deploy.sh records the workout app's git tree after each phone install
and skips the phone when the tree is the same, so a PC-only change
deploys with the phone unplugged. No record means the phone is needed.
Todo: 2556952d
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
-n auto (16 workers) peaked at 3.2 GiB and the kernel OOM-killed the
ci-pytest gate step in capped.sh's 4 GiB slice. 4 workers fit in 2 GiB
and finish in 19 s instead of 58 s under the CPU cap.
Todo: 2556952d
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
kuhy, 2026-11-18: shutdown 18:00, a run ending 19:58 must give 20:00
and the book's +1h 21:00 -- what the daily reset derives. The
late-credit rule (credit's own time + reward, 2026-10-17) made it
21:58 and 22:58. Unpaid gates now hold the PC behind the evening
chain until 23:00 instead, so the rule and its credit-time plumbing
go: _late_credit, the at= anchors, _done_at. The preview flags any
live-vs-derived MISMATCH again.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWWLoUxxSPf1m8KUVo8
Every locker screen, the status window and the credit toast now show
glyphs (screen_locker/_glyphs.py) and numbers, coloured by state.
Sentences that carried a reason (verifier message, sync diagnosis,
sick refusal, form rejection) go to the log at warning instead.
New pre-commit hook text-free runs scripts/verify_text_free.py.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Paints every locker screen under Xvfb and fails on any painted word.
Red now: 18 screens. --shots DIR saves one PNG per screen.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
textgate scan 319 -> 0: docs/explanations (dropped explain-diff
output, not published, nothing links it), Tk window titles, the
MaterialApp title, '+kg' -> '+' plus the weight icon, and
spoken-only Semantics labels hoisted so the gate sees them.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Painted chrome becomes icons or symbols; old words stay as
spoken labels or code comments. OS window titles kept. Tests
build today from the local day.
Todo: ed07cff3
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Each slice runs under 30 s; together they run all 35 default-stage
hooks once. CI (pre-commit.yml) is unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWZ9CKZsNxtD5Jr9bYD
Real fixes: split over-long functions (lock-decision chain, manual
sync, validation, phone verification, status), hoist a dead lazy
import, context-manage Xvfb and temp dirs, add encodings, share the
silent-failure argv loop.
lock_grab_handoff read lock._recovery, which gatelock's LockWindow
does not have (its public property is recovery); a contract test now
checks every attribute the handoff reads against the real class.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWYdwRAtMg7ki6TKxuU
The by-minutes bonus path clamped at 24:00 (_MIDNIGHT) and relied on
adjust_shutdown_schedule.sh's RESTORE_CEILING (23:00). It now clamps at
earned_time.shutdown_ceiling_for(day) via the new _ceiling.ceiling, so a
moved WAKE_MINUTES cannot be passed by a bonus. The shell's 1380 stays
the hard upper bound; ceiling() is min()'d with it, and a test pins the
Python constant to the script's value. _preview_sim and the invariant
test reuse that constant instead of a second 23 * 60.
Failure mode: if the earned_time lookup raises (ArithmeticError,
LookupError, TypeError, ValueError), ceiling() falls back to 23:00
loudly (logger.warning), never silently and never above the helper's
bound. That is fail-open relative to a lowered ceiling, bounded by the
shell clamp; an earned_time without the API (0.3.0) uses its constant.
Red first: new tests (21:00 ceiling, failing lookup, caps-at-ceiling)
failed on c3a5d14, pass now. 14-day replay of the real ledgers
(2026-10-18..31) through old and new code, plus an all-done-at-22:30
counterfactual: identical, every day 23:00; live files unchanged.
TODO-cap-bonus-at-day-ceiling.md removed (done).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWYdwRAtMg7ki6TKxuU
The whole-suite check (~96 s capped) outlived the 60 s action cap. Manual-stage ids flutter-test-coverage-shard-0..3 each record their bucket's lcov under the HEAD tree hash; flutter-test-coverage-merge fails unless all four exist for the same tree, then applies the unchanged 100% bar and lib-file guard to the union. CI (--shard/--merged) is unchanged.
