Outputs

This page documents every output artifact produced by the pipeline, with particular emphasis on the final merged catalog columns and their physical units.

Directory Layout

With default settings, outputs.work_dir contains:

work_dir/
├── EPSFs/                    # ePSF artifacts
│   ├── <band>/
│   │   └── r00_c00/
│   │       ├── epsf.npy          # ePSF array (normalized)
│   │       ├── epsf.png          # ePSF stamp visualization
│   │       ├── epsf_growth_curve.png      # Encircled-energy diagnostic
│   │       ├── epsf_residual_diagnostics.png  # Star minus ePSF residual diagnostics
│   │       ├── meta.json         # ePSF build metadata
│   │       ├── psfstars.png      # PSF star selection overlay
│   │       ├── psfstars_bgsub.png  # PSF star overlay on background-subtracted patch
│   │       ├── background_diagnostics.png  # Raw/background/bg-sub patch panels
│   │       ├── star_local_background_diagnostics.png  # Per-star raw/annulus/bg-sub diagnostics
│   │       ├── extracted_stars.png   # Extracted star cutout montage
│   │       └── used_stars.png        # Fitted star cutout montage
│   └── epsf_summary.json    # Summary across all bands and cells
├── patches/                  # Patch geometry definitions
│   ├── patches.csv
│   └── patches.json
├── patch_payloads/           # Compressed per-patch input data
│   └── r00_c00_pr00_pc00.pkl.gz
├── outputs/                  # Per-patch Tractor fit results
│   └── r00_c00_pr00_pc00/
│       ├── r00_c00_pr00_pc00.log          # Patch fit log
│       ├── r00_c00_pr00_pc00.runner.log   # Subprocess stdout/stderr
│       ├── r00_c00_pr00_pc00_cat_fit.csv  # Patch fit catalog
│       ├── meta.json                       # Patch metadata + PSF audit
│       ├── patch_overview.png              # Data/Model/Residual overview
│       └── cutouts/
│           ├── src_{ID}.png                # Per-source montage (named by source ID)
│           └── ...
├── cropped_images/           # Crop diagnostic plots
│   ├── white_before_crop.png
│   └── white_after_crop.png
├── bright_mask/              # Bright unmatched-source mask diagnostics
│   ├── white_bright_mask.fits       # Raw boolean mask (1 = masked)
│   └── white_bright_mask_overlay.png  # White-stack overlay with mask and detections
├── overlay/                  # White-stack overlay plots
│   ├── white_overlay.png
│   └── white_overlay_zoom.png
├── wcs.fits                  # Runtime WCS snapshot (working pixel frame)
├── ZP/                       # Zero-point calibration (if zp.enabled)
│   ├── input_catalog_*_with_Gaia.csv  # Augmented input catalog
│   ├── gaia_augmentation_overlay.png  # Source selection overlay
│   ├── zp_vs_mag__<band>.png              # Per-band ZP diagnostic plots (constrained y-range)
│   ├── zp_vs_mag__<band>_fullrange.png   # Per-band ZP diagnostic plots (adaptive y-range)
│   ├── zp_summary.csv                # Per-band ZP summary
│   └── zp_all_bands__per_object.csv  # Per-star ZP values
├── config_used_{timestamp}.yaml      # Snapshot of the config used for this run
├── {command}.log                     # Per-command log file (e.g. run.log, merge.log)
└── output_catalog.csv        # Final merged catalog

ePSF Patch Diagnostics

Each EPSFs/<band>/<epsf_tag>/ directory contains ePSF QA artifacts:

File	Description
psfstars.png	PSF-star selection overlay on the raw ePSF cell image.
psfstars_bgsub.png	PSF-star selection overlay on the background-subtracted ePSF cell image.
background_diagnostics.png	Three panels: raw patch, estimated background map, raw-minus-background. Written only when epsf.save_patch_background_diagnostics: true.
star_local_background_diagnostics.png	Per-star diagnostics (one row per used star): raw stamp, annulus sample pixels, background-subtracted stamp. Written only when epsf.save_star_local_background_diagnostics: true and local-background subtraction is enabled. If id_label is missing/duplicated, this file is skipped with a warning.
epsf_growth_curve.png	Encircled-energy growth curve of the final ePSF stamp.
epsf_residual_diagnostics.png	Median star/model/residual panel plus normalized residual histogram.

Per-cell meta.json includes local_bkg_diagnostics.local_bkg_diag_status to indicate whether star_local_background_diagnostics.png was written (ok) or skipped (skipped_*).

