Methods acceptable for registration once listed administrative pre-registration conditions are completed. No new major or minor methodological errors identified in version 0.12.0.
Round 11 comments
C11-1: Resolved. Competing risks, protocol/protocol.qmd:615 now distinguishes reported counts from curve inversion. Approximation (1/d_k) applies only to reported first-event counts with valid person-time. Curve inversion requires reported or author-supplied joint covariance across causes and interval endpoints, transformed through (J\Sigma J^\mathsf{T}), with dependence retained during pooling and uncertainty propagation. Missing qualifying inputs disable that route while preserving eligible alternatives. Subdivided intervals reuse one hazard draw.
Numerical verification recovered hazards 0.4 and 0.2 from compatible cumulative incidences. For round 11’s Kaplan-Meier example, revised delta method recovered variance 0.016975, matching Greenwood propagation rather than incorrect (1/90=0.011111). Joint uncertainty is consistent with established Aalen-Johansen covariance estimation. Survival package methods.
Declining other reviewers’ count-reconstruction formulas was justified. Curve increments and published risk counts do not identify within-interval censoring histories; knowing event counts would still not establish variance of curve-inverted hazards. Conservative covariance requirement may restrict availability of absolute estimates, but supplies defensible eligibility and fallback rules.
C11-2: Resolved. Absolute effects, protocol/protocol.qmd:963 now explicitly uses each covering arm’s reported Aalen-Johansen cumulative incidence when that route supplies reference risk. Hazards are required only for hazard-derived routes. Scenario quantiles therefore no longer impose an incompatible prerequisite on reported-incidence conversion.
Neither C11 proposal was declined or only partly adopted.
New findings
None requiring replacement text or values. Associated revisions to extraction items, Kaplan-Meier eligibility, shared interval draws, exponential-route count definitions, document date, and amendment record are consistent with adopted corrections.
All requested files inspected, including programmatic inspection of all 861 registry-candidate rows. Cross-table checks found no identifier, primary-outcome, applicable-window, reference-SD, Turner-mapping, nonrandomized-harms coverage, registry-count, or search-alias discrepancies across 637 applicable outcome/population window pairs and 43 agent mappings. Version comparison confirmed substantive tables and search implementation unchanged from version 0.11.0.
Search renderer reviewed statically. R execution and live database recall were not assessed. Administrative checklist remains a pre-registration completion requirement, not grounds for withholding methodological acceptance. No files edited, created, or deleted.
VERDICT: ACCEPT