Claude-Session: https://claude.ai/code/session_01WzxdWYmRsPBNf6gi2LVjFA
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- addopts: -n 4 --dist worksteal (fixed, not auto: 24 workers would break
capped.sh's 4 GiB). CI's own -n auto still wins.
- coverage core = sysmon (PEP 669): ~40% less tracing overhead, same data.
- shutdown ceiling invariant: one plain-function patch of earned_units per
day instead of a MagicMock per unit (~5000 per case); every day and every
case still checked.
- poke/web server tests: serve_forever(poll_interval=0.01), so shutdown()
no longer waits 0.5 s per test.
- autouse _no_real_ntp: check_clock_skew sent real UDP to pool.ntp.org from
every verification test, and a lost reply cost a random test 5 s.
- temperature hard-timeout test: pins the default to 5 s and times the
mechanism at 0.1 s instead of waiting 5 real seconds.
- _xvfb_reexec: Xvfb -displayfd instead of xvfb-run's fixed 3 s sleep.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The single ci-mirror manual hook took ~100 s (pre-commit --all-files plus a
serial clean-venv pytest, ~85 s), past the 60 s per-action cap. ci_mirror.sh
now takes a subcommand (venv|precommit|pytest|all; default all = the old
one-shot behaviour) and each step is its own manual hook, ordered so the
venv rebuild runs first in its own call. The clean-venv pytest runs
-n auto --dist worksteal; branch coverage stays at 100% (pytest-cov merges
the workers).
Measured (timeout 58): ci-venv 27.0 s cold / 0.1 s warm, ci-precommit
32.6 s, ci-pytest 12.6 s, flutter-test-coverage 46.5 s, web-coverage 1.7 s.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PAhbNszs5JKdtnD4uBi3xp
Next to the one-tap "I'm sick" (unchanged: keeps nothing, unlocks), a small
secondary link lays sickdays 0.3.0's embeddable KeepPanel over every
surface, in-process so it gets input under the lock's grab. Closing the
overlay reveals the untouched screen underneath, so no countdown or phone
check is restarted or skipped.
After the panel's own write + sync: sick with the workout lock standing
down -> sick floor, decision, close (shared with the one-tap path, no second
mark or push); sick with the workout kept -> sick floor and a note, lock
stays; cleared / unchanged / cancel -> the panel just closes.
Tests stub KeepPanel and redirect resolve_paths and sync_quietly in every
sickdays.* module, sickdays.gui._panel included. The fit check measures the
panel on a throwaway pool. Pins bumped to sickdays-v0.3.0.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PAhbNszs5JKdtnD4uBi3xp
The workout lock, its explanation, the status view and the workday penalty's
penalised day now ask _sick.stands_down_workout: a sick day that keeps the
workout locker behaves as a normal day. The penalty's ring day is judged by
the alarm switch. Shutdown (20:00 floor, grace floor, base, target, preview
scripts) keeps is_sick: the floor belongs to the day, not to a locker. The
one-tap button still keeps nothing. Test pool fixture and preview harness now
redirect every sickdays resolve_paths binding, so keep edits never reach the
real pool. Pins bumped to sickdays-v0.2.0.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PAhbNszs5JKdtnD4uBi3xp
A sick day is now a date in the shared sickdays pool (utils sickdays 0.1.0)
and nothing else. Removed: the 7/30/90-day budget, the escalating lockout
countdown, the justification form, the commitment prompt and its double
debt, the workout-debt counter, the -60 min shutdown cut and its
sick_day_state.json, the post-sick-day --verify-workout mode, the sick
auto-upgrade, scripts/revoke_sick_day.py and their tests.
- Lock screen: "I'm sick" on the retry, no-phone and both Tue-Thu screens;
one tap -> sickdays.add_sick_day(today) -> sick floor -> best-effort
background sickdays.sync_quietly -> unlock. Refusals (free day,
unreadable pool) are shown and the lock stays.
