The great responsory is the deep tail

Resolution: Killed

Caveats: Killed mechanically on the registered clause precedence, and the honest weights all point the same way. (1) THE KILL IS TOTAL, NOT A NEAR-MISS: SUPPORTED required singleton_share_R >= 2 x singleton_share_A = 0.7726; observed R = 0.3903 vs A = 0.3863 - ratio 1.0103 against a 2.00 pin, a 1.98x shortfall. Every previous multiplicative kill in the calibration series was a direction-right magnitude-miss; this one is not even that in substance: the realized separation sits at parity (R above A by 0.40 percentage points), the nearest-to-null observation the series has produced. The wagered structural effect - the great responsory as a dramatically thinner-transmitted repertory than the antiphon - is absent from this instrument, not attenuated. (2) NO DIRECTIONAL REMNANT SURVIVES: the primary clause's sign is nominally satisfied (R 0.4pp above A) at a magnitude with no evidential weight, and the companion clause - R median breadth 'lower than A by a wide margin', registered ex ante as report-only because unquantified - finds identical medians of 2.0 sources. The full distributions agree end to end: 2-source share A 15.13% vs R 14.56%; deep-tail 20+-source share A 20.19% vs R 20.44% (the conjecture named the great responsory 'the deep tail'; the actual deep tail is genre-flat too, and marginally R-heavy at the top rather than R-thin). No reading of this artifact leaves a directional claim worth carrying forward. (3) SINGLETON SHARE LOOKS LIKE CORPUS STRUCTURE, NOT GENRE (bounded claim): the two genre shares sit 0.40pp apart and both within ~3.2pp of the corpus-wide 41.79% share the house had measured on this same dump before registration (disclosed in known_priors), while the same prior compute's office-CLASS strata varied enormously (Mass 38.4%, Office-hours 44.2%, office-blank 72.9%). On this instrument, breadth varies along cataloguing/coverage axes and is flat across the one genre axis now measured - consistent with cross-source concordance breadth in Cantus being a property of which sources the member databases have indexed, not of a chant's liturgical genre. Scope discipline: this is measured for A vs R, within office scope, on one frozen dump covering ~84.9% of the live network; 107 other genre codes are untested and no claim about genre in general is licensed by this artifact. (4) NO REGISTERED-CONSISTENT READING REACHES THE BAR: a flip to SUPPORTED needs +2,193 R singletons (2,238 -> 4,431 of 5,734, holding A fixed) or -2,385 A singletons (4,818 -> 2,433, holding R fixed); the disclosed method degrees of freedom move IDs in the single digits to low hundreds. (a) Source column: srclink is the only shared column that actually keys into sources.csv (strict 1:1, 2,278/2,278, zero orphans either direction) and is therefore the registered 'source-identifier column'; the rejected alternative siglum carries just 5 duplicated values, so even that off-reading could collapse at most 5 source pairs and mint at most a handful of new singletons across both genres. (b) The mid-run tie-vs-merge precedence fix touched exactly 5 IDs and the post-fix exclusion partition was verified exact over all 53,282 IDs; any alternative handling of those 5 moves the shares in the fourth decimal place. (c) The office-fold choice (offices of merged-away verse suffixes not folded into stems; recipe-silent, stated by the compute) could only churn membership at the MASS boundary - and the artifact's own figures imply office scope excluded zero R IDs at all (R's stated pre-office-scope total, 5,734, equals its final n), so the choice had no purchase on the R side. (d) The exact-code genre reading ('A'/'R'; compound codes like InR, CmR, RSV not folded in) is what the registration's 'the value carried by a strict majority of its non-blank genre-column rows' pins; folding compounds would measure a different population, not a reading of this registration. (e) Even the maximally adversarial OFF-recipe stack - counting all 411 excluded genre-ties as R singletons - yields R share 0.4311 and ratio 1.12. Nothing available on this instrument closes to within a factor of ~1.8 of the pin; the outcome is not method-fragile. (5) THE VERSE-MERGE ASYMMETRY CUTS TOWARD A, AND R STILL ONLY TIED: by the prediction's own rule, only R stems union in verse-suffix sources (5,422 of 5,734 R IDs received at least one merged verse; every realized chain 1-deep). That construction can only raise R breadth and depress R's singleton share, so the single degree of freedom with nominal headroom toward the pin is UN-merging the verses - a reading forbidden by the prediction's verbatim text ('verses are counted with their responsory') and by the pinned merge op. The artifact does not report the pre-merge share, so the merge's realized effect is unmeasured; what is bounded is the direction (merging can only lower R's singleton share, i.e. the construction was, if anything, generous to A in this comparison - and R's share still came out above A's, by a hair), and the scale a rescue would need: even a total off-recipe un-merge would have to convert ~2,193 R stems back to singletons, i.e. roughly 40% of all merged stems owing their entire multi-source status to verse-contributed sources alone - a regime nothing in the artifact suggests. The asymmetry is the registration's own construction and grounds no grievance against the kill. (6) GUARDS, INTEGRITY, ORDERING: both sha256 hashes matched the registration exactly (verified twice, byte sizes matching the release listing); the 500-item design floor cleared by 11.5x on the thinner genre; registered_at 2026-07-19T11:23:15Z (status open, no prior resolution row, is_calibration=True under the leaked/adjacent triage rule), artifact computed_at 11:52:16Z the same day - ordering verified read-only in the DB. (7) CALIBRATION: the sixth multiplicative-magnitude pin to die (>=3:1 at 1.69, >2x at 1.28, <1/3 at 0.87, 10x at 9.4, >=5x at 3.78, now >=2x at 1.01) and the first in which the underlying effect is absent rather than undershot. The registration saw the arithmetic in outline and wagered anyway with disclosure: known_priors stated that a 2x separation atop a ~44% pooled office share needs something like A near 35% / R near 70%, and flagged the multiplicative class as the house's historically overpinned one. The lesson sharpens accordingly: earlier multiplicative kills were correct mechanisms with overconfident magnitudes; here the proposed mechanism (proper offices multiplying responsories into single-source depth) simply does not register in the measured corpus at any magnitude. No v2 chase: re-registering on another genre pair, another corpus, or a folded-code population after seeing genre-flatness would be threshold-shopping; if the breadth-is-coverage-structure observation in (3) is worth pursuing, it should be posed as its own ex-ante conjecture, not as a rescue of this one.