The original southern field and a second ζ Cancri pointing are both recovered. Sparse independent checks still limit distortion claims.
The original southern field is retained. A separate pattern identifies ζ Cancri at RA 123.05973°, Dec +17.62437°, consistent with the margin. Equal-looking paired images can exchange brightness rank, so choosing the brighter member does not define a consistent exposure frame. This is a registration problem before it is a glass-distortion problem.
The same star often occurs at two displaced positions. Fifty accepted pairs define a useful grouped source list.
The recovered footprint is only about 0.764 × 0.611 degrees. Source morphology and grouping matter more than the apparent size of the scanned glass.
The revised layer uses local contrast in both polarities, expanded inspected writing regions and contextual recovery of short fragments and round digits. Known photographic targets and sampled spectral apertures are protected. It is not OCR and does not reconstruct hidden sky or invent pen strokes.
The selection covers 1.39% of this scan (410,525 pixels). Faint marks, scratches and contacts between ink and photographic images remain ambiguous. The cyan overlay shows the wider exclusion footprint used by the photographic remainder; the stroke mask is supplied separately.
The writing PNG is a white, contrast-rendered stencil of bright and dark observed marks. The observed-pixel PNG preserves their stretched scan appearance. Both use original scan coordinates. The photographic remainder displays excluded pixels as black holes. It is neither a clean-sky reconstruction nor a new photometric product. Keep the mask alongside any measurements. Native original-intensity TIFF/FITS products, where generated, remain in the analysis workspace and are not duplicated in this handoff.
This revision recovers short letter fragments, round digits and inspected dark marks, with broader writing regions. The stroke selection grows from 100,554 to 410,525 pixels (4.08×). The photographic remainder uses a separate 3-pixel exclusion margin, covering 2.27% of the scan.
The wider selection deliberately favors removing writing. Nearby stars, scratches and ambiguous fragments can also be excluded. Explicit photographic protection regions retain known target images and sampled spectra; this does not establish perfect separation elsewhere.
The blind grouped-source solve uses index 4206-09, with 23 solver matches and odds 1.09 × 10⁴⁴. Its center is 08h 17m 25.72s, −11° 46′ 33.30″. Common shifts of (33.02, 1.85) and (29.73, 0.64) binned pixels correspond to roughly twice those displacements on the original scan. A separate label-assisted solve identifies ζ Cancri with 13 solver matches and four additional reference checks; it has its own WCS.
The DSS2 image is resampled into the recovered plate orientation and footprint. Its photographic epochs, bands, resolution and contrast differ. Brightness and morphology need not match. It is context from the recovered WCS, not an additional independent fit check.
The original southern field is retained. A separate pattern identifies ζ Cancri at RA 123.05973°, Dec +17.62437°, consistent with the margin. Equal-looking paired images can exchange brightness rank, so choosing the brighter member does not define a consistent exposure frame. This is a registration problem before it is a glass-distortion problem.
Center RA 123.05973123°, Dec +17.62436532°. After excluding reference stars near solver inputs, 4 additional associations have median/p90 residuals 1.880/2.644 native pixels. Shifted controls yield 0, 0, 0, 0, 0, 0 associations.
Only four additional stars are available: this falls below the predeclared eight-star validation gate. The 13 solver matches, four extra checks and written target support the identification, but do not establish precision distortion calibration.
This is a separate coordinate frame for one sky/exposure component. Residuals use a finite association gate and no historical proper-motion propagation; they validate location rather than isolate glass distortion.
Across these scans, repeated photographic images need not imply the same mechanism. Shutter timing, guiding offsets, objective optics and different pointings can leave related patterns. Establish consistent component membership before fitting a smooth warp. No result here isolates glass deformation from telescope optics, emulsion, scanner geometry and centroid bias.
Twenty-five reference associations yield 19 training stars and just five held-out stars. The TAN median is 3.299 native pixels (about 1.49 arcsec); the 90th percentile is 10.301 pixels (about 4.66 arcsec). Five test stars give very limited information about the residual distribution.
Retain the field WCS as a coordinate solution for exposure-group centers. There is no defensible spatial glass correction from this sample, and no resampled plate product is claimed.
| Model | Held-out median | Held-out p90 | Retained |
|---|---|---|---|
| TAN | 3.299 | 10.300 | Yes |
The reference footprint describes where the model is constrained. A smooth polynomial outside that footprint can look plausible while remaining unmeasured.
Reference catalogs were not propagated to the plate epoch. Proper motion, centering errors, repeated images, optical projection, emulsion, the glass and the scanner may all contribute to residuals. One scan cannot separate these physical causes.
Batch held-out stars were excluded from distortion fitting, but may have participated in the initial field identification. They are not a wholly independent astrometric experiment.
Select a numbered pin or a note. Locations are editorial review pointers, not measured confidence contours; plate-wide limitations also apply.
Pairs may blend or lose one member. A grouped center can move even if the true star position does not.
Five held-out stars cannot establish a stable local distortion map. Refer to the coverage plot rather than assuming uniform precision.
The ink extraction is a reviewed candidate layer; faint strokes and photographic overlaps remain ambiguous.
Measure both exposure components with a shared point-spread model, infer a consistent translation, and obtain more epoch-corrected reference stars across the field before fitting curvature.
The sky identification is convincing; the distortion precision remains limited by sparse checks and repeated-image centroids.
All scientific numbers are snapshots of the saved analysis products. The original 16-bit scans remain the measurement source. JPEG derivatives, contrast stretching and masked previews are inspection tools; do not derive photographic density or absolute response from their appearance.