- Shutdown: the reset resolves with earned_time.resolve(..., sick=True)
(max(20:00, ladder), capped 23:00) and saves the sick lift with the grace
lift, so later credits absorb it instead of stacking on 20:00. A day
marked after the reset is lifted by the new _sick_floor live pass (every
5-min tick, --sync-only, --apply-bonuses, and the tap itself).
- Workout lock and grace floor stand down on a sick day; workouts that day
credit normally through the sync paths.
- wake-alarm workday_penalty.json is void when its issued_date or
penalty_date is a sick day.
- Status window, i3 line, CLI --status, MCP flags and web reasons read
sickdays.
- scripts/migrate_sick_history.py imports sick_history.json's dates via
sickdays.import_legacy, verifies and removes the file; idempotent, run by
install.sh.
- Pins: earned-time 0.11.0, freedays 0.4.0, sickdays 0.1.0 (tags pending);
pre-commit 4.7.0, ruff 0.17.1 (freshness gate).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Xfj1sxKH9VBPqpcAni8Jry
A 5+ workout week no longer banks (count - 4)h, and the every-4-week streak
milestone +1h is gone. The streak counter and the 09:00 early-bird extension
stay. The applying path is untouched, so 2026-W41's already-banked 2h keeps
applying through Sunday 2026-10-11; old weekly_shutdown_bonus_hours keys load
and are preserved as-is. The next-week status preview no longer promises
bonus hours.
Claude-Session: https://claude.ai/code/session_01WfVGhY8XcsQsC7qz9UU8xb
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The PC push aborted before it ever tried Firebase whenever the GitHub PAT
was missing ("Workouts NOT synced: no sync token"), so once the PAT was lost
this machine published nothing although Firebase was configured and live.
The phone has not written the GitHub repo since 2026-08-21.
- sync_client() / sync_client_or_reason(): firebase_client_for("screen_locker")
only, proven with can_access_remote(), healed once from a sibling app's
session, else a warning naming what will not sync (push AND pull) plus a
recorded degraded source. No config -> warning, no fallback.
- _manual_push and scripts/tombstone_sync_record.py take their client from
sync_client_or_reason(); the push result carries the reason.
- Deleted: read_sync_token, remote_client, the mirror client, _push_outcome
(the half-landed-push case no longer exists), SYNC_TOKEN_FILE and
SYNC_REPO_OWNER/NAME, the sync_token test fixture. Status backend is
"firebase" or "none".
- crdt-sync pinned to v0.9.2 in pyproject, requirements.txt and
install_systemd.sh (the installer still said v0.9.0).
kuhyx/syncs/screen-locker-sync is left untouched as an archive; no code
reads it.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018RHwVPyiiQMPh4M3JQwQku
- A day's RunnerUp walks (_Walking.tcx) count once their moving time sums to 40 min
- Rest days: declared ahead, 2 per ISO week, HMAC-signed; pay the first workout unit
- Grace floor: shutdown >= first task + 60 min (capped); skipped on a sick day
- Workout ledger: one signed credit row per log credit slot and rest day
- Day-aware ladder amounts from earned_time 0.4.0; preview_shutdown dry-run script
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QrjUTo1yhgAsUUasdw4bpG
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Tests (~214 s, gated by flutter-coverage):
- Flutter suite runs as 8 parallel buckets of whole test files
(check_flutter_coverage.sh --shard I/N); flutter-coverage merges their
lcov files (--merged N) and applies the unchanged 100% bar and
every-lib-file-present guard. `flutter test --total-shards` splits test
cases inside each file, so every shard would still compile all 77 files.
Proved locally: the merged lcov matches an unsharded run line for line
(3312/3312), and a missing shard fails. flutter-action caches the SDK.
- pytest: uv install, -n auto -q -p no:sugar (local: 25.9 s -> 10.6 s,
100% coverage both ways).
