# RUN-050 — The novelty rule this arm cannot advance, and the route matrix it does not follow **Arm:** 數學戰士「墜衡」 / AMRAL Research Lab **Subject:** `26_Novelty_Search_Log` and `01_Phase2_Route_Matrix` — two documents about discipline rather than arithmetic **Tools:** `src52_novelty_and_routes.py`, `src11_gate_drill.py` **Logs:** `src52-novelty-and-routes.json`, `src11-gate-drill.json` **Result: `26`'s box — **`NO HIT ≠ NOVELTY PROOF`** — is the finding, and it stands. It lists four things that must happen before priority is claimed, and **this arm can do none of them**: all four are searches against external literature databases, this tree has no external access, and the standing rule forbids routing unpublished AMRAL work to a third-party provider while it lives only on GitHub. So the box is reported as *not scoreable here*, which is what it is. Across 50 reports, **21 sentences carry a novelty term: 11 are refusals and 10 are classified quotations** — a first, cruder version of the scan called the five quotations it then saw assertions, which is RUN-027's failure mode again, so every quotation is now pinned with its reason and any new unaccounted mention turns the check red. **No round of this line claims novelty.** And `01`'s route matrix, measured against what this line ran: its **PRIMARY GO** route is the most covered (**8 of 8** mapped documents), its **STOP** route is untouched — and **two rounds worked routes marked HOLD and *separate Phase later***, RUN-026 on rank 1 and RUN-023 on rank 2. That is named rather than defended: `01` orders routes by extensibility of the family theorem, a verification arm orders by what has been written down, and the two orderings are different on purpose.** --- ## `26`'s box, and why it cannot be scored here `26` searched arXiv for the curve, the equation, and the family, found nothing directly matching, and then wrote the only sentence that governs what may follow: $$\boxed{\text{NO HIT}\neq\text{NOVELTY PROOF}}$$ with four things still required before priority is claimed: 1. MathSciNet / zbMATH / Google Scholar citation chaining; 2. searching Fouquet–Wan's citing papers; 3. a number theorist checking whether it is a direct corollary of something general; 4. checking the 2026 preprints. **This arm can do none of the four.** They are literature work against external databases; this tree has no external access; and the standing rule on unpublished AMRAL research forbids sending it to a third-party provider while it exists only on GitHub. The honest report is that the box is **not scoreable here** — not that it passed, and not that it failed. `28_Submission_Gate` listed five boxes and RUN-030 delivered the one that was this arm's. This is another that is not. ## No round of this line claims novelty | | | | --- | ---: | | reports scanned | **50** | | sentences carrying a novelty term | 21 | | refusals | **11** | | classified quotations | **10** | | unaccounted | **0** | Five of the quotations are `RUN-019` quoting Phase 0's freeze rule, `RUN-030` quoting `28`'s checklist and its own "not ours" column, and three in `RUN-047` reporting what `27` defers and what labels the documents carry. **The other five are in this report.** A round that audits the novelty rule cannot avoid naming it — the subject line, the box, and the scan's own tally row all trip the term. They are pinned individually like the rest: **a round that exempted its own report would have written itself a loophole**, and the check would then be blind to exactly the document most likely to overclaim. **A first version of this scan called every one of the first five an assertion.** A crude refusal-word test cannot tell a quotation from a claim, which is exactly the pattern RUN-027 named after four rounds of it: *a scan built on the shapes the author remembers writing rather than the shapes the corpus contains*. Rather than widen the word list until the red went away, each quotation is pinned with its reason — and **any new unaccounted mention turns the check red**, which is how this round's own five were found. ## `01`'s routes, against what this line ran | route | verdict | documents this line made a subject | | --- | --- | ---: | | Higher 2-power descent | **STOP as mainline** | — | | Non-semistable + odd-`p` patchwork | GO as baseline | **4 of 4** | | Fouquet–Wan arbitrary reduction | **PRIMARY GO** | **8 of 8** | | BCS ordinary non-semistable | supporting | **1 of 1** | | Full rational 2-torsion | HOLD | — | | Analytic rank 1 | HOLD | — | | Prime conductor | niche | — | | High rank 2+ | separate Phase later | — | The mapping of documents to routes is a reading, not a field in the corpus; what is computed is the count given that reading, and the gate says so. The shape is what it should be: the primary route is where the documents are and where the rounds went, and the route marked STOP has no documents and no rounds. ## Two rounds on routes the matrix does not prioritise | round | route | verdict | what it did | | --- | --- | --- | --- | | **RUN-026** | Analytic rank 1 | **HOLD** | the rank-1 BSD identity at 37a1 and 43a1, closed with every term computed in this tree | | **RUN-023** | High rank 2+ | **separate Phase later** | the rank-2 identity at 389.a1, closing at ratio 1.0 | Neither was an attempt to extend the family theorem. Both verified an identity a document stated, which is what this arm is for — and RUN-026 is the round that found `canonical_height` returning its worst extrapolation level, which corrected a regulator this line had been carrying since RUN-017. **`01` orders routes by extensibility; a verification arm orders by what exists and is checkable.** Those are different orderings and neither is wrong. The measurement is worth making because assuming they agree would be a way of mistaking a project plan for a work log. ## The drill **200 defects, 200 caught by the check named for each**, 0 uncaught, 0 caught by the wrong check, 39 controls undisturbed, over 78 checks. The 4 planted for this gate, each turning `novelty-and-routes` red and nothing else: | planted defect | went red | | --- | --- | | 26's box is reported as a step this arm can take | `novelty-and-routes` | | the classified novelty quotations are unpinned | `novelty-and-routes` | | 01's STOP route is reported covered | `novelty-and-routes` | | the two rounds on non-prioritised routes are hidden | `novelty-and-routes` | ## What this round does not claim * **Nothing about novelty.** `26`'s box is reported, its four steps are listed as outside this arm's reach, and no verdict is offered. * **The route mapping is a reading.** Assigning corpus documents to `01`'s eight routes is judgement; only the counts are computed, and a different reading would move them. * **Coverage is not merit.** That the PRIMARY GO route is best covered says where the documents are, not that the work there is more valuable. * **The novelty scan reads sentences, not meaning.** It checks that every novelty-term sentence is a refusal or a pinned quotation; a claim phrased without any of the four terms would pass it unseen. * **`01`'s verdicts are the corpus's**, not this arm's. Whether STOP is the right call for higher 2-descent is not a question this round examines.