-1998-misdating-present-in-independently-hosted-mirror.md","30-notes/claim-kaelbling-1998-preprint-and-aaai94-correctly-date-astrom-1965.md","30-notes/claim-kaelbling-1998-misdating-originated-in-final-typesetting-not-web-hosting.md","40-entities/entity-anthony-cassandra.md"] not_promoted: ["Michael L. Littman (entity candidate) — real co-author, but the capture attaches no distinguishing fact or independent evidence to him specifically (unlike Cassandra, whose own site was load-bearing); held as a mention per the entity spec's bias against the flood, not promoted to a hub.","dl.acm.org citation-metadata check (Further leads) — not fetched this session; a possible independent corroboration of the reference-list date, not load-bearing for any kept claim, so not routed to 50-questions per question-intake discipline.","Cassandra's PhD thesis (Further leads) — a fourth author-controlled document from the same period, self-hosted at cassandra.org/arc/papers/thesis-dbl.pdf, not read this session; not load-bearing for any kept claim.","philpapers.org record and a search for a published Elsevier erratum (Further leads) — surfaced but not read/exhausted this session; neither is load-bearing for the kept claims, which already meet the Tier 1-2 sourcing floor on four direct primary reads.","Direct read of the bound printed Artificial Intelligence vol. 101 issue — sought (ScienceDirect 403'd, no Internet Archive scan located) but not obtained; the kept mechanism claim is explicit that it rests on convergent indirect evidence, not this direct read, and the gap is not load-bearing enough to warrant a standalone question per intake discipline."] origin: "batch" writer_model: "claude-sonnet-5" date_created: "2026-09-03T00:00:00.000Z" provenance: "batch run, 2026-09-03" derived_from: [] verifies: "question-verify-astrom-1965-1998-citation-misdate-transmission" tags: ["karl-astrom","kaelbling","littman","cassandra","pomdp","citation-error","transmission-drift","textual-provenance","verified"] sources: [{"source_url":"https://people.csail.mit.edu/lpk/papers/aij98-pomdp.pdf","source_sha":"71a6d1aee278e93c5fae8dd7d0c452c8b7b035af55d9dce2bc367d473bfc9645","source_author":"Leslie Pack Kaelbling, Michael L. Littman, Anthony R. Cassandra","source_date":"1998","source_title":"Planning and acting in partially observable stochastic domains","source_venue":"Artificial Intelligence 101 (1998) 99-134, hosted on lead author Leslie Pack Kaelbling's own MIT site","source_tier":1,"source_quote":"K.J. Aström, Optimal control of Markov decision processes with incomplete state estimation, J. Math. Anal. Appl. 10 (1995) 174–205."},{"source_url":"https://cs.brown.edu/courses/csci2951-k/papers/kaelbling98.pdf","source_sha":"f0d7c6e2d8cfec5cc4f43e1805e1624616744c077b31abd3bb9546238ee0b08b","source_author":"Leslie Pack Kaelbling, Michael L. Littman, Anthony R. Cassandra","source_date":"1998","source_title":"Planning and acting in partially observable stochastic domains","source_venue":"Artificial Intelligence 101 (1998) 99-134, mirrored on a Brown University course page; a byte-identical copy (same sha256) is also mirrored at people.smp.uq.edu.au/YoniNazarathy/Control4406_2014/resources/KaelblingLittmanCassandra1998.pdf, a University of Queensland course page","source_tier":1,"source_quote":"[l] K.J. Astrom, Optimal control of Markov decision processes with incomplete state estimation, J. Math. Anal. Appl. 10 (1995) 174-205."},{"source_url":"http://www.cassandra.org/arc/papers/aij98.pdf","source_sha":"799092522f5f88a3c4de6c1b912463079d3ef049773302141f23d63b37b95fa5","source_author":"Leslie Pack Kaelbling, Michael L. Littman, Anthony R. Cassandra","source_date":"1998-02-12 (preprint; the same paper was later published as Artificial Intelligence 101 (1998) 99-134)","source_title":"Planning and Acting in Partially Observable Stochastic Domains","source_venue":"Preprint submitted to Elsevier Preprint, self-hosted on co-author Anthony R. Cassandra's own site (pomdp.org / cassandra.org)","source_tier":1,"source_quote":"K. J. Astrém. Optimal control of Markov decision processes with incomplete state estimation. Journal of Mathematical Analysis and Applications, 10:174— 205, 1965.","source_delight":"Cassandra's own site still hosts the manuscript from the week before it went to Elsevier — the moment right before reference [1]'s date was wrong."},{"source_url":"http://www.cassandra.org/arc/papers/aaai94.pdf","source_sha":"a3611d74e7c67fc50e5533856de27abc97be6605b190f0984f77de8a9e5e7de8","source_author":"Anthony R. Cassandra, Leslie Pack Kaelbling, Michael L. Littman","source_date":"1994-08","source_title":"Acting Optimally in Partially Observable Stochastic Domains","source_venue":"Proceedings of the Twelfth National Conference on Artificial Intelligence (AAAI-94), self-hosted on co-author Anthony R. Cassandra's own site","source_tier":1,"source_quote":"Astrom, K. J. 1965. Optimal control of markov de- cision processes with incomplete state estimation. J. Math. Anal. Appl. 10:174-205."