Latest decision review: completed training pilot and changed validation evidence
2026-10-02T23:51:06-04:00 · Dated inventory, not a live monitor
Current app activity and actual work state are separate. Idle tasks can have running Condor jobs. Recent completed and interrupted work is below. Includes tasks outside the project sidebar.
Paper 1 critical path: verified neighbour ledger → actual matched training → final recipe/validation, with targeted photometry developed in parallel. Paper 2 gain and WCS work does not block Paper 1.
Paper 1 finish plan · Recovery · Structured status
Removal-ledger repair
Owner task idle, but C4 ledger DAG 345 is still running. Latest source/recipient state counts are in decision_evidence.json; full-field acceptance remains pending.
Next: Deliver immutable completed-cell ledgers, partial-survival accounting and the full C4 validation result.
Existing reportMatched neighbour-aware ePSF training
Corrected C4 A/B training completed. Primary held-out gain inconclusive; only one bright intact validation core. Learned B does not approach RAW-faint reference. Old flux-unit-mismatched runs superseded; no magnitude-shape closure.
Next: Design an intact-core, better-powered comparison using verified regional ledgers; audit large residuals and flux/prior conventions before further A/B training. Local validated cells suffice to start design.
Existing reportTargeted faint-star photometry
star extract-band/prepare-band and mask cache fixes are installed in main through 69d311a. Corrected 1000-target, two-epoch results and conditional injection tests exist; owner report still in progress. Implementation no longer the main gap.
Next: Accept owner-reviewed final production-interface report, catalogue reference convention, failure/support accounting and profile-mismatch limits; then validate against the final paper generation.
Temporal ePSF and instrumental photometry
Latest phase-shift null benchmark falsely rejects independent slow synthetic trends. Earlier group flags are not standalone evidence of instrument failure. Physical-reference calculations and extraction candidates exist; clean light curves remain unestablished.
Next: Validate false-positive rate and power of the coherence statistic on realistic independent variability and shared calibration effects before interpreting real-data flags or adding corrections.
Existing reportPaper integration and recovery
100 obsolete held jobs retired with products preserved. F1 diagnostic probe exited 0; C1 continuation 320 also exited 0 and its native/F4 products exist. Execution success is not final paper science acceptance.
Next: Inspect recovered products and accept the ledger/training/extraction handoffs into one final recipe; keep preparation decisions explicit.
Existing reportEarlier joint-fit adapter and report
Idle with interrupted recent turns. Its Gaia-neighbour joint-fit adapter, splits and F1 baseline checks are reusable, but its README saying RUNNING does not mean augmented training occurred. The current matched-training owner is the main atlas task.
Next: Reuse audited implementation and evaluation splits through the active training owner; do not resume a duplicate training campaign.
Existing reportPhysical WCS and visualization
Idle; last visualization request was interrupted. VA/guide-direction and temperature-associated distortion work exists, with startup/roll limitations. The requested guide-direction 2D display is not confirmed complete by the latest task turn.
Next: Finish the pending display and separate measured pointing, roll, VA and temperature/time components; keep startup prediction limits visible.
Existing reportNonlinearity correction tests
Completed investigation; attribution and independent-sector transfer are inconclusive. No global correction accepted and no production images changed.
Next: Retain as a limitation/hypothesis; only reopen a bounded independent test if the final paper claims require it.
Existing reportNonlinearity literature and calibration provenance
Completed source review distinguishing measured ground response, SPOC application and unestablished residual calibration accuracy. No validated universal magnitude cutoff or new correction follows.
Next: Use as citation/provenance input, not as proof that residual electronic nonlinearity explains the measured ePSF trends.
Raw-FFI interpretation review
Completed review. Distinguishes evidence for brightness-dependent shape from identification of a physical brighter-fatter mechanism; atlas selection/estimator limitations remain relevant.
Next: Keep this distinction in captions/claims; do not promote a non-detection into a magnitude-independent PSF claim.
Existing reportTemporal status review
Not loaded; last completed answer summarized unresolved group trends and signal attenuation. This is a review of the temporal task, not an additional independent experiment.
Next: Use the active temporal owner's evolving record for current numbers.
Existing reportOutline and publication figures
Not loaded; last completed edits include the requested layouts, penalty explanation and development section-4 plates. Historical input/provenance warnings and convergence limits remain visible.
Next: Resume final figure/table production when accepted inputs arrive; maintain the existing canvas rather than creating another manuscript structure.
Existing reportNo agent was reassigned or messaged by this inventory. App activity is distinct from running computation, completed training and scientific acceptance. Private transcripts are not published here.