1 reporting.scorer.mac_q1_synthesis

reporting.scorer.mac_q1_synthesis

Conditional, physical-scene post-hoc synthesis of sealed Mac Q1 evidence.

Native owners admit candidate support and compute state metrics. This module reduces states equally within scenes and scenes equally; bootstrap intervals resample scenes with fitted weights fixed. They exclude seed, checkpoint selection and unobserved-population uncertainty. Recording positions describe collector prefixes, not causal interventions on trajectory depth.

1.1 Functions

Name	Description
build_mac_q1_synthesis	Build immutable analysis tables from authenticated completed evaluations.
synthesis_frame	Return a detached frame; modifying it cannot change sealed results.

1.1.1 build_mac_q1_synthesis

reporting.scorer.mac_q1_synthesis.build_mac_q1_synthesis(
    saved_report,
    *,
    root,
    draws=10000,
    seed=20261006,
    legacy_selected_update_receipts=False,
)

Build immutable analysis tables from authenticated completed evaluations.

Candidate indices, original labels and unique physical-state identities must pair exactly across final, selected and constant roles. Engineering results supply census status only. Optional metrics retain metric-specific support. MSE decomposition is native: mean(error²) = mean(error)² + var(error). Native version-two products retain admitted selected-update counts. Legacy products leave that marker unavailable by default. Explicit legacy_selected_update_receipts=True reopens receipt-bound legacy attempts when their original paths are available; it never upgrades the saved historical format or changes the numerical observations. Concentration and phase fractions describe observed error mass, not causal importance. All confidence intervals are exploratory pointwise percentile scene bootstraps; no checkpoint is reselected and no model is evaluated.

1.1.2 synthesis_frame

reporting.scorer.mac_q1_synthesis.synthesis_frame(snapshot, table_name)

Return a detached frame; modifying it cannot change sealed results.