Forecasting the payoff of another view in a finite inspection queue: recorded data and analysis code
Yanpei LI
Recorded-data and analysis-code companion for Forecasting the payoff of another view in a finite inspection queue.
The package contains derived recognition scores, recorded outcomes and service quantities, object-role manifests, frozen protocols, analytical mark catalogs, and the code needed for the documented numerical replay. The main command reconstructs the quantitative main-table analyses, including six retained negative endpoints, and runs both modes of the numerical supplement. A separate command regenerates all four manuscript figures. The finite-window calculations use separate grid and coupling implementations; the stationary calculation checks 43 positive boundaries.
The evidence concerns the specified finite data banks and simulated inspection contracts. The original registered 32-stream common-error comparison remains unresolved. Later deterministic error enclosures are conditional numerical bounds, not confidence intervals for new objects. The Washington intervention does not constitute a separate replication of common forecasting failure; reserved iCub evaluation objects remain unscored. Complete-cost policy losses are retained.
Source datasets are attributed with their identifiers, study selections and applicable terms in SOURCE_DATA.md and THIRD_PARTY_SOURCES.json. Original images and videos, fitted neural checkpoints, and a complete raw-pixel training pipeline are outside this numerical package. Upstream scripts in recipes/ are reference material. COVERAGE.md distinguishes analyses actually executed by the portable commands from included but unexecuted supporting material.
Version 1.1.0 adds a six-stage offline evidence audit and three retrospective cases, and separates file identity from bounded plot-content verification. The PDF comparison normalizes only exact zero numeric tokens in supported page/Form streams; nonzero values, text and binary content remain checked. Equivalent numeric input representations are handled consistently by the audit. The package does not validate real-world arrival or stability declarations, authorize policy switching, or establish a new beneficial controller. Original scientific inputs, reference figures and negative conclusions remain unchanged from version 1.0.0.