Per-Patch Output Directory

Each patch produces a subdirectory under outputs/ named by its patch tag (e.g. r05_c02_pr04_pc00). Contents:

File	Description
<tag>.log	Tractor optimization log: iteration-by-iteration dlnp values, convergence status, PSF choices.
<tag>.runner.log	Full subprocess command line and captured stdout/stderr. Useful for debugging crashes.
<tag>_cat_fit.csv	Per-patch fitted catalog. Contains all input columns plus fit result columns.
meta.json	Patch metadata including PSF audit (which bands used ePSF vs fallback), patch geometry, fitting mode, and source counts (n_sources for base sources, n_halo_sources for halo neighbors included in fitting but not in the output CSV). In multi-band mode, also includes optimizer summary (niters, converged, hit_max_iters). When staged fitting is enabled (patch_run.enable_staged_fit: true), includes staged_fit_enabled and staged_fit_stages (or staged_fit_stages_by_band for single-band) with per-stage iteration counts and convergence status.
patch_overview.png	Three-panel plot: Data (white stack), Model+Sky, Residual. Source positions shown as crosshairs.
cutouts/src_{ID}.png	Per-source cutout montage for base sources (one row per panel: Data, Model+Sky, Residual; one column per band). Crosshairs show original (lime, dashed) and fitted (magenta, solid) positions. Filename uses the source ID (e.g. src_00019.png, src_gaia_12345.png). If ID is absent, falls back to zero-padded index (src_000000.png). Filesystem-unsafe characters in the ID are replaced with underscores.
cutouts/src_halo_{ID}.png	Per-source cutout montage for halo sources. Same layout as base source montages, but crosshairs use cyan (fitted) and yellow (original) instead of magenta/lime. Halo sources are fitted as part of the patch model but their results are not included in the output CSV.

---

Final Merged Catalog

File: output_catalog.csv (or as set by outputs.final_catalog)

The merge stage performs a left join from the base catalog onto the concatenated patch fit results, using the merge key (ID if present, otherwise RA+DEC). The base catalog is the augmented catalog (when zp.enabled) or the original input catalog (otherwise). Either way, it includes all sources — both active and excluded — with their flag columns.

As a result:

• Every row in the base catalog appears in the output, even if the source was excluded from fitting.
• Fit columns are empty (NaN) for sources that were excluded or not assigned to any patch.

Input Catalog Columns (preserved)

All columns from the base catalog are preserved in their original form. These include ID, RA, DEC, TYPE, and any FLUX_*, ELL, THETA, Re, POS_ERR columns you provided, plus gaia_source_id and exclusion flag columns added during the load and augmentation stages. When zp.gaia_pos_err_pix is set, newly injected Gaia rows in the augmented base catalog also carry POS_ERR.

Exclusion Tracking Columns

These columns are added by the pipeline to indicate why a source was excluded from fitting.

Column	Type	Description
excluded_crop	bool	True if the source was flagged by crop filtering (fell outside the crop margin).
excluded_saturation	bool	True if the source was flagged by saturation-cut filtering (near saturated pixels).
excluded_any	bool	True if excluded by any reason (logical OR of all exclusion flags).
excluded_reason	string	Human-readable exclusion reason. Values: "crop", "saturation", "crop+saturation", or "" (not excluded).
affected_by_bright_mask	bool	True if the source's position falls inside a bright-source mask ellipse. The source is still fitted, but its surrounding pixels may be partially or fully masked (invvar = 0), potentially degrading fit quality. Only present when bright_mask.enabled: true.

Fitted Position Columns

The column names depend on the fitting mode (enable_multi_band_simultaneous_fitting).

Multi-band simultaneous fitting (default, true): position is shared across bands.

Column	Type	Unit	Description
x_pix_patch_fit	float	pixels	Fitted x position in the local patch coordinate frame.
y_pix_patch_fit	float	pixels	Fitted y position in the local patch coordinate frame.
x_pix_white_fit	float	pixels	Fitted x position in the full (post-crop) white-stack coordinate frame. Computed as x_pix_patch_fit + x0_roi.
y_pix_white_fit	float	pixels	Fitted y position in the full (post-crop) white-stack coordinate frame. Computed as y_pix_patch_fit + y0_roi.
RA_fit	float	degrees (ICRS)	Fitted right ascension, computed from (x_pix_white_fit, y_pix_white_fit) via WCS.
DEC_fit	float	degrees (ICRS)	Fitted declination, computed from (x_pix_white_fit, y_pix_white_fit) via WCS.

