The day's last verified progress row for the source wins, and both done
and target come from that one row. max(done) with max(target) could pair
a stale target with a new done when a gate re-derives its target within
a day (book-guard does). The rule reads neither created_at nor device,
which the workout ledger does not write.
A later row that fails verification, has the wrong kind, source or day,
or has unreadable numbers is skipped; no good row is still None. The
public API, the version and every other behaviour are unchanged.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWZLyjnLjB1THD6ckou
A gate can now sign "done N of target M for day D" and resolve() pays a
no-answer that share of a yes (60 * 1 // 3 = 20), floored, for shutdown,
gaming and sick days. A yes still pays full; "could not check" still pays
nothing. With no progress row anywhere every output is 0.23.0's, pinned by
a digest grid and, when installed, the real 0.23.0 as oracle.
Every failure reads as no progress, so the day pays by its yes/no rule and
never more:
- no row, a bad or missing signature, an edited row
- another day or another gate's source
- numbers typed as ints, floats, signed, padded, empty, non-ASCII or huge
- a zero target (no plan, no share)
- an unreadable key or ledger (logged, never raised)
- the gate being off today
- an earner with no progress source (tutor, focus)
A bad row beside a good one is skipped. Done above target is capped.
The focus earner now uses the same share arithmetic, same floor. A progress
row is not a credit: maturity and the done answers never see it. The floor
helpers moved to _floor.py and the tutor row contract to DOCS-tutor.md to
stay under the 250-line cap.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWVfmEXxPymCs5bo3Zs