- lock-popup-safety: uv, and no apt step -- xvfb-run and setup-python's
tkinter are already on the runner (screens-fit relies on both already).
pre-commit (~122 s): matrix of mypy / pylint / flutter-analyze / rest, each
installing only its own toolchain, with a per-group hook cache.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0188kZdx3bb56QM8JFUnbHuM
The between-set break never finished once the phone's screen was off. Three
independent causes (worked through on-device on 2026-09-12; left
uncommitted awaiting the go-ahead, committed here as-is plus one pin bump):
1. The countdown was not wall-clock: `_breakRemaining--` on a
Timer.periodic, which Android throttles and Doze suspends in a
backgrounded isolate, so the counter stalled and _onBreakFinished()
never fired. Now a BreakClock deadline on the wall clock.
2. An expired break was silently dropped on restore (`if (remaining > 0)`
only) — now cued and logged.
3. No notification infrastructure at all: no package, no
POST_NOTIFICATIONS / FOREGROUND_SERVICE / WAKE_LOCK.
Now: flutter_foreground_task runs the countdown in a service isolate that
survives BACK and the screen lock; flutter_local_notifications shows an
ongoing "Workout A · break m:ss / Next: …" notification with −1 rep / Done
actions and a "Break over" alert on its own channel (raw break_end sound,
vibration). Button presses go through a persisted BreakIntent queue the
screen drains on resume, so a press with the app killed is not lost.
Finish and Reset stop the service; a startup reaper handles the crash case.
Verified on the phone with the screen off and locked: service alive and
ticking while Dozing, both notifications on the lock screen, channel
importance 5 with the break_end sound. Not verifiable from adb (receiver is
RECEIVER_NOT_EXPORTED): an actual action-button tap, and a true recents
swipe — see stronglift_replacement/TODO-workout-app-v2.md.
Also fixed en route: the break-end sound had never played in any widget
test (AssetSource needs path_provider, which nothing faked, and the failure
was swallowed by catchError); the queue's prune wrote back a stale list.
This commit additionally:
- takes the freshness gate's bump flutter_local_notifications 22.3.1;
- restores web/package.json to HEAD (the tree held a stale pre-bump copy);
- gives scripts/check_flutter_coverage.sh a FLUTTER_TEST_CONCURRENCY knob:
the default one-isolate-per-core run peaks over 4 GiB and is OOM-killed
under the resource cap. Default behaviour unchanged.
Gate: flutter analyze clean, 511 tests, 3187/3187 lines (100%), pre-commit
green. Not re-installed on the phone for the 22.3.1 pin: no device attached
at commit time.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01C7vgacLhb4WJjxuVj2QLEP
manual:2026-09-13T14:30 -- the synced copy of the football that this
morning's day-key rekey let back in under a second id -- made every
workout-sync tick log "Manual-workout budget exhausted -- not ingested".
Tombstoning it once did not stick: sync_log pulls only the OTHER devices'
logs and _manual_push re-derives the PC's log from log.json each tick, so
the next push overwrote the tombstone ("tombstoned", then "LIVE" five
seconds later). Deletion is monotonic only for records still in the merge.
- New _sync_tombstones: {id: {deleted_hlc, reason, tombstoned_at}} in
screen_locker/sync_tombstones.json (gitignored with its siblings). The
HLC is minted once and stored so re-pushing does not churn the store.
The ledger path is resolved at call time, not as a default argument, so
the test suite's redirect actually applies -- verified by hashing the
real ledger before and after the run.
- _manual_push.records_from_workout_log merges the ledger over the live
derivation, so every push carries the tombstones.
- scripts/tombstone_sync_record.py <id> [--apply --reason ...]: dry run
reports live/tombstoned; --apply writes the ledger, runs one tick and
reports what the store holds afterwards. Rerunnable.
- tests/_isolated_state.py redirects SYNC_TOMBSTONES_FILE (both bindings).