Single-band independent fitting (false): position is fitted independently per band. For each band {band}:

Column	Type	Unit	Description
x_pix_patch_{band}_fit	float	pixels	Fitted x position in the local patch frame for this band.
y_pix_patch_{band}_fit	float	pixels	Fitted y position in the local patch frame for this band.
x_pix_white_{band}_fit	float	pixels	Fitted x position in the white-stack frame for this band.
y_pix_white_{band}_fit	float	pixels	Fitted y position in the white-stack frame for this band.
RA_{band}_fit	float	degrees (ICRS)	Fitted right ascension for this band.
DEC_{band}_fit	float	degrees (ICRS)	Fitted declination for this band.

Fitted Flux Columns

For each band {band} present in the images (matching FILTER header values). Column names are the same in both fitting modes:

Column	Type	Unit	Description
FLUX_{band}_fit	float	scaled counts (nominal ZP = zp_ref)	Fitted flux in the given band. Images are scaled so that the nominal ZP equals zp_ref, but see note below.
FLUXERR_{band}_fit	float	scaled counts (nominal ZP = zp_ref)	Flux uncertainty (1-sigma), derived from the Tractor's parameter variance at the optimized solution.

Flux system and approximate vs calibrated magnitudes

All fitted fluxes are in the scaled system defined by image_scaling.zp_ref (default 25.0). The formula m_approx = -2.5 * log10(FLUX_{band}_fit) + zp_ref gives only an approximate AB magnitude. This approximation assumes ZP_AUTO in the FITS header is the exact zero-point, but in practice ZP_AUTO has errors, and the Tractor photometry method differs from whatever method was used to derive ZP_AUTO. For calibrated AB magnitudes, use the MAG_{band}_fit column (when zp.enabled), which applies the ZP derived from Gaia-matched stars: MAG_{band}_fit = ZP_median - 2.5 * log10(FLUX_{band}_fit). The offset ZP_median - zp_ref reflects primarily the error in ZP_AUTO. When ZP calibration is well-behaved, this offset is small. Note that ZP_median and ZP_AUTO live in different flux systems (scaled vs original), so comparing them directly is not meaningful — see Pipeline Behavior — ZP Calibration.

Fitted Morphology Columns

Multi-band simultaneous fitting (default, true): morphology is shared across bands.

Column	Type	Unit	Description
stype_fit	string	—	Source model used for fitting: "star", "exp", "dev", or "sersic".
sersic_n_fit	float	dimensionless	Fitted Sersic index (only for sersic type; NaN otherwise).
re_pix_fit	float	pixels	Fitted effective (half-light) radius in the Tractor's internal parameterization.
ab_fit	float	dimensionless	Fitted axis ratio b/a (range 0–1).
phi_deg_fit	float	degrees	Fitted position angle in the Tractor's internal convention (modulo 180).
ELL_fit	float	dimensionless	Fitted ellipticity: 1 - ab_fit. Comparable to the input ELL column.
Re_fit	float	pixels	Fitted effective radius (same as re_pix_fit; provided for naming symmetry with input Re).
THETA_fit	float	degrees	Fitted position angle, approximately converted back to SExtractor convention: phi_deg_fit - 90. Comparable to the input THETA column.

Single-band independent fitting (false): morphology is fitted independently per band. For each band {band}:

Column	Type	Unit	Description
stype_fit	string	—	Source type (same across bands, from input TYPE column).
sersic_n_{band}_fit	float	dimensionless	Fitted Sersic index for this band.
re_pix_{band}_fit	float	pixels	Fitted effective radius for this band.
ab_{band}_fit	float	dimensionless	Fitted axis ratio for this band.
phi_deg_{band}_fit	float	degrees	Fitted position angle for this band.
ELL_{band}_fit	float	dimensionless	Fitted ellipticity for this band: 1 - ab_{band}_fit.
Re_{band}_fit	float	pixels	Fitted effective radius for this band.
THETA_{band}_fit	float	degrees	Fitted position angle for this band (SExtractor convention).

Morphology columns for point sources

For stype_fit = "star", all morphology columns are NaN (in both fitting modes).

Optimizer Diagnostic Columns

Multi-band simultaneous fitting (default, true): one set of optimizer diagnostics per source.

