FCLT Evidence Table v40 — Section 1 Attribution Audit: The Screened Set Falls from Eighteen Rows to Six

This version records a line-by-line attribution audit of Section 1, in which every screened proximity row was searched for in the paper it cites. Twelve of eighteen do not survive the search, and the screened count falls to six. Eleven rows cannot be traced to a measurement in the paper they name. Six cite the wrong paper entirely; two of those have since been relocated to their true sources — the solar-cycle row to P92, where it reproduces, and the quasinormal-mode row to P73, where the recorded deviation of 0.041 appears nowhere among that paper’s four values. Four cite the correct paper but report no such measurement; P85 states explicitly that no analysis of the dataset in question has been performed. One is an identifier collision. A twelfth row is withdrawn as circular: P93’s own text gives a characteristic equation whose root is the constant the row reports proximity to. The root cause is the one Section 1c of v36 already named — a prediction registry numbered P1 through P14 colliding with paper numbers P01 through P107. It is recorded as a new failure class, R12 registry collision, and it is not a first occurrence: P73 Section VI(b) records three ghost rows removed for the same reason on 16 June 2026, before v36 was built. A parallel sweep of the deviation metric found a single affected cell rather than the systematic error anticipated. P05 Section IV.A and Table 1 print the correct fractional deviation of 0.0011 for the galaxy axis ratio; only the z near zero row of its Table 2 prints the absolute difference 0.0007, and v36 transcribed that one. The exact fractional value is 0.001188. This moves the flat-lattice probability of the corpus’s strongest row from one in 344 to one in 203, and the redshift-gradient figure quoted in P05, P12 and P13 from thirty-one-fold to approximately nineteen-fold. No previously falsified row returns. Three further items are recorded. The cosmological constant row, cited as foundational in six papers, appears in no prior version of this table and is entered as not a test following P06 v3. The derivation audit declared closed at v37 is reopened for one item: D4 appears nowhere in its disposition table, and P73 Section VI(d) records it as the P15 v3 Born Rule. And four new failure classes — R11 convergent restatement, R12 registry collision, R13 free-slot fit, R14 exponent-rule non-uniqueness — join R10 as applied ahead of the protocol that will contain them. The consequence is stated plainly. The nineteen-row rival-constant scoring prepared in August is not deposited, because most of those rows were never eligible to enter it. The Rival-Constant Catalogue v1 remains frozen and valid; what fails is the row set it was applied to. The narrower claim that survives is this: of the Section 1 rows traceable to a measurement in a named section of a named paper, none is admissible for adjudication between rival constants at the uncertainties those papers report. That is six rows, not nineteen. Rows still awaiting source verification are marked attribution-unresolved rather than deleted, so that the unfinished part of the audit is visible in the table itself. Zero of the six remaining rows attain significance under Benjamini-Hochberg, as was true of eighteen. Because three corpus totals circulate in v39, the result is reported across all of them; nothing turns on the choice. Supporting files reproduce every number. Code MIT, content CC BY 4.0.