Verified: --apply -> "tombstoned"; a second ordinary tick -> "already
tombstoned" with the same HLC; workout-sync.service now logs "Sync record
manual:2026-09-13T14:30 is tombstoned -- skipping it" and the budget line
for that id is gone. log.json untouched (the local entry stays; it shares
the original's credit per duplicate_credits.json).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01C7vgacLhb4WJjxuVj2QLEP
The i3bar showed 21:00 on 2026-09-13 with a counted workout on disk. The
football was logged at 00:02 and credited against the previous day's
config (already at the 23 ceiling, so it bought nothing); the 09:15 base
reset then wrote 21/21 and only re-layered the weekly bank, never asking
what today's log had already earned. Live credits are read-add-write, so
every workout credited between local midnight and the first locker tick
of the day lost its hours to the reset that followed.
- _shutdown_base.reset_to_base_if_new_day takes the log and writes
min(23, base + hours today's entries earned) in one write, using the
same credit rule as the weekly count. Idempotent: still once per day.
- _workout_credit names FIRST_WORKOUT_BONUS_HOURS=2 /
EXTRA_WORKOUT_BONUS_HOURS=1 and derives earned_shutdown_bonus_hours()
from them; _shutdown and the +1h path use the constants instead of
literals so the replay and the live paths cannot drift.
- _weekly_check.credit_key: a synced copy whose sync_record_id names an
entry already filed on the same day shares that entry's slot. This
morning's rekey (967e3cd) let the sync pull the football back in as
manual:2026-09-13T14:30, which counted as a second credit (weekly 4/5
instead of 3/5) and would have made the replay ask for 24.
- annotate_duplicate_credits.py passes the day's id index so the ledger
records the manual duplicate the same way it records phone/PC pairs.
Verified through the real path: cleared last_reset_date, started
workout-locker.service -> "--restore 23 23 5", "Daily base reset:
Mon-Wed=23, Thu-Sun=23 (base 21/21 + 2h already earned today)",
shutdown_countdown.sh prints "23:00"; a second start changed nothing.
leetcode-guard grants no shutdown time anywhere (grepped screen-locker,
digital-wellbeing, i3 config) — the 23:00 comes from the workout alone.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01C7vgacLhb4WJjxuVj2QLEP
A manual workout logged at 00:02 CEST on 2026-09-13 was filed under
2026-09-12: every day-key derivation used datetime.now(tz=UTC), and at
that instant UTC was still yesterday. The 09:18 lock check looked for
2026-09-13, found nothing, and locked the PC — for a workout that was on
disk. The budget counter (a trailing window) counted it fine, which is
why the status showed 1/2 used while the locker still locked. Any log
between local midnight and 02:00 CEST hit this; it had happened before.
- New screen_locker/_day.py: today_str() is the one definition of "what
day is it" — the local calendar day the user sees on the clock, which
is also what the phone app, RunnerUp verification, the status window
and the freedays pool already used. All 14 UTC sites (log writer, lock
predicates, explain chain, early-bird, wake state, shutdown/sick state,
sick budget, manual CLI, phone verification) now go through it, writers
and readers together so no new midnight window opens between them.
- write_signed_entry dedups by workout_id across ALL days: a workout id is
globally unique, and an entry moved to its correct day must not let its
synced twin (still stamped with the old day) re-ingest as a duplicate.
- scripts/rekey_log_entry.py: move a wrongly-filed entry to its real day
(the HMAC never covered the day key, so it still verifies).
- test_local_day_key.py freezes the clock at the exact failing instant.
Replayed the real timeline (log at 00:02:35, evaluate at 09:18:53, weekly
3/5) against HEAD in a throwaway worktree vs this tree:
HEAD -> logged under 2026-09-12, LOCK
fixed -> logged under 2026-09-13, no lock (already_logged)
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FcU5dCcUNHMAMzUNChdrrJ
The grab axes are grab=GrabPolicy(kind=..., disable_vt=...,
preempt_weaker_holder=...) and every colour read goes through
config.palette; nothing about the lock's behaviour changed. Also carries
the pylint hook on 4.0.8 with the source-only scope config; the gate
stays at 8.0 until the remaining findings are cleared.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HHCbkJWXHjtTUaqvnHRAqA