The Yield Belt Pays Out

Status: Open

Status is derived only from the shepherd-authored triage/prediction data above -- community submissions and claims are a separate overlay and can never change it (see the participation panel below).

This is a proposed connection between two domains, generated by a language model. It is not an article and not evidence: it sits below the evidence/publication boundary. A quantitative prediction and a named kill-dataset are attached (when registered) so the claim stays falsifiable rather than merely evocative.

A conjecture about this corpus itself

This is one of five meta-conjectures the generating model posed, blind, in the corpus’s final wave — predictions not about the pre-print world but about this collection of 1001 conjectures: how its verdicts would distribute, where its novelty would concentrate, whether its own governing architecture would survive its own data. Its kill-dataset is the corpus’s adjudication records, which is why it sits outside the ordinary triage ladder (“out of scope” below): it could only be judged once everything else had been.

Those records now exist, and on 11 July 2026 the model formally resolved this conjecture against them. Verdict: killed

Pure-Q2 waves (W09,W10,W11,W12,W22,W25) show 7/213 novel_unlocated = 3.3%; Q1 (W07+W08 dense-reference block per the architecture's own Section 4 assignment, plus W13) shows 1/135 = 0.74%. Ratio 4.4x meets the 2x bar but chi-square(1) = 2.38, p ~ 0.12 (exact binomial ~ 0.12) — not significant at 0.05; under the alternative operationalization (W13-only Q1: 2.9%) the ratio is 1.15x and fails the 2x bar outright. Secondary ordering fails: measured Q2 3.3% > Q1 0.74% > Q3 (W14) 0% = Q4 (W15+W18) 0%, not the pre-specified monotone order.

The test is runnable — quadrant assignments exist in the pre-registration and novel_unlocated verdicts exist in the records — and the primary clause (>=2x AND chi-square p<0.05) fails under every defensible operationalization: the favorable one passes the ratio but not significance, the unfavorable one passes neither. Operationalization choices flagged: pure-quadrant waves only (straddles W16,W17,W19,W20,W21 and cross-cuts W23,W24 excluded); W07/W08 read as Q1 per the architecture's explicit dense-stratum-reference language; novel-and-testable read as shepherd novel_unlocated. The rubric-shift caveat bears heavily: the strict leaked-focused rubric compressed novel rates roughly six-fold below the pre-registered 18-24% Q2 predictions, gutting the event counts (8 novels across 348 items) and hence the test's power — this kill indicts the compound clause as measured under that rubric, not necessarily the two-axis theory, whose signature survives in the leaked gradient. But the conjecture staked its verdict on the novel-rate chi-square, and on the recorded data that clause fails.

All five self-judgments, with the full exhibit →

Claim (verbatim)

This corpus of 1001 conjectures was generated under an explicit two-variable theory: that a language model's genuinely novel and killable claims concentrate in the "yield belt" where sources are digitized but questions have not yet been operationalized by scholars, while heavily worked fields tempt the model into restating the known. That organizing theory is itself a falsifiable prediction about this very dataset, and it should be held to the same standard as the conjectures it organized. I conjecture that when every item has been triaged and adjudicated, the yield-belt quadrant's rate of "novel and killable" verdicts will exceed the rate in the digitized-and-operationalized quadrant by at least a factor of two, with the four quadrants ordered exactly as the theory predicts. The mechanism is that operationalization saturates a field's question space, so a model trained on that field can mostly only reinvent, whereas digitized-but-unoperationalized material supports claims that are both fresh and checkable. If the gradient is flat, the corpus's own quadrant map — its central intellectual claim — is broken.

Prediction clause (verbatim)

When all 1001 conjectures are classified by quadrant and triaged, the proportion receiving a "novel and testable" triage verdict in the digitized-unoperationalized quadrant will be at least 2 times the proportion in the digitized-operationalized quadrant. Primary clause (the verdict follows it): the >= 2x ratio, significant by a chi-square test at p < 0.05. Secondary clause: the four quadrants' novelty rates are ordered digitized-unoperationalized > undigitized-unoperationalized > digitized-operationalized > undigitized-operationalized or as otherwise pre-specified by the quadrant model, monotonically.

Kill-dataset (verbatim)

The One Thousand and One Conjectures corpus and its triage/adjudication records; kill is a statistical test (chi-square on triage verdict by quadrant).

Nobody has run this test. The kill-data is named above. If you can run it — or you know the paper that already settles it — claim the kill or submit the prior. Kills and priors are credited here, by name, as they come in.

In the atlas

This conjecture is bridged, as an L1 lead, onto these Inferpedia subject pages.

Provenance

Run: Fresh agent generation · model: claude-fable-5

Composed blind by claude-fable-5 with zero tool use, emitted as a single JSON text message per the fresh-lane blindness protocol.

Novelty / leakage triage

out of scope (a meta-conjecture about the corpus itself)

Meta-conjecture whose subject and kill-dataset are the 1001-corpus's own triage/adjudication records, not an external scholarly claim; per W26 rules this is out of shepherd scope.

Predictions

No prediction registered yet.

Weigh in

No community feedback yet.

Add your take

Posted immediately (spam is removed). Community feedback is never an adjudicated verdict and never changes this conjecture's triage label or status above.