Portable contents
Share the fourteen named HTML files and the complete assets directory. Keep their relative locations unchanged. index.html is a convenient entry point. Open any page in a current browser; no web server, network connection, Python installation or external font is required. External reference links are optional reading.
The original TIFFs and large original-intensity FITS/TIFF analysis products are not duplicated in this package. Each report records the original SHA-256 and dimensions. Shared JPEGs are display derivatives; the transparent annotation and binary mask PNGs retain the native scan dimensions unless explicitly marked rectified.
Measurement chain
The analysis uses a local Astrometry.net 0.98 build with 4100-series Tycho-2 and 4200-series 2MASS indices. Inspected annotation masks reduce handwriting contamination; compact round detections provide source lists. Repeated-exposure plates require grouping rather than treating each photographic image as an independent star. Solver odds describe a particular model search, not a universal probability that every subsequent calibration claim is correct.
Saved catalog associations are split for fitting and held-out model checks. Batch held-out stars may have participated in field identification; UV1642’s 232 check stars were absent from the blind input. Held-out data were used to choose model complexity, so these residuals are internal model-selection estimates, not an external accuracy certification. Matching gates can exclude large residuals. The UV1642 check sample was selected within 3.5 binned pixels of the initial WCS.
TAN is the tangent-plane coordinate model. SIP adds polynomial pixel-coordinate terms. WCS headers map original scan coordinates to sky coordinates; a WCS does not by itself resample an image. UV1642 has a separate cubic SIP-to-TAN rectified preview. Its local residual interpolator is a separate experiment, not part of that rectification. Cubic interpolation is not flux conserving.
The saved field fits do not propagate catalog proper motions to the plate epoch. A separate illustrative Castor epoch check does propagate its system proper motion; it is not a refit of the field WCS. Optical projection, glass, emulsion, scanner geometry, centering and repeated-exposure grouping all affect empirical residuals. A single scan cannot isolate glass-only distortion. Border values outside reference support are extrapolations.
Layer interpretation
Revision 2 uses local contrast in both polarities, expanded inspected writing regions, contextual recovery of short and round fragments, and explicit photographic protection. The white writing PNG is a contrast-rendered stencil; the observed-pixel PNG preserves stretched scan appearance. A separate exclusion mask pads selected strokes by three native pixels. The previous mask and photographic remainder are retained for comparison. It is neither OCR nor a physical separation of superposed light. Transparent writing PNGs are display selections; they may retain scratches or miss faint disconnected fragments. A star under an ink stroke cannot be reconstructed from this one observation. The photographic remainder uses the expanded exclusion mask and shows excluded pixels as black, with no inpainting. Additional faint marks, stars near text and scratches can be included. Existing distortion fits have not been recomputed by the mask revision. New component identifications and additional-star location checks are documented separately.
The OP photographic bars and the UV1901 structured target images were explicitly protected after inspection. Per-plate masks remain reviewable hypotheses, not ground-truth labels. UV1642’s catalog-proximity intersection mask flags possible ambiguous locations and does not identify hidden stars with certainty.
Spectral profile extraction
IB5894 has three A–B–A records extracted with constrained specreduce traces; the old two-track extraction is superseded. OP966 and OP967 have repeated banded photographic records. Quadratic or linear center traces are fitted on 2× binned images. Final profiles sample the original 16-bit TIFF with linear interpolation across the trace, one sample per native Y pixel. Adjacent sidebands estimate the OP background; CSV files retain aperture means, background and background-subtracted response, with OP sideband scatter where available. Very low-contrast OP candidates are rejected. Coordinates, apertures and cautions are recorded in spectra.json.