Column	Type	Description
opt_converged	bool	True if the optimizer converged (dlnp dropped below patch_run.dlnp_stop).
opt_hit_max_iters	bool	True if the optimizer did NOT converge and effectively exhausted iteration budget (iterations ≥ n_opt_iters - flag_maxiter_margin).
opt_niters	int	Number of optimizer iterations actually performed.
opt_last_dlnp	float	Last delta-log-probability value from the optimizer.
flag_bound_stalled	bool	True if any hard-bounded parameter for this source is at or near its limit after optimization. This indicates the optimizer may have been stuck at a constraint boundary rather than converging to a true optimum.
flag_bound_stalled_which_parameter	string	Comma-separated list of parameters at their bounds, encoded as {param}@{side}. Examples: pos.x@upper, shape.logre@lower, sersicindex@upper. Empty string if flag_bound_stalled is False.

Single-band independent fitting (false): per-band optimizer diagnostics. For each band {band}:

Column	Type	Description
opt_converged_{band}	bool	Whether the optimizer converged for this band.
opt_hit_max_iters_{band}	bool	Whether the optimizer exhausted iterations for this band.
opt_niters_{band}	int	Number of iterations for this band.
opt_last_dlnp_{band}	float	Last dlnp for this band.
flag_bound_stalled_{band}	bool	Whether any bounded parameter is at its limit for this band.
flag_bound_stalled_which_parameter_{band}	string	Parameters at their bounds for this band (same encoding as multi-band).

Patch and PSF Audit Columns

Column	Type	Description
patch_tag	string	Patch identifier (e.g. r05_c02_pr04_pc00).
epsf_tag	string	Parent ePSF cell identifier (e.g. r05_c02).
psf_min_epsf_nstars_for_use	int	Quality gate threshold used for this patch.
psf_used_epsf_band_count	int	Number of bands that used a real ePSF.
psf_fallback_band_count	int	Number of bands that used a fallback PSF.
psf_low_star_band_count	int	Number of bands where ePSF existed but was rejected due to low star count.
psf_fallback_bands	string	Comma-separated list of bands that used fallback PSF.
psf_low_star_bands	string	Comma-separated list of bands with low-star ePSF rejection.
psf_frozen_bands_for_optimizer	string	Comma-separated list of bands where GaussianMixturePSF params were frozen for optimizer stability.
psf_fallback_reasons_json	string (JSON)	JSON dict mapping band → fallback reason string.

Reliability Summary Columns

These columns provide a single-column summary of source fit reliability, computed during the merge stage from the exclusion, optimizer, bound-stall, and bright-mask flags above.

Column	Type	Description
fit_is_reliable	bool	True if the fit is considered trustworthy. False if any hard-fail condition applies (see below).
fit_quality_issues	string	Comma-separated list of all quality concerns for this source (both hard and soft). Empty string if none.

Hard-fail conditions (set fit_is_reliable = False):

Tag	Trigger
excluded	excluded_any is True.
no_fit_result	Source is not excluded but has no fitted position (patch crash or unassigned).
hit_max_iters	opt_hit_max_iters is True (any band in single-band mode).
bound_stalled	flag_bound_stalled is True (any band in single-band mode).
bright_mask	affected_by_bright_mask is True.

Soft warnings (recorded in fit_quality_issues only, do not affect fit_is_reliable):

Tag	Trigger
not_converged	Optimizer did not formally converge but did not hit max iterations either.
all_psf_fallback	All bands used an analytic fallback PSF instead of ePSF.

Internal Coordinate Columns

These columns are added during patch input building and carried through to the output. They are useful for debugging spatial assignments.

Column	Type	Unit	Description
x_pix_white	float	pixels	Source x position in the white-stack frame (from WCS projection of input RA/DEC).
y_pix_white	float	pixels	Source y position in the white-stack frame.
x_pix_patch	float	pixels	Source x position in the local patch frame (= x_pix_white - x0_roi).
y_pix_patch	float	pixels	Source y position in the local patch frame.

Gaia Augmentation Column

When zp.enabled: true, the augmented catalog adds this column:

Column	Type	Description
gaia_source_id	string	Gaia DR3 source_id for sources matched to (or injected from) the GaiaXP synphot catalog. Empty string for non-Gaia sources.