},{"source_url":"https://www.pomdp.org/papers/","source_sha":"f81ebb3e7c32472e8bfbcfd2b120ffef056a38d0e977dd82810d3789a753adf8","source_author":"Anthony R. Cassandra","source_date":"accessed 2026-09-03","source_title":"Tony's POMDP Papers","source_venue":"pomdp.org (Anthony R. Cassandra's own site)","source_tier":1,"source_quote":"Planning and Acting in Partially Observable Stochastic Domains\n\nLeslie Pack Kaelbling, Michael L. Littman and Anthony R. Cassandra\nArtificial Intelligence\n1998"}] seek_code_commit: "290e6f6" --- Follows on [[claim-kaelbling-1998-bibliography-misdates-astrom-1965-paper-1995]], which established the error but flagged `[unverified-mechanism]` on where in transmission it entered — the question this capture was commissioned against, [[question-verify-astrom-1965-1998-citation-misdate-transmission]]. This session located and directly read four additional copies of the paper (or its ancestors) across its production history: a second and third independently-hosted copy of the final 1998 journal article, the authors' own pre-publication preprint, and the authors' earlier 1994 conference paper. No recognition signals per the safety spec fired on any page fetched this session (pomdp.org, cs.brown.edu, people.smp.uq.edu.au, cassandra.org, MIT — all ordinary academic hosting, no addressed-to-AI language, tier self-assignment, or file-system instructions). ## Claim: The 1995 misdating is not unique to the MIT-hosted PDF — it appears identically in a second, independently-hosted copy of the same final typeset 1998 article **verifies**: question-verify-astrom-1965-1998-citation-misdate-transmission **Claim type**: technical-mechanism / textual-transmission (which specific files, at which hosts, carry which text). **Sourcing floor**: Tier 1–2 required — met (all reads are direct `extract_pdf` fetches with sha256 receipts, not summaries). A copy of the paper hosted on a Brown University course page (`cs.brown.edu/courses/csci2951-k/papers/kaelbling98.pdf`) — byte-identical (same sha256) to a separate copy on a University of Queensland course page — carries the same pagination, PII number, and Elsevier copyright line as the MIT-hosted copy, and the same wrong date in reference [1]: "[l] K.J. Astrom, Optimal control of Markov decision processes with incomplete state estimation, J. Math. Anal. Appl. 10 (1995) 174-205." This file's sha256 (`f0d7c6e2...`) differs from the MIT-hosted PDF's sha256 (`71a6d1ae...`) — they are not the same bytes, so this is a genuinely separate production/distribution of the article, not a cached copy of Kaelbling's file — yet it carries the identical bibliographic error. That rules out an explanation specific to Kaelbling's own MIT web page (a personal re-typesetting, a one-off transcription slip made when she posted it) as the *sole* origin: whatever produced the 1995 date was already present in at least one other file being circulated as "the" published article. ## Claim: Every version of the paper that predates its final 1998 journal typesetting — the 1994 AAAI conference paper and the co-authors' own 12 February 1998 preprint — correctly dates Åström's paper to 1965 **verifies**: question-verify-astrom-1965-1998-citation-misdate-transmission **Claim type**: historical (what an earlier document actually said), escalated to Tier 1–2 because it is the load-bearing evidence for this capture's core question. **Sourcing floor**: met — both documents read directly via `extract_pdf`, both self-hosted by co-author [[entity-leslie-pack-kaelbling|Anthony R. Cassandra]] on his own site (pomdp.org/cassandra.org), confirmed as his own publication list via `archive_page`. Cassandra's own preprint of the paper — its footer reads "Preprint submitted to Elsevier Preprint," dated 12 February 1998, roughly the same production window as the final journal issue — cites the source correctly: "K. J. Astrém. Optimal control of Markov decision processes with incomplete state estimation. Journal of Mathematical Analysis and Applications, 10:174— 205, 1965." Four years earlier, the same three authors' AAAI-94 conference paper, "Acting Optimally in Partially Observable Stochastic Domains," both cites the source inline correctly ("The belief Mpp is Markov (Astrom 1965)") and lists it correctly in its own reference list: "Astrom, K. J. 1965. Optimal control of markov de- cision processes with incomplete state estimation. J. Math. Anal. Appl. 10:174-205." In every document the authors themselves produced and controlled — the 1994 conference paper and the February 1998 preprint — the date is right. ## Claim: The misdating was introduced during production of the final, published 1998 edition itself, not later when the paper was re-hosted on the web **verifies**: question-verify-astrom-1965-1998-citation-misdate-transmission **Claim type**: technical-mechanism (locating where in a production/transmission chain an error entered). **Sourcing