[xrfez]

Phase 11: survey QC on top of parseSEG 0.9.0's survey geometry. Data in, values out (D1, D14): parseSEG reads and writes SPS, SEG-Y and SEG-D.

What's new

• Station-first join (parseSEG 0.9.0): traces match the survey by their source and receiver stations; the FFID is a fallback only. This is the Morgantown case: FFID 151 is XPS record 77.
• survey/patch:
  • patch description of an X survey (a lines × channels template, or a distance box with clipped edges);
  • X generation from a template or a distance field (roll, shift or static edges; source placed by coordinates or by numbers; holes in R start new runs);
  • compareRelations for agreement with a supplied XPS.
• qc/fold:
  • surveyGrid, an automatic bin grid: receiver azimuth, natural bin, centred on the midpoints' dominant phase, stray stations trimmed;
  • pre-plot fold maps (offset classes, azimuth sectors, offset range) from X or O relations;
  • actual fold from live traces in any order;
  • foldDifference.
• qc/record: completeness of a field record against the XPS. The record is found by source station or FFID, with the other key checked; traces match by station or channel. It reports missing ranges, unlisted stations, duplicates and per-line counts; summarize totals many records.
• qc/amplitude:
  • per trace: RMS and max in dB, DC, non-finite, clipped runs and spikes, with dead/constant/clipped/spiky/DC flags;
  • per record: median and MAD, weak/noisy outliers by robust z;
  • a ProcessPool form.
• Docs: guide page docs/guide/survey-qc.md (runnable examples), API reference, README.

Morgantown (local example files, skipped in CI)

• Join: all 3954 live traces joined by station.
• Completeness: XPS 77, with 1127 expected, 1096 present, 31 missing and 2858 unlisted, in 0.65 ms.
• Fold: the fold difference against XPS 77's pre-plot reproduces those counts bin by bin.
• Pre-plot fold: the whole survey (11.2 M pairs) takes 1.58 s in release.
• Amplitude QC: 27.5 ms on a pool of 4 for 3957 × 4001 samples.

Review

A fresh-agent review gate found 10 issues. Fixed:

• spike-sum cancellation after huge spikes;
• stray stations inflating the grid;
• a NaN FFID reported as a mismatch;
• station indices dropped from the match;
• Inf counted as a clip;
• argument guards.

Multi-component fold counting is documented. The parseSEG relations cost is deferred.

Version 0.2.0 (also covers Phase 10, PR #2).

🤖 Generated with Claude Code