Zero-Point Calibrated Magnitude Columns

When zp.enabled: true, the ZP computation stage adds these columns for each band {band}:

Column	Type	Unit	Description
MAG_{band}_fit	float	mag (AB)	AB magnitude: ZP_median - 2.5 * log10(FLUX_{band}_fit). NaN for sources with non-positive flux.
MAGERR_{band}_fit	float	mag	Magnitude error: sqrt((2.5/ln10 * FLUXERR/FLUX)^2 + ZP_err^2). Propagates both flux error and per-band ZP uncertainty. ZP_err is either zp_err_mad or zp_err_std depending on zp.zp_err_method (default "all_mad"). NaN if flux error is missing.

---

Diagnostic Plots

Bright-Source Mask Diagnostics

When bright_mask.enabled: true (default), the bright_mask/ directory contains:

File	Description
white_bright_mask.fits	Boolean FITS mask (1 = masked pixel). Combines bright unmatched-source ellipses with any-band saturation mask.
white_bright_mask_overlay.png	White-stack image with the mask shown as a semi-transparent red overlay. Four marker categories are shown (see below). A legend with counts is included.

The overlay plot uses two independent matching passes with bright_mask.match_radius_pix:

Marker	Color/Shape	Meaning
Open circle	Lime	SEP matched — SEP detection that has a matching active (non-excluded) catalog source nearby. These are known sources; not masked.
X	Red	SEP unmatched — SEP detection with no catalog match. These bright contaminants generate the mask ellipses.
Diamond	Blue (deepskyblue)	Cat matched — input catalog source (including excluded sources) that has a nearby SEP detection.
Triangle	Orange	Cat unmatched masked — catalog source with no nearby SEP detection whose position falls inside the bright mask (affected_by_bright_mask = True).

Note: "SEP matched" and "Cat matched" are computed from opposite directions and differ subtly. The SEP-to-catalog match only considers active sources (excluded_any = False), so a SEP detection near an excluded source appears as "SEP unmatched" even though the excluded source appears as "Cat matched". Multiple SEP detections can match the same catalog source, and vice versa.

White-Stack Diagnostics

File	Description
cropped_images/white_before_crop.png	White-stack image with crop box overlaid (green rectangle). Generated before crop is applied. Only when crop.enabled: true and crop.plot_pre_crop: true.
cropped_images/white_after_crop.png	White-stack image after crop. Only when crop.enabled: true and crop.plot_post_crop: true.
overlay/white_overlay.png	Current working white stack with source positions color-coded by TYPE: cyan=STAR, magenta=GAL/EXP/DEV/SERSIC, yellow squares=UNKNOWN, red X=saturation-excluded, gray=crop-excluded (legend only). Generated when overlay.enabled: true, regardless of crop.enabled.
overlay/white_overlay_zoom.png	Zoomed region of the overlay. Generated when overlay.zoom_enabled: true. Out-of-bounds areas are blank-filled.
wcs.fits	Runtime WCS snapshot in the working pixel frame (post-crop if crop enabled, full-frame otherwise). Used by merge to compute RA_fit/DEC_fit.

Patch Overview

patch_overview.png in each patch output directory shows a three-panel view (Data, Model+Sky, Residual) of the white-combined patch images. When per-filter patch data is available, the white combination uses initial (pre-mask) inverse variance weights derived from the sky sigma, so that saturated and bright-masked pixels receive physically correct weights rather than falling back to equal-weight. The red overlay and display scaling use the final Tractor mask (non-finite | saturation | bright mask): pixels with invvar = 0 in all bands are shown with a semi-transparent red overlay, and the color stretch is computed from unmasked pixels only. Source positions are marked:

Base sources:

• Orange X — original (input catalog) positions
• Deepskyblue X — fitted positions
• Cyan line — connecting line from each original position to its fitted position(s)

Halo sources (fitted for neighbor modeling but not output to CSV):

• Yellow diamond — original positions
• Lime diamond — fitted positions
• Lime line — connecting line from original to fitted position

The title reports both base and halo counts (e.g. n_sources=2 + 3 halo).

In single-band mode, each source has multiple blue markers (one per band, with reduced alpha/linewidth to avoid clutter), each connected to the single orange marker by a cyan line. In multi-band mode, there is one blue marker per source.

Source Cutout Montages

cutouts/src_{ID}.png (or cutouts/src_halo_{ID}.png for halo sources) shows a grid of panels: one column per band, three rows (Data, Model+Sky, Residual). The filename uses the source's ID value (e.g. src_00019.png, src_gaia_2518065009826205440.png, src_halo_gaia_12345.png). If ID is absent, a zero-padded index is used as fallback.

