⁠The Elephantine Filter: Empirical Intercalation⁠

Revision 3 withdraws a mathematical claim made in Revision 2 and restructures the paper’s findings. 1. The Revision 2 bridging check, which grounded the exclusion of AP-6, contained an arithmetic error. It compared the 78 elapsed lunations between AP-5 and AP-6 against Nisan-to-Nisan month totals (maximum 75) and declared the gap impossible under every admissible pattern. AP-5 is Elul, month 6; AP-6 is Kislev, month 9. The true gap equals the Nisan-to-Nisan total plus three. The corrected requirement is 75 months, which equals the admissibility maximum. Twelve of the 57 patterns can bridge the gap in principle. The impossibility claim is withdrawn. 2. The primary result is now the full fifteen-document corpus test: 0 of 57 admissible patterns fit all usable datelines. The four-survivor result (patterns 38, 39, 40, 41) is conditional on the removal of AP-6 and is labeled as such. 3. A leave-one-out uniqueness test is added. Of fifteen single-document removals, only AP-6’s yields any live patterns, and the live set is exactly the four-pattern family. AP-6’s anomalous status is an output of the analysis, not an input assumption. 4. The headline four-pattern family is unaffected, verified by rerun. The error was found during an independent verification pass on the Revision 2 code. The analysis engine (elephantine_analysis_v3.py) reproduces every figure above under assertion. PILOT STATUS. This is a working paper and methods pilot, not a finished historical-calendar study. Two source-verification steps remain outstanding: a line-by-line check of all fifteen datelines against the Porten-Yardeni TAD edition, and a check of the Parker-Dubberstein intercalation record for Xerxes years 15 through 21 against the AP-5 to AP-6 bridge conflict. SHA-256 checksums: elephantine_analysis_v3.py: 42a1b0f665673980d32fde2adbba6d4c083df13901044c869ca3a7410cee48df — elephantine_pattern_results_v3.csv: 3793f0abb9a8f04f42b43fbae9cef91c8afb3420804a82b2ddbc4cac7a1a4476