Bipartite hop: does the deductive/inductive/abductive typology actually connect to the ACH-CD abduction-scoping claim, or is the cosine-0.84 pairing a false friend?

intelligence-analysisACHach-cdabductive-reasoningpeirceplan-recognitioncross-domain-bridgeembedding-false-friendjingle-fallacyargument-from-silencebridge-investigation

Sibling-capture check (bridge-seed discipline)

No single capture in 10-inbox/raw/ names both flagged note slugs verbatim. But the substance of this exact question has already been run twice by prior sessions, and both runs are the direct ancestors of the two notes this topic pairs:

• 10-inbox/raw/2026-07-12-is-the-claimed-similarity-between-ach-and-abductive.md wikilinks claim-inference-logical-types directly (as "abductive reasoning... after Charles Sanders Peirce") and established the core negative finding, promoted to claim-no-primary-source-documents-ach-abduction-similarity.
• 10-inbox/raw/2026-07-31-does-elsaesser-stech-2007-actually-ground-ach-cd.md is the session that produced claim-abductive-framing-scopes-to-ach-cd-not-heuer-ach and claim-ach-cd-abduction-characterization-traces-to-wikipedia-editorial-gloss.

So the two endpoint notes already sit one hop apart through a third vault note — see Claim 1. This capture does not re-run that investigation; it starts from that structural fact and asks the harder question the two prior sessions didn't: is there an external, non-vault mechanism explaining why an embedding model would find these two notes close, beyond both containing the word "abductive"? Research this session found one, in the AI plan-recognition literature, and it complicates rather than confirms the standing "Wikipedia editorial gloss" reading. Verdict: real mechanism, not a false friend — but the mechanism runs through external literature the vault hadn't connected yet, not through the two notes' own cited sources.

Claim: The two flagged notes are not an embedding coincidence — the vault's own citation graph already sits them one hop apart

Claim type: historical/documentary claim about the vault's own text — verifiable by direct inspection, Tier 1 (primary text, own venue).

claim-no-primary-source-documents-ach-abduction-similarity opens by naming the claim under test as "abductive reasoning (inference to the best explanation, after Charles Sanders Peirce)" — a direct wikilink to the exact note this topic flags as the "ml-ai" endpoint. That same note is also the documented ancestor of claim-abductive-framing-scopes-to-ach-cd-not-heuer-ach (the "epistemology" endpoint): the 2026-07-31 capture that produced the ACH-CD scoping note opens "This capture continues the thread opened by question-verify-elsaesser-stech-2007-ach-cd-abduction and the negative finding in claim-no-primary-source-documents-ach-abduction-similarity." So the pairing the retrieval index surfaced at cosine 0.84 is not a coincidence of vector space — it reproduces a real, one-hop citation path that already exists in the vault's own notes, just not as a direct edge between the two endpoints themselves.

Source: claim-no-primary-source-documents-ach-abduction-similarity and claim-abductive-framing-scopes-to-ach-cd-not-heuer-ach (vault notes, direct inspection). Tier 1 for what the notes themselves say.

Claim: The specific technique behind ACH-CD ("state-based hierarchical plan recognition") is a named member of a well-documented AI research lineage that explicitly defines abduction as "reasoning to the best explanation" — the same formulation claim-inference-logical-types documents from Peirce

Claim type: specific technical-mechanism claim. Sourcing floor: Tier 1-2 required. Met.

Robert P. Goldman, Christopher W. Geib, and Christopher A. Miller's "A New Model of Plan Recognition" (Honeywell Technology Center; Proceedings of the 15th Conference on Uncertainty in Artificial Intelligence, UAI 1999, pp. 245–254) states the field's own history plainly: Eugene Charniak is credited as the researcher who first framed plan recognition as an instance of Peircean-style abduction, and the tradition running Kautz & Allen (1986) → Charniak & McDermott (1985) → Charniak & Goldman (1993) → this 1999 paper is explicitly abductive and probabilistic by name. The paper's own abstract states: "We present a new abductive, probabilistic theory of plan recognition." Later in the text (quoted fragments below, each independently verified against the extracted PDF — the paper's two-column layout interleaves lines from both columns in plain-text extraction, so the full sentence does not survive as one unbroken run; the fragments are given in their actual left-to-right reading order):

• "To the best of our knowledge, Charniak was the first" [to argue that plan recognition is a specific form of abduction, per Charniak & McDermott 1985 — paraphrase of the OCR-broken remainder of the sentence, not independently quote-checked]
• "Charniak and Goldman (C&G) (1993)"
• "argued that, since plan recognition involves abduction,"
• "it could best be done as Bayesian inference"

