---
title: "Read mechanism against mechanism, ACH and expert-system backward chaining are a false friend: a flat matrix scan versus a goal-directed recursive proof search"
type: "observation"
status: "seedling"
bridge_verdict: "false-friend"
audit_status: "synthesis — [unverified-synthesis]: the false-friend verdict is Seek's own comparative reading of the two Tier-1 mechanism notes below; no source asserts the comparison, consistent with [[claim-no-documented-comparison-of-ach-to-backward-chaining]]'s finding that none exists. The two mechanism descriptions are separately sourced (both Tier 1, capture-verified) in their own notes. // 2026-10-18 (scheduled cross-model audit, auditor claude-opus-5, writer claude-opus-4-8): CONFIRMED, no defect in the note. Both underlying primaries were independently re-fetched and read in full this pass (van Melle ch. 4, sha 69fc0bfae367…; Heuer 2007, sha ce60834078bc…), and each mechanism description this note leans on is verbatim-supported in its own note's source — so the synthesis rests on two confirmed legs even though the comparison itself remains unsourced, exactly as flagged. source_quote located in the cited vault capture. The note's account of its own history checks out against that capture and Cali's 2026-10-08 review callout inside it: the Jevons 'quote' was the vault's own Tier-4-origin prose, and 'simultaneous / all at once' is not in Heuer — both correctly demoted here rather than inherited. TWO THINGS RECORDED, NOT CORRECTED. (1) source_tier 1 is for a vault-internal capture of Seek's own reasoning (source_url is a `vault:` path), which is Tier 1 only in the sense of a first-hand record of its own author's thinking — it is not external support for the claim. Honest as written, because audit_status and `flags` both say so in the same breath, but a reader scanning frontmatter alone could misread it. (2) Per the 2026-10-18 audit of [[claim-mycin-backward-chaining-is-goal-directed-recursive-subgoal-search]], 'depth-first' is this vault's characterisation and not van Melle's word; that note now carries the three qualifications the MYCIN chapter adds (preview mechanism, unity path, meta-rules). None of them weakens the contrast drawn here — all three make backward chaining *more* focused, not less — so this note's 'goal-directed recursive proof search' versus 'flat exhaustive consistency scan' fork stands."
audits: ["2026-10-18 claude-opus-5"]
source_url: "vault:10-inbox/raw/2026-10-07-bipartite-epistemology-computing-no-source-was-found-documenting-an-explicit.md"
source_title: "Bipartite hop: ACH's matrix elimination and expert-system backward chaining are a mechanism-level false friend"
source_author: "Seek (writer_model claude-sonnet-5)"
source_date: "2026-10-07T00:00:00.000Z"
source_tier: 1
source_quote: "Sharing the word \"eliminate\" is not sharing the mechanism."
provenance: "Promotion from 10-inbox/raw/2026-10-07-bipartite-epistemology-computing-no-source-was-found-documenting-an-explicit.md, 2026-10-17"
origin: "batch"
writer_model: "claude-opus-4-8"
derived_from: ["10-inbox/raw/2026-10-07-bipartite-epistemology-computing-no-source-was-found-documenting-an-explicit.md"]
date_created: "2026-10-17T00:00:00.000Z"
tags: ["embedding-false-friend","cross-domain-bridge","ACH","backward-chaining","expert-systems","reasoning-by-elimination","mechanism-comparison","bridge-investigation"]
flags: ["[unverified-synthesis — Seek's comparative reading of two primaries; no source draws the comparison]"]
seek_code_commit: "d40ee39"
---


The vault has long held that [[claim-ach-step-5-instructs-analysts-to-disprove-not-prove|ACH's Step 5]]
*rhymes* with [[claim-inference-classic-ai-engines|backward chaining]] — both get
called "elimination," both were developed in the 1970s — and that
[[claim-no-documented-comparison-of-ach-to-backward-chaining|no source has ever drawn the comparison]].
That earlier note is a documented-absence finding: nobody *wrote it down*. This
note does the thing the absence left undone — it compares the two mechanisms
directly, from their Tier-1 primaries — and finds the resemblance is a
mechanism-level false friend.

Backward chaining, per MYCIN's own technical account, is a **sequential,
goal-directed, recursive proof search**: pick one target, retrieve the rules
whose conclusions bear on it, and chain through premise subgoals until each
resolves ([[claim-mycin-backward-chaining-is-goal-directed-recursive-subgoal-search]]).
ACH is a **flat, exhaustive consistency scan**: enumerate every hypothesis up
front, score each evidence item against all of them in a matrix, and eliminate
by tallying inconsistency — no rule base, no conclusion-indexed retrieval, no
recursive subgoaling ([[claim-ach-matrix-scores-each-evidence-item-against-every-hypothesis]]).
Sharing the word "eliminate" is not sharing the mechanism: one focuses on a
single proof tree, the other refuses to focus at all. This joins the vault's
tracked false-friend series ([[observation-ach-abduction-lucky-widrow-cosine-pairing-is-embedding-false-friend]]),
but is reached by reading the algorithms rather than by failing to find a
citation.

The verification that would move this past an echo — a primary or scholarly
source that actually draws the ACH↔backward-chaining comparison — is tracked at
[[question-ach-backward-chaining-structural-echo]]. The through-line note
[[observation-reasoning-by-elimination-from-stoic-logic-to-symbolic-ai]] still
lines ACH and backward chaining up as one lineage of "eliminate the impossible";
the mechanism read here suggests that lineage is coarser than the algorithms
underneath it, and the parent note's own Step-5↔chaining span remains the
unsourced one.

> [!note] Seek's commentary:
> The capture wanted a tidier ending than the evidence gives. It proposed that
> ACH's *true* structural twin is Jevons's Logic Piano — a combinatorial machine
> that strikes out the inconsistent combinations all at once — and quoted a line
> about the piano to seal it. Two problems, both Cali's catch. The quoted line is
> the vault's *own* prose from [[claim-jevons-logic-piano-1869-mechanical-inference-machine]],
> Tier-4 Wikipedia origin, not a primary — it can't carry a headline. And the
> hinge of the Jevons analogy is "simultaneous / all at once," which is exactly
> the word that isn't in Heuer: his matrix is worked row by row. Pull that word
> and the Jevons twin loosens from a finding back into a lead. So I'm keeping the
> part that survives contact with the primaries — ACH and backward chaining are a
> false friend — and leaving Jevons as a question for a session that can read the
> Logic Piano's own mechanism, not the vault's gloss of it. The fork the capture
> drew (a combinatorial/matrix family versus a goal-directed/recursive one) is a
> good hunch; it just isn't sourced yet, and a bridge is only as strong as its
> softest span.
> — Seek