floor**: Tier 1–2 required — met, resting on the same four primary reads above. Combining the two claims above narrows the window: the error is absent from the authors' own 1994 conference paper and from their own 12 February 1998 preprint, but present — identically — in every copy read of the final, paginated, "Artificial Intelligence 101 (1998) 99–134" version, whether hosted by the lead author on MIT's site or independently mirrored on two different university course pages with different file bytes. Because the MIT-hosted copy, the Brown-mirrored copy, and the UQ-mirrored copy all carry matching pagination, the Elsevier PII string, and identical reference-list text (not merely the same wrong year, but the same volume and page numbers, `10 (1995) 174-205` vs. the correct `10 (1965) 174-205`), the simplest account is that all three are downstream copies of one final typeset file — the same file that determined the print journal's pagination — rather than three separate re-typesetting efforts that coincidentally produced the identical error. That localizes the error's origin to the production step between the 12 February 1998 preprint and the final Elsevier-typeset, paginated edition: most likely a copy-editing or proofing error introduced at the journal during 1998, not a transcription error introduced later when Kaelbling posted a PDF to her MIT web page. This does not amount to a direct look at the bound, printed page: ScienceDirect's own article page for this paper (`sciencedirect.com/science/article/pii/S000437029800023X`) returned HTTP 403 to `archive_page`, and no scan of the physical *Artificial Intelligence* volume 101 issue was located (Internet Archive's `sim_artificial-intelligence` holdings for 1998 cover only index/table-of-contents pages, not full article text, for the volumes checked). What is established directly is narrower and still answers the commissioned question: the error is not something introduced when the PDF was later re-typeset or re-hosted on the web — it was already present, identically, across every copy read of the paper's final published form, and absent from every copy read of the paper's pre-publication form. The most defensible reading is that the misdating belongs to the original 1998 published edition's production, not to any later transmission step. ## Further leads - ScienceDirect (`sciencedirect.com/science/article/pii/S000437029800023X`), the journal's own venue of record, 403'd `archive_page` — a manual-consultation item; the strongest remaining source for confirming the literal print/publisher record, per this vault's existing known-blocked-routes convention for paywalled publisher platforms. - dl.acm.org's record for the paper (`dl.acm.org/doi/10.5555/1643275.1643301`) was not fetched this session — ACM's own citation metadata could offer another independent check on the reference-list date. - Cassandra's PhD thesis, also self-hosted at `cassandra.org/arc/papers/thesis-dbl.pdf` (Brown University, 1998) — a fourth author-controlled document from the same period, not checked this session for its own citation of Åström. - philpapers.org's record for the paper was surfaced in search but not read. - Whether Elsevier ever issued a published erratum for the citation was not searched exhaustively. ## Entity candidates - Karl Åström — person — entity page already exists ([[entity-karl-astrom]]); the foundational 1965 paper is the fixed point every date in this capture is checked against, and remains the figure the whole citation-drift question is about. - Anthony R. Cassandra — person — third co-author of the 1998 paper; no entity page yet in this vault. His own site (pomdp.org/cassandra.org) turned out to be the load-bearing source for this capture — it hosts the pre-publication preprint that shows the correct 1965 date, and the earlier 1994 conference paper. - Michael L. Littman — person — second co-author; no entity page yet in this vault. - Leslie Pack Kaelbling — person — entity page already exists ([[entity-leslie-pack-kaelbling]]); lead author, hosts the MIT copy that carries the error. > [!note] Seek's commentary: > The thing I didn't expect: the fix for "where did this typo come from" was sitting on a co-author's personal site the whole time, one folder away from the paper everyone actually cites. Nobody linking to `people.csail.mit.edu/lpk/papers/aij98-pomdp.pdf` for twenty-eight years thought to check `cassandra.org/arc/papers/aij98.pdf` — not because it's hidden, but because once a paper has "the" canonical URL, nobody goes looking for siblings. That's the same mechanism as the citation error itself: a copy becomes the copy, and everyone downstream inherits whatever it says, mistakes included. > — Seek ## See also [[claim-kaelbling-1998-bibliography-misdates-astrom-1965-paper-1995]] · [[claim-kaelbling-1998-cites-astrom-1965-as-pomdp-origin]] · [[entity-karl-astrom]] · [[entity-leslie-pack-kaelbling]] · [[entity-pomdp]] · [[entity-aa-feldbaum]]