TEagle 4.1.0: Results & Deviations

TEagle 4.1.0: Results & Deviations Results for seven registered panels and 18 comparative task frames associated with the benchmark registration. The complete exploratory biological series adds ten records in seven dependence blocks: four intended-product witnesses from five domain-guided attempts, one Hermes no-pair outcome, and five separately defined generic-route witnesses. All 41 primer-pair rows and 11 native product rows are retained. TRAS1 verification is domain-bounded. The package contains PDF/HTML reports, CSV/XLSX tables with JSON schemas, processed scientific data and readable reproduction scripts. Original failures, unrun conditions and later corrections remain separate. Original S and original complete-workflow acceptance remain unestablished. The finite dependent frame, missing curation inputs and incomplete global costs bound interpretation; product witnesses are computational evidence. Numbered deviations Deviation 1: Structural orientation-pair checker variable shadowing The independent checker changed a variable subsequently used in grouping keys and failed after outputs had been observed. The affected panel contained 1,408 cases in 704 orientation pairs. Corrected-equation and independent retained-output assessments subsequently supported limited retrospective numerical validation without native reruns. Original prospective acceptance remained unestablished. The original validation failure and unaccepted S condition were retained. Deviation 2: TEsorter dependency-path failure and deferred corrected execution The original launch environment omitted executable search paths required by the pinned dependencies. Of 1,722 invocation slots, 871 ended in process errors and 851 were deferred without execution. A separate corrected condition supplied the dependency path while retaining scientific inputs, intended options, software and reference resources. Corrected source-label batch and isolated results were reported separately from the original errors and skips. The later ten-record biological-classification results likewise retained their finite frame and dependence. Original launch errors do not measure classification accuracy. Deviation 3: PASTEC configuration spelling and empty runtime-file validation Case-sensitive configuration names were altered during writing, and the first attempt failed before classification. Original validation also rejected declared empty package files. One attempted failure and 1,721 deferred invocation slots were retained separately. The corrected condition restored intended option spelling without changing parameter values and allowed an empty runtime file only when its declared path, size and content identity agreed. Supplemental source-label scoring did not promote the original validation to success. The exploratory LF biological-classification condition is disclosed separately below and does not establish original PASTEC BIO_CLASS execution. Deviation 4: RepeatClassifier orientation-aware supplemental label recovery The original collection adapter rejected exact reverse-complement outputs. The affected subset contained 134 invocations and 1,610 observations for 1,477 unique queries: 1,333 forward matches, 276 reverse-complement matches and one palindrome compatible with both orientations. Post-output recovery required whole-sequence identity while preserving labels, multiplicity and native orientation. Repeated identifiers were concordance checks; conflicting labels could not receive consensus. This supplemental subset did not replace original invalid-output states, redefine the full benchmark denominator or establish independent biological validation. Deviation 5: Revised supplementary GUI measurement method After the original observations, two supplementary families measured narrower quantities. Family A tested freshness protection after specimen edits using unchanged and sham controls and observed backend design dispatch. Family B tested transfer of custom oligos and settings to the actual local-PCR request and completed record. The families shared 24 base cases. A had 24/24 accepted cells; B had 23/24 and one terminal-uncertain cell. A later timing condition distinguished positive-control submission intervals from recording-critical intervals while retaining the 2,250 ms critical or unclassified gap bound and five-second no-dispatch interval. The uncertain original cell had an unexplained 2,609 ms recording gap and was neither a functional failure nor a pass. A separate technical replacement did not change that original outcome. The original five observational routes retained null detection, and 43 methodological holds remained unrun. Families, controls and replacements were not pooled. Deviation 6: MCHelper corrected-module and recovery-counter condition The supplemental condition replaced module A with U and supplied an external classifier reference. A source correction initialized the recovery counter at completed checkpoint N rather than N minus one, retaining the intended 16-round limit. Continuation from the completed 16-round checkpoint added no BEE rounds. This modified exploratory condition was distinct from original-A results and unmodified software. External-reference component rights and training exposure remained unresolved, and reference agreement was not independent biological truth. The nucleotide-to-amino-acid BUSCO mismatch precluded a gene-filter validity claim. Deviation 7: Display-only 4.1.1 correction with targeted verification The five domain-guided cases displayed included-region start/length values as endpoints, while computational constraint transfer remained correct. The decision was to ship a display-only correction as 4.1.1 and retain the unchanged scientific benchmark results. One separate ART-CH clone-5 source-interface probe displayed the correct interval, 1,379–1,772, and reproduced the same analysis, scientific settings, five primer rows and 295-bp molecule. Targeted verification replaced a full benchmark repeat because scientific code and parameters were unchanged. Original 4.1.0 display failures remain recorded. The 4.1.1 installer also completed version/license navigation and cancellation before installation; no fresh installation was performed. Later unnumbered technical corrections and exploratory conditions are reported alongside their task results and in the report. They do not retroactively preregister observations or replace failed originals. Incomplete global resource costs remain unknown.