Ten selected OP tracks now have provisional Balmer-reference wavelength coordinates. Hγ/Hδ anchor an inverse-square dispersion relation, and three unused lines test the assignment before all five are refitted. These reference coordinates do not independently measure observed wavelengths or redshift. No IB5894 wavelength solution is accepted; a bounded RASCAL iron-atlas trial is documented. No physical flux or photographic-density calibration is implied. Individual scanner/emulsion response and saturation remain uncalibrated. Overview curves are independently normalized for pattern inspection. IB5894 uses bounded, non-crossing spatial lanes, a nine-pixel fractional aperture, and two external background sidebands that exclude all three records. Its middle trace is extrapolated within its lane where the record is faint. Native Y is retained; cross-dispersion rectification interpolates the scanner values once.
Images and coordinates
Whole-plate JPEGs preserve stored scan orientation. The 1st–99.7th percentile linear stretch is recorded per scan and clipped to 8 bits; the largest display derivative fits within 2600 × 2600 pixels. This is not photographic-density calibration. Detail crops are recorded as fractional original-coordinate boxes; long margin crops are rotated and mirrored for reading. Original-coordinate layers and WCS headers are not changed by that presentation choice.
DSS2 color images were retrieved from CDS HiPS2FITS using the recovered WCS, with a vertical flip from the service JPEG into stored scan-row order. They have different bands, epochs and resolution from the historical plates. They are context images generated from the solution and must not be counted as additional independent fit validation. The position-only catalog panels use uniform dots, not synthetic catalog photometry.
Coverage plots dim the area outside the convex hull of saved associations. Cyan marks held-out stars; white rings mark held-out residuals above two native scan pixels. For pi4197-98 the plot contains solver correspondences only. Numbered uncertainty pins are editorial review locations, not fitted error contours.
Provenance and reproduction
Each assets/PLATE directory contains source.json, the extraction record, and any solution, WCS, reference-support data or survey provenance available. Identified fields also include small reference-catalog FITS subsets and saved association/correspondence tables. UV1642 includes a numerical warp grid and the separate local-residual model with its application recipe. UV1642 additionally contains its full distortion-model comparison; R4073 includes the rejected candidate and field-hint record; UV1693 includes source-repetition and pixel-control audits. manifest.json records the SHA-256 and size of every package file except the manifest itself, the QA record and this explanatory file (which are verified by the archive checksum).
UV1693 is newly identified from actual anchor members of repeated triples. UV1642 and pi4146–47 have additional recovered sky components, each with a separate WCS and location-check record. The ζ Cancri component has only four additional stars and does not pass the stricter eight-star validation gate. All component checks exclude references close to solver inputs before association. UV1693 and Kap 41 use a finer reference index for those checks. These are location checks through finite matching gates, not new distortion fits.
The source scripts shipped under analysis-source are the exact local pipeline and report generators used for this snapshot. They document commands and algorithms but require the original TIFFs, local environment, saved results and astrometric indices to recompute the analysis; the HTML package alone is an evidence handoff, not a replacement for those inputs.
Typical workspace commands: .env/bin/python scripts/lab_report_assets.py all, .env/bin/python scripts/lab_report_figures.py, then .env/bin/python scripts/plate_spectra.py, .env/bin/python scripts/plate_annotations_v2.py and .env/bin/python scripts/build_lab_reports.py. The original pipeline is documented in the included workspace notes. Native scan pixels and FITS one-based coordinates must be handled consistently; compact-source extraction generally uses 2× binned images.
User-supplied provenance initially described northern-hemisphere plates from the main Yerkes telescope around 1930. This is not independently established for every scan. Visible inscriptions appear to span 1901–1950, including UV1642’s apparent 1916 date. These visual readings and the disparate morphologies make per-plate logbook checks essential. Dates and telescope identity are not inferred from filenames alone.
Credits
Astrometry.net provides the local field solver. Tycho-2 and 2MASS indices supply reference positions as recorded in the analysis. Survey context: Digitized Sky Survey / STScI, distributed via CDS HiPS and HiPS2FITS. See the per-image request records and the STScI DSS archive and CDS service documentation.
Iosevka Mono is bundled as a WOFF2 subset from the locally installed Iosevka Nerd Font Mono regular and semibold faces, under the SIL Open Font License. Report palette: black, light grey and limited cyan. No runtime external dependencies are used.