TEagle 3.8.0 benchmark evidence: sanitized aggregate results and provenance

This aggregate-level reproducibility and evidence package contains sanitized benchmark results for TEagle 3.8.0, a native Windows single-locus workflow for transposable-element analysis and PCR assay planning. The development corpus has 135 rows from 122 accessions: 129 produced output, 6 failed, and 34 containing-record cases were stratified. The primary analysis contains 94 distinct inputs, 93 gradable at class/order rank. One additional distinct input was outside the class/order endpoint. 66 were answered and all 66 matched the curated class/order label. TE-only withholding was 16/82, and 11/11 controls were correctly not called. In the held-out set, 34/35 TE cases were answered and correct, 1 was withheld, and 8/8 controls were correctly not called. Comparator accounting covered 116 successful accessions: 82 paired, 33 excluded despite comparator output, and 1 without comparator output. Default-database paired counts were both correct 64, TEagle only 2, TEsorter only 9, neither 7 (exact McNemar P = 0.0654297); matched-database counts were 61, 5, 9, and 7 (P = 0.4239502). The assay reconstruction found products for 17/17 cases, including 14 exact size matches and 16/16 concordant single-locus cases; 15 were preworked, 1 blind, and 1 multilocus. The provenance extract covers all 172 completed outputs, not the 6 failures. Exact raw reruns require retrieval of the cited accessions, TEagle 3.8.0, and the documented comparator environment. Underlying source records, raw sequences, primer strings, genomic coordinates, per-case sequence-bearing outputs, and comparator intermediates are not redistributed.