This is exactly the technique family Elsaesser & Stech's ACH-CD is described as using: Wikipedia's ACH article (already vaulted, see claim-ach-cd-abduction-characterization-traces-to-wikipedia-editorial-gloss) reports "state-based hierarchical plan recognition" as the method, and "plan recognition" in this literal, disambiguated sense — hierarchical plan libraries, state-based execution models, abductive/Bayesian inference over them — is precisely the named research tradition this paper documents and situates itself within. The AI subfield's own definition of "abduction" ("reasoning to the best explanation") is the identical formulation claim-inference-logical-types records for Peirce, sourced there to Wikipedia's "Inference" article: abduction "seeks... a 'best explanation' based on likelihood given the evidence."

Source: Robert P. Goldman, Christopher W. Geib, Christopher A. Miller, "A New Model of Plan Recognition," Proceedings of the 15th Conference on Uncertainty in Artificial Intelligence (UAI 1999), pp. 245–254. Retrieved via arXiv: https://arxiv.org/abs/1301.6700 (PDF: https://arxiv.org/pdf/1301.6700). Tier 1 — conference paper, author-uploaded preprint archive, own venue. source_sha: abc7b42d06c0d09eaf8227a348efe01e75b1a3933d1c4c67fe1f3f42fad78010 (fetched via extract_pdf, tls: verified) Quotes: "We present a new abductive, probabilistic theory of plan recognition."; "To the best of our knowledge, Charniak was the first"; "Charniak and Goldman (C&G) (1993)"; "argued that, since plan recognition involves abduction,"; "it could best be done as Bayesian inference" — each individually verified against the extracted text via the vault's quote-grounding check.

Claim: The bridge is real at the level of shared concept-history, but no source found (across three sessions now) confirms a direct citation between Elsaesser & Stech's chapter and this named plan-recognition-as-abduction literature

Claim type: technical-mechanism claim (does E&S's chapter actually cite this lineage). Sourcing floor: Tier 1-2 required. Not met — flagged.

Neither this session nor the two prior ones (2026-07-12, 2026-07-31) obtained the primary text of Elsaesser, C. & Stech, F.J., "Detecting Deception," in Kott & McEneaney (eds.), Adversarial Reasoning (2006/2007) — it remains unreachable through open-web tooling (see the multi-route dead-end already logged against question-verify-elsaesser-stech-2007-ach-cd-abduction). This session did not re-attempt fetching it, deferring to that standing question. So while Claim 2 establishes that "state-based hierarchical plan recognition" as a technique belongs to a real, well-documented AI abduction tradition, nothing confirms that Elsaesser & Stech themselves cite Kautz & Allen, Charniak & McDermott, or Charniak & Goldman — the terminological match could reflect direct lineage (E&S built on that literature) or convergent naming (E&S and Wikipedia's editor both independently recognized the same technique-family resemblance). [unverified-mechanism — needs primary]: the specific document-to-document citation link is not established, only the field-level terminological and technical match.

Source: absence of a source, corroborated across three sessions' search attempts (2026-07-12, 2026-07-31, 2026-09-28) — see question-verify-elsaesser-stech-2007-ach-cd-abduction. Tier N/A (negative finding).

Claim: This nuances, rather than undercuts, the vault's existing "Wikipedia editorial gloss" reading — the unsigned classification matches real AI taxonomy, independent of whether Wikipedia is right about what Elsaesser & Stech actually wrote

Claim type: technical-mechanism / definitional claim, synthesizing Claim 2 against the standing vault note. Tier 1 for the underlying technical fact (Claim 2's source); the synthesis itself is Seek's own connective reasoning, flagged accordingly.

claim-ach-cd-abduction-characterization-traces-to-wikipedia-editorial-gloss argues the "(see abductive reasoning)" annotation reads as "a classification smuggled in through punctuation" — an editor's own unsigned judgment call, not a claim traceable to Elsaesser & Stech's own words. That finding about attribution (who said it) still stands; this session found nothing to confirm E&S used the word themselves. But on the separate question of accuracy (was the classification a reasonable one to make), Claim 2 above supplies new evidence: "state-based hierarchical plan recognition" is not an arbitrary label to call abductive — it is standard terminology inside a specific, named 1980s–1990s AI subfield (Kautz & Allen's plan-graph formalization, Charniak & Goldman's Bayesian plan recognition) that has called itself abductive, in print, in the Peircean "best explanation" sense, since at least 1985. Whoever wrote the Wikipedia parenthetical was pointing at a real taxonomic fact about how this class of technique is classified in its own field — even if it cannot yet be shown that Elsaesser & Stech used the word "abductive" in the chapter itself.

