{"path":"research/linked-convergent-and-what-we-are-actually-asking.md","content":"# Does the system tell parts from evidence by itself, and how accurately?\n\n**Date**: 2026-09-22 · **Type**: measurement against a pre-registered prediction · **Founder\ninstruction**: *\"can and does the system work this out for itself now, does it accurately distinguish\nbetween the types of parts? And what does argumentation philosophy/theory say about all this?\"*\n\n**⚠ The theory half was already in the corpus and this doc does not restate it.** Read\n[weakest-link-arithmetic-and-the-merge-hunch.md](weakest-link-arithmetic-and-the-merge-hunch.md)\n§ 1d and § 8 first (the linked/convergent tradition with citations, Yu & Zenker's circularity\ncritique, Freeman's structural reply, and the parts-versus-evidence reframe), and\n[a-cut-through-a-concept-is-an-argument.md](a-cut-through-a-concept-is-an-argument.md) § 5 (a dozen\nrival criteria, Goddu's case for dropping the distinction, and why reviewer disagreement is the\nexpected case). A first draft of this doc re-derived all of it from a fresh web search; the founder\ncaught it.\n\n**What is actually new here: the pass ran, and it confirms § 8's prediction.**\n\n---\n\n## 1. The prediction, made before the pass existed\n\n§ 8 of the weakest-link doc registered the three-way pass (necessary / corroborative /\nmiscast-as-support) as *padding-defense option C*, and predicted its headline:\n\n> a corroborative-looking child is usually an evidence edge mislabelled as decomposition\n> (three studies are witnesses, not parts)\n\n## 2. The measurement, 197 edges, 2026-09-22\n\n| Who wrote the edge | n | mean relevance | genuine parts | miscast as parts | fits nothing |\n|---|---|---|---|---|---|\n| `auto_connect` | 84 | 0.17 | **0** | 36 | 48 |\n| `split` | 44 | 0.42 | **30** | 4 | 10 |\n| `human` | 39 | 0.32 | 9 | 18 | 12 |\n| `extraction` | 30 | 0.49 | 6 | **20** | 4 |\n\n**The prediction holds on the provenance it was aimed at.** `extraction` writes the highest-relevance\nedges of any source and **20 of 30 are miscast** — the child bears on the parent and is not a piece\nof it. That is § 8's sentence, measured.\n\n**And it separates two defects the prediction did not distinguish.** `auto_connect`'s 84 are not\nmiscast parts but non-relations (mean relevance 0.17, zero genuine parts) (⚠ corrected 2026-10-07: these links were stored premise first, so the judge was asked whether a conclusion is a piece of its premise; the zero records that, and says little about the pairs, of which 16% held no relation when read again: `auto-connect-upgrade.md` § 2c), which the relatedness gate\nruled the same day fixes going forward and the already-ruled re-check retires. The gate does nothing\nfor `extraction`, because those pairs *are* related.\n\n## 3. The deterministic detector is Freeman's structural criterion, and its silence is the predicted grey zone\n\n§ 8 names the pre-signal — *child shares the parent's subject and fragments its predicate → part;\nindependent same-shape proposition → evidence* — as a **structural** criterion of the kind Freeman's\n2023 reply defends against Yu & Zenker's circularity critique. Run alone on all 197 it reads **27\nagree, 18 disagree, 152 no signal**.\n\n**So the 77% silence is not a defect of the detector — it is the grey zone Yu & Zenker predict**, and\nthe reason the pass keeps a confession value. The **18 disagreements** are where two independent\nreadings conflict and are the cheapest first human pass.\n\n## 4. Accuracy is not established, and cannot be from inside\n\n`MATCH ()-[r:DECOMPOSES_INTO]->() WHERE r.support_semantics IS NOT NULL` returns **0**. No human has\never set one, so there is nothing to score against — and § 5 of the cut-through-concept doc already\nsays why a score would be the wrong object: **reviewer disagreement is the expected case rather than\na defect.** The field's own instrument is agreement, not accuracy.\n\n**Consequence for the founder's word \"accurately\": the answerable question is whether a reader\nagrees with a re-filing, which only the review sitting produces.** Nothing else here will settle it.\n\n## 5. One open question the corpus may not have: is there a fourth cell?\n\nFreeman argues that what Walton's Suspension test picks out is **coordinatively compound** structure\nrather than linked structure: premises that could each support the conclusion alone yet function\ntogether through parallel reasoning. The live schema has three cells (`necessary`, `corroborative`,\n`alternative`) plus `does_not_fit`.\n\n**Whether that is a gap is genuinely unclear and should not be asserted either way**: it may already\nbe `corroborative` under another name, or `alternative`. **74 of 197 edges came back `does_not_fit`\nand nobody has read them** — that pile is where the answer would show, and reading it is the cheap\nmove before any schema question is opened. Unruled; not a proposal yet.\n\n---\n\n## Cross-references\n\n[weakest-link-arithmetic-and-the-merge-hunch.md](weakest-link-arithmetic-and-the-merge-hunch.md)\n§ 1d, § 4, § 8 (the tradition, the circularity critique as a non-objection, and this pass's\nregistration) · [a-cut-through-a-concept-is-an-argument.md](a-cut-through-a-concept-is-an-argument.md)\n§ 5 (a dozen criteria; disagreement as expected) ·\n[retrieval-instruments-beyond-cosine.md](retrieval-instruments-beyond-cosine.md) § 4 (Yu 2025's\nscheme-instance criterion, checked against the graph and not computable today) ·\n[session35-the-verification-night.md](session35-the-verification-night.md) § 4 (the run) ·\n`deliberus/decomposition_judgement.py` · `deliberus/support_semantics.py` (module header)\n"}