October 4, 2026. Exact archived-model replay, not final model acceptance.
Conclusion: structured residuals remain across all five fields. This baseline does not establish intrinsic brightness dependence, calibrated statistical significance, physical flux bias or noise-level closure. It separates the full A3 model from the local achromatic neighbour pilot and supplies native pixels for independent tests.
The source fits use archived F=4 bootstrap hp_d, a 6×6 spatial ePSF grid and 17 colour coefficients, Gaia-XP u with recorded fallback, lp7s=3e8, no brightness-width term and no fitted background. Every saved final stage reaches its step cap. Newer template_v3 products have not yet been propagated into these observations. Source IDs, saved states, unique-pixel ownership and independently reconstructed scene models pass the export audit.
The primary grids use catalogue Tmag and the actual fitted colour u. All displayed stacks are centred on saved catalogue/scene coordinates, without fitting away centroids. Exact fitted WCS plus chromatic renderer origins are preserved separately in geometry.npz. Robust stack clipping and population differences can affect appearance; these images are descriptive.
Own-source finite-stamp template sums are not forced to one; the original full-profile rendering and normalization are preserved. Full model divided by target amplitude includes overlapping source light. No plotted residual projection is claimed as an absolute flux-bias estimate.
Method and implementation record · All image inspection hashes · Independent export audit
| Field | Scene sources | Raster stars | Active unique pixels | Bins with ≥5 stars | Stop |
|---|---|---|---|---|---|
| C4 | 19,120 | 14,777 | 2,317,814 | 81/81 | stage 3 cap, 1999 |
| F1 | 14,733 | 11,477 | 1,963,587 | 81/81 | stage 3 cap, 1999 |
| F2 | 15,521 | 9,498 | 1,707,467 | 81/81 | stage 3 cap, 1999 |
| S22 | 12,543 | 10,023 | 1,667,055 | 81/81 | stage 3 cap, 1999 |
| C1 | 10,984 | 8,962 | 1,645,586 | 81/81 | stage 3 cap, 1999 |
Matched population controls · Numerical estimator controls · Template-only response · Two-shift construction
Held-core colour-off positive control
Fixed-width sensitivity construction record · Fixed-scene row identity audit
Structured core and wing residuals persist across position. Red columns show asymmetric patterns; the bright and faint rows have different counts and population mixes.
Input hashes, recipe and replay · Bin counts · Coordinate convention
Core/ring structure appears in blue and intermediate colour populations. Red-star residuals toward the right of the CCD are strongly asymmetric; a radial width metric cannot summarize the full pattern.
Input hashes, recipe and replay · Bin counts · Coordinate convention
Blue and intermediate populations retain core/ring structure. Red populations retain spatially varying multi-lobed residuals; the sparsest bright/red panels need particular caution.
Input hashes, recipe and replay · Bin counts · Coordinate convention
A broad positive residual background occurs in the right-middle spatial cell, alongside central and asymmetric components. A width-only attribution would miss this component.
Input hashes, recipe and replay · Bin counts · Coordinate convention
Core residuals and spatially varying red-star asymmetries remain. The plotted differences are population residuals, not independently measured magnitude-specific ePSFs.
Input hashes, recipe and replay · Bin counts · Coordinate convention