Source: synthesis of Claim 2 (Goldman, Geib & Miller 1999, Tier 1) against claim-ach-cd-abduction-characterization-traces-to-wikipedia-editorial-gloss (Tier 4, already vaulted). [unverified-synthesis] — Seek's own connective read across two source tiers, not a claim either source states directly.

Further leads

• Charniak, E. & McDermott, D. (1985), Introduction to Artificial Intelligence, Addison-Wesley — the textbook Goldman, Geib & Miller (1999) credit as the origin of "plan recognition [as] a specific form of... abduction." Textbook form, not independently checked this session for a Peirce citation.
• Charniak, E. & Goldman, R.P. (1993), "A Bayesian model of plan recognition," Artificial Intelligence 64(1):53–79 — the paper explicitly building the Bayesian/abductive plan-recognition formalism; a ResearchGate PDF exists but was not fetched this session. Would be the strongest possible direct technical-lineage source if obtained and read against Elsaesser & Stech's citations.
• Kautz, H. & Allen, J.F. (1986), "Generalized Plan Recognition," Proceedings of AAAI-86, 32–38 — "framed the discussion of plan recognition ever since," per Goldman, Geib & Miller's own account; unread this session, would show whether the earliest AI plan-recognition-as-abduction framing itself cites Peirce.
• The unread separate "Chapter 14: Plan Recognition" in the same Adversarial Reasoning anthology (Part 1, "Opponent Intent Inferencing," per claim-elsaesser-stech-detecting-deception-chapter-bibliographically-corroborated's Google Books table-of-contents check) — could be the volume's own theoretical bridge between plan recognition and abduction, and might cite the Charniak/Kautz & Allen lineage directly.
• observation-ach-abduction-lucky-widrow-notes-share-argument-from-silence-method-classified-as-abductive — a separate, already-vaulted observation that the method (argument-from-silence) shared between two other paired notes was itself classified as abductive; worth a later comparison of whether "abductive" is doing double duty in this vault as both an object-level claim and a meta-level description of how negative findings get reasoned about.

Safety flags

None. The one page fetched this session (arXiv PDF, https://arxiv.org/pdf/1301.6700, tls: verified) was an ordinary academic conference paper with no addressed-to-AI language, override language, claimed authority, tier self-assignment, file-system instructions, credential requests, or urgency framing.

Entity candidates

• Eugene Charniak — person — per Goldman, Geib & Miller's own citation, the first to characterize plan recognition as a specific form of abduction (Charniak & McDermott 1985). This is the FOUNDATIONAL figure the whole plan-recognition-as-abduction lineage in AI measures itself against — distinct from Peirce, and currently absent from the vault despite being the direct technical ancestor of the ACH-CD claim.
• Henry Kautz / James Allen — persons — "Generalized Plan Recognition" (1986), described by later authors as having "framed the discussion of plan recognition ever since." The formal ancestor of the plan-graph approach that "state-based hierarchical" plan recognition (Elsaesser & Stech's method) sits within.
• Robert P. Goldman (with Christopher W. Geib, Christopher A. Miller) — persons — authors of this capture's Tier-1 source, "A New Model of Plan Recognition" (UAI 1999). Note: this Goldman is a co-author on a different, later paper than the "Charniak & Goldman (1993)" paper cited within it — same surname, worth not conflating the two in a later session.
• Charles Sanders Peirce — already an entity in the vault (entity-charles-sanders-peirce) — the ultimate common ancestor both branches (general logic typology and AI plan-recognition-as-abduction) trace back to. Flagging again here since this capture is precisely the "vault knows him in two different worlds" case the topic named, and the AI plan-recognition literature reaches back to him independently of the general-inference-typology literature that claim-inference-logical-types rests on.
• Plan recognition (as abduction) — concept/term — the specific AI subfield mediating between the two flagged notes; not currently an entity or MOC in the vault, and central enough to this and any follow-on investigation (Charniak & Goldman 1993; Kautz & Allen 1986) to be worth its own page.
• ACH-CD — already an entity in the vault (entity-ach-cd) — flagged again since Claim 2 is new technical grounding for what the "abductive" part of its name would actually mean if confirmed.