Fix the marking and accuracy columns, and watch each cycle for parsing errors

[RISCfuture]

Testing the parser against the current distribution (cycle 20260802, published 2026-09-14) reported zero errors — while misreading one field on every record. This fixes that, then adds the workflow that would have caught it.

The bug

Per the FAA README shipped inside the distribution, the tail of each record is:

Column (1-idx)	Spec	Parser read it as
98	Horizontal Accuracy (1–9)	✓ correct
100	Vertical Accuracy (A–I)	marking ✗
102	Mark Indicator (P W M F S N U)	faaIndicator — declared, never read ✗

MarkingType's raw values were A–I, coinciding exactly with the vertical accuracy code set, so every record parsed "successfully" with a wrong value and vertical accuracy was never exposed. A column tally over all 652,785 records confirms it: column 100 holds only A–I, column 102 only P W M F S N U. The test fixture pinned the wrong answer — … R 5 D M … asserted marking == .paintAndFlags, reading the D that means ±50 ft height tolerance.

The distribution data is correct; the parser was wrong.

Changes

• HorizontalAccuracy (1–9) and VerticalAccuracy (A–I) replace AccuracyCategory, which mixed both code sets. Breaking.
• MarkingType rewritten to the real codes. Breaking.
• Obstacle.verticalAccuracy added; marking now reads column 102.
• Cases are named for the code the FAA publishes. These are not a file-format encoding — the FAA has used Accuracy Codes since 1979 and still writes them as "Accuracy Code 1A" (PARC recommendation); the spec's own columns read Code and Tolerance, never "category".
• DOF Cycle Watch workflow, modelled on SwiftNASR's, parses each cycle as it is released and opens an issue on parse errors, count drift, or a wrong currency date.
• SwiftDOF_E2E --report / --baseline produce the JSON the workflow reads; the tool now exits non-zero when any line fails to parse.

Verification

• All 652,785 records cross-checked against an independent read of the raw columns — horizontal accuracy, vertical accuracy, marking, lighting, action, verification, AGL, MSL, quantity: 0 mismatches.
• The previous distribution (20260607, 641,207 obstacles) also parses with 0 errors.
• All 11 supplemental files parse with exact record counts and 0 errors.
• Cycle arithmetic checked against the FAA's published schedule, including DOF (Next); live probe returns 200 for released cycles and 404 for the unreleased one.
• Drift and parse-error detection both exercised with negative tests.
• Drift tolerances calibrated on real data: a normal 56-day cycle moves the total 1.8%, against a 5% tolerance, so a clean cycle reports no drift.
• swift test (64), swiftlint --strict, swift format lint --strict, actionlint, yamllint, and DocC all clean.

Notes

• The combined 4D / 1A accuracy code is not exposed as a property; say the word if it would be useful. The two codes dominate the file (322,448 and 159,603 obstacles).
• Unrelated FAA data inconsistency spotted: obstacle 01-000525 is in DOF.DAT but missing from the per-state 01-AL.Dat extract.

🤖 Generated with Claude Code

[RISCfuture]

Superseded — these commits went directly into main ahead of Modernize for Swift 6.4 as 0cde9b0 and 1b28a73.