Markers for the center source (base):

• Magenta solid crosshair — fitted position
• Lime dashed crosshair — original (input) position

Markers for the center source (halo):

• Cyan solid crosshair — fitted position
• Yellow dashed crosshair — original (input) position

Markers for other base sources in the cutout:

• Deepskyblue X — fitted position
• Orange X — original position
• Cyan line — connecting line

Markers for other halo sources in the cutout:

• Lime diamond — fitted position
• Yellow diamond — original position
• Lime line — connecting line

Masked pixels (where invvar = 0, including bright-mask and bad-pixel regions) are shown as a semi-transparent red overlay on all three panels (Data, Model+Sky, Residual). The color stretch is computed from unmasked pixels only, so faint sources near masked bright contaminants are displayed with full dynamic range.

In single-band mode, each band column shows that band's own independently fitted position for the crosshairs and markers.

ZP Diagnostics

When zp.enabled: true, the ZP/ directory contains:

File	Description
input_catalog_*_with_Gaia.csv	Augmented input catalog with Gaia-matched gaia_source_id column and (if inject_gaia_sources: true) injected Gaia source rows. When zp.gaia_pos_err_pix is not null, newly injected Gaia rows also include POS_ERR.
gaia_augmentation_overlay.png	White-stack overlay showing original sources (cyan circles), Gaia-matched originals (lime squares), injected Gaia sources (magenta circles, only when inject_gaia_sources: true), saturation-excluded Gaia sources (red X), saturation-excluded input sources (red X), crop-excluded input sources (orange X), and the selection bounding box (yellow dashed).
zp_vs_mag__{band}.png	Per-band ZP diagnostic scatter plot (constrained y-range): ZP vs GaiaXP synphot AB magnitude. Shows unreliable (gray diamonds), sigma-clipped (red X), and kept sources with error bars. When zp_err_method: "bright_std", kept sources are split into high-SNR (black circles) and low-SNR (grey circles) based on the active SNR threshold; the shaded band shows the std error. When zp_err_method: "all_mad", all kept sources are shown as black circles and the shaded band shows MAD. Pale grey vertical bands mark magnitude ranges outside the GaiaXP catalog coverage. x-axis: plot_mag_lo..plot_mag_hi; y-axis: median +/- plot_zp_offset.
zp_vs_mag__{band}_fullrange.png	Same as above but with adaptive y-axis range (based on data points, not error bars).
zp_summary.csv	Per-band summary: band, n_raw, n_kept, n_clipped, n_unreliable, clip_sigma, zp_median, zp_err_mad (MAD of all kept sources), zp_err_std (std of high-SNR kept sources), n_zp_err_src (number of sources used for std), zp_err_snr_used (actual SNR threshold used after any fallback). Both error columns are always populated regardless of zp_err_method.
zp_all_bands__per_object.csv	Per-star ZP values for all reliable Gaia-matched sources across all bands, with is_kept/is_clipped flags from sigma clipping. Unreliable sources are not included.

---

Merge Behavior Details

The merge stage:

1. Reads the base catalog (augmented catalog if ZP is enabled, otherwise the original input catalog). Exclusion flag columns (excluded_crop, excluded_saturation) are already present from the load stage and are used directly.
2. Computes excluded_any and excluded_reason from the flag columns.
3. Collects all *_cat_fit.csv files from per-patch output directories.
4. Concatenates fit results and handles duplicate merge keys (drops duplicates with a warning; rare boundary sources may appear in two patches).
5. Performs a left join: base_catalog.merge(fit_results, how="left", on=key_cols).
6. Computes sky coordinates from fitted pixel positions using the runtime WCS snapshot (wcs.fits) generated during load_inputs(). In multi-band mode: RA_fit/DEC_fit. In single-band mode: RA_{band}_fit/DEC_{band}_fit for each band.

Merge Key Selection

• If ID column exists → use ID as the sole merge key.
• Otherwise → use (RA, DEC) as a composite merge key.

What "merge-missing" Means

The merge log reports merge-missing (left join rows without fit): N. This count includes:

• Sources excluded by crop filtering.
• Sources excluded by saturation filtering.
• Sources with NaN coordinates that could not be assigned to patches.
• Sources in patches that failed or crashed.

Check the excluded_* columns to distinguish intentional exclusions from unexpected failures.

Merge Logging

The merge stage logs:

• Output file path
• Row and column count
• Exclusion counts (if exclusion columns are present): crop, saturation, any
• Merge-missing count (left-join rows without fit results)