The noetome, measured

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.

Leaked (re-derived-published) rates track kill-database scale, not production: cuneiform W22 13.3% (CDLI: >500,000 known artifacts, >360,000 electronically catalogued), exact sciences W20 16.7%, law W21 13.3%, medieval literature W07 14% (Trismegistos-class corpora ~964,000 text entries; Cantus ~500,000 chant records) — versus 0% leaked AND 0% novel in South Asia (W18 Americas, W23 women, W24 oral-written likewise 0% leaked), although South Asia alone holds an estimated ~10 million extant manuscripts (NAMAMI database ~3.4M entries) against Buringh & van Zanden's ~8 million manuscript copies for the entire medieval West's production. All 10 actually-resolved prediction records (7 conjectures) closed against digitized databases (DBBE, Pinakes, papyri.info/CDLI-derived samples, colophon lists); zero closed against production-estimate evidence.

The pre-registered Steiger test is not computable as posed — the triage records contain no killability field (I proxy it with leaked rate and with which predictions actually resolved: both flagged choices), per-stratum conjecture counts were fixed by the registered allocation rather than measured (a circularity), and no published per-region pre-print production series exists at stratum granularity — so per the adjudication instruction the primary clause is judged QUALITATIVELY, and I say so. Judged on substantive content, the direction is unambiguous: every checkable signal in the corpus (leaked rate, novel pockets, actual resolutions, in-house kill quotas) sits where 20th-21st-century digitization is thick, and the strata that dominate honest production estimates (South Asia foremost) yield zero checkable signal. Caveats: this was also the architecture's own Section 3.5 prior, so allocation was partly self-fulfilling — though the verdict rates were measured blind; and W16 Africa is a counterexample in one direction (15% novel atop small databases), driven by thin-literature unlocatability rather than database mass. The rho>=0.6 / rho<=0.3 / Steiger-p secondary machinery remains formally unrun; a numeric replication would require post-hoc killability scoring and an assembled production-estimate series.

• CDLI, About (>500,000 artifacts, >360,000 catalogued)
• Trismegistos Texts (~964,000 entries, Oct 2024)
• Cantus Database (~500,000 chant records)
• DBBE (>12,000 occurrences)
• Buringh & van Zanden 2009, JEH 69(2) (~8M+ medieval Western manuscript copies)
• National Mission for Manuscripts (~10M estimated Indian manuscripts; ~3.4M database entries)

The shepherd taxonomy is leaked / adjacent / novel_unlocated / out_of_scope: an 'untestable / no adequate kill-dataset' verdict was issued 0 times in 851 triaged items, so the predicted modal failure mode of the blank strata is unmeasured everywhere. Only 'leaked' maps to 'already established' (dense strata: W13 scholasticism 2/35 = 5.7%, W11 Byzantine 1/35 = 2.9%), and if 'failures' means leaked-only the >=40% clause is vacuously 100% — the chi-square of failure-type by stratum-class cannot run on a one-column failure taxonomy.

This is the paradigm case of a prediction whose categories the triage never used: the conjecture requires a three-way failure taxonomy (falsified / already-established / untestable) but the records encode only redundancy (leaked), with adjacent explicitly defined as un-run rather than failed. Worse, the campaign's own Section 5.4 blankness protocol engineered the predicted failure mode out of existence — Q4 waves (W15, W16, W17, W18) were generated against closed menus of pre-vetted kill-datasets, so an untestability verdict structurally could not arise in exactly the strata where the conjecture predicted it would dominate. Operationalization choices flagged: blank strata read as W15/W16/W18/W24, dense as W13/W11 per the conjecture's own naming. What would resolve it: a second-pass kill-feasibility audit tagging every adjacent item with whether its named kill-dataset actually exists at the required granularity and access level; the chi-square could then run on leaked vs kill-infeasible by stratum class, and the missingness-map claim would become testable.

Mapping the architecture's quadrant assignments to an ordinal operationalization score (Q1=4, Q3=3, Q2=2, Q4=1, straddles averaged) across the 17 quadrant-assigned waves W07-W22+W25, Spearman rho between operationalization rank and leaked rate = +0.38 (t ~ 1.59, df 15, one-tailed p ~ 0.066) — below the pre-registered rho>=0.6 and not significant at 0.05. Secondary clause fails decisively: the two named densest strata rank 12th (W13 scholasticism, 5.7%) and 14th (W11 Byzantine, 2.9%) of 17 in reinvention rate; the actual top two are W20 exact sciences (16.7%) and W07 medieval literature (14%).

The test is genuinely runnable — leaked is the rubric-stable 'already established' measurement and the pre-registration supplies an operationalization ordering — and it fails on both clauses. Operationalization choices flagged: wave=stratum; quadrant-to-ordinal mapping is mine (no numeric index was pre-registered, only the quadrant placements); cross-cut waves W23/W24 excluded for lacking quadrant assignments — including them as low-operationalization points raises rho to only ~0.45, so the kill is robust to that choice. The substantive finding is sharper than a mere miss: reinvention tracks the DIGITIZATION axis, not the question-operationalization axis — the corpus's two pre-registered axes dissociate in its own failure data, with the most question-saturated strata (scholasticism, Byzantine transmission) among the LEAST reinvented and the digitized-quantitative ones (exact sciences, cuneiform at 13.3%) the most. Caveat: wave-level leaked counts are small (0-7 per wave), making individual ranks noisy — but noise cannot rescue a point estimate of 0.38 against a 0.6 bar plus a secondary clause whose two named strata land in the bottom third. Note the direction is at least positive; the calibration-curve idea is weakened, not reversed.

Only 7 conjectures (10 prediction records; cj-079's three variants counted once, a flagged choice) have been resolved against kill-datasets — 0.7% of the corpus. Directional accuracy 4/7 = 57.1% (right: entropy-floor, precision-recession, zipf, yule; wrong: dealers-razor, universality-class, timurs-fingerprint), binomial P(X>=4 | n=7, p=0.5) = 0.50; numeric-threshold coverage is also 4/7 = 57.1%, so the predicted >=20-point direction-over-magnitude gap measures 0 points. At n=7 the stated binomial test can only reach p<0.05 at 7/7 (0.0078; even 6/7 gives 0.0625), so the primary clause was structurally undecidable on this sample.

The resolved sample is too small for the stated test in the strict sense: no achievable outcome short of perfection could have satisfied the p<0.05 binomial clause, and the 95% CI on 57.1% spans roughly 18-90%, consistent with both the conjecture and its negation. A second as-posed defect: this corpus's kill-clauses bundle direction and magnitude into single supported/killed verdicts (all 4 supported items satisfied their stated thresholds by construction; all 3 killed items failed direction itself), so the predicted dissociation — right arrow, wrong number — had almost no way to register in the records; it occurred in 0 of 7 cases. Honesty note: this verdict is NOT a flattering dodge — the point pattern runs against the conjecture (57.1% direction misses the 60% bar, 57.1% coverage exceeds the 40% ceiling, gap 0 not 20+), and had these proportions held at n>=40 the verdict would be killed. What would resolve it: 40+ resolutions with sub-clause scoring that separates directional sign from numeric-interval capture at adjudication time — a scoring change future waves should adopt. Counting all 10 records instead of 7 conjectures gives 7/10 direction (p=0.17) and 7/10 coverage; same conclusion.