Measure the Greek grant list's coverage; it is complete for the harvest

[brianglass]

Closes the loose thread left by #230: the Greek fixed-date grant list carried a warning that it was "incomplete by construction", implying a follow-up harvest was needed.

That warning was an expected-value argument, not a measurement. Measuring it shows the list is already complete with respect to the ten years harvested, so no further harvest is needed.

Coverage

	of 366 calendar dates
probed two or more times	329
probed once	12
never probed	25

The measurement has a trap, which I hit first

Asking "where does the app say strict?" is circular — a granted date no longer says strict, so every grant looks like a gap. Done that way it reported 113 blind dates, most of which were the grants themselves.

The right test asks the season, not the output: a grant leaves a trace only where the season's own floor is strict and its cap would let a claim reach wine and oil. That's what grant_coverage.py does, and the docstring says why.

Both structural blind spots checked, neither is a gap

• Dec 25 – Jan 4 and Jan 6 are fast-free, so no grant can exist there.
• Aug 1–14, the Dormition fast, caps claims below rank 7 to strict, so a grant would be invisible. Rather than leave that an unknown, it was checked against goarch.org directly: across ten years its Dormition weekdays are strict 93 times and fish 7 — and all seven are Aug 6, the Transfiguration, which the shared data already grants.

Of the 12 thinly-probed dates, two showed wine and oil on their single probe — Feb 17 and Jun 3 — and both are already right: Feb 17 through its override, Jun 3 from the Paschal season's own allowance rather than from any grant at all.

What's in the PR

• tools/fasting/grant_coverage.py — so the question can be re-asked after any future harvest instead of re-deriving the argument
• docs/fasting-refactor-scope.md — the corrected claim, with the measurement

More years would only buy probes for those 12 thin dates; the harvest route stays documented in tools/fasting/goarch_audit.py.

Documentation and one tool. No application code or data changed. 197 tests pass.

🤖 Generated with Claude Code

https://claude.ai/code/session_019nQRjJpPqW89xqzAf1vsEX