[futureweb]

Open follow-ups from the reviews of #36 and #39, wanted before release 1.0.0b2. No new tools, no
parameter or catalogue changes, no extra API requests.

1. Lap-press steps in the execution analysis (R36-11)

The duration or distance of an until_lap_press step is only a planning estimate, but
analyze_workout_execution treated it as the planned length. Lap-press steps are now open:

• "ended by lap after (planning estimate )" instead of "-59 s vs plan",
  "shorter than planned" or a cap at the estimate with "extra time inside the plan";
  "ended with the recording" when the last step runs to the end of the activity.
• Target adherence, time in target and the other metrics are evaluated as before.
• The plan clock restarts at the actual lap, so the timed steps after it are judged from there (no
  "started 5:00 earlier than the plan timeline" after a short warm-up).
• Alignment: any length costs little; ties go to the length nearer the estimate. With device
  auto-laps the step ends on the lap closed by the press (a partial lap), never on a full auto-lap.
  Its end boundary is never moved, split or extended by trailing laps. Its estimate no longer makes
  auto-laps of the same length look like lap presses. Carving a lap-less step uses the plan bounds,
  so it is skipped for a lap-press step.
• JSON: row lap_press (ended_by, elapsed_s, moving_time_s, estimate_s/estimate_m,
  actual_m), planned.lap_press, summary lap_press_steps/lap_press_estimate_s. The plan total
  is marked "≈" with the estimate share.
• get_activity_report key finding names the lap-press steps. get_activity_intervals(include_planned_types=true)
  shows "until lap press, ended by lap after …" and never beyond_plan_s for them.

Code paths checked with synthetic tests (tests/test_execution_lap_press.py):

• exact laps: warm-up 12:00 / 16:01 / 25:00 against a 17:00 estimate;
• device auto-laps (1 km, counter reset at the press): warm-up 7:00 / 10:00 / 20:00 against 10:00;
• carried-over boundary: a timed warm-up ending inside an auto-lap, then a lap-press step;
• lap-press recoveries inside repeats, with step laps and with auto-laps;
• no-streams interval mapping, the auto-lap detection, a last step running to the end of the recording;
• a fuzz test with 30 random plans × 2 lap layouts. Lap-press steps last 0.3-3× their estimate. Every
  timed step stays exact, has no start shift and has no length or timeline note.

The warm-up and auto-lap cases also fail on main, so the tests cover the change.

Live, read-only: activity i196751426, paired with event 144186380 (lap-press warm-up with a 17:00
estimate, 4×(3+3 min), 3 min, 10 min).

Before (main):

Plan: 11 steps, 54:00 planned | Actual: 12 intervals, 58:40 in total | matched 11, …
[plan 1] warmup 17:00 @ 110-175 W 'Press lap Einrollen bis Startpunkt' -> Press lap Einrollen bis Startpunkt (WORK) 16:01 from 0:00, -59 s vs plan, in range (110% of target)
compact:  [plan 1] warmup 17:00 @ 110-175 W: 16:01, 157 W, HR 126/143, 73% in target, in range (110% of target)
report:   - Plan: 11/11 steps executed; work steps 301/288/292/289/278 W vs 275-285 W planned (…); steps with clear deviations: 1
intervals:  [1] Intervals.icu WORK | plan step 1 warmup 17:00
JSON: rows[1..10].start_offset_s = -59.0 (every step after the warm-up "earlier" than the plan clock)

After:

Plan: 11 steps, ≈54:00 (incl. 17:00 planning estimate of 1 lap-press step) planned | Actual: 12 intervals, 58:40 in total | matched 11, …
[plan 1] warmup until lap press @ 110-175 W 'Press lap Einrollen bis Startpunkt' -> Press lap Einrollen bis Startpunkt (WORK) from 0:00, ended by lap after 16:01 (planning estimate 17:00), in range (110% of target)
compact:  [plan 1] warmup until lap press @ 110-175 W: ended by lap after 16:01 (planning estimate 17:00), 157 W, HR 126/143, 73% in target, in range (110% of target)
report:   - Plan: 11/11 steps executed; work steps 301/288/292/289/278 W vs 275-285 W planned (…); steps with clear deviations: 1; lap-press steps (open length, not judged): plan 1 warmup ended by lap after 16:01 (planning estimate 17:00)
intervals:  [1] Intervals.icu WORK | plan step 1 warmup until lap press, ended by lap after 16:01 (planning estimate 17:00)
JSON: rows[1..10].start_offset_s = 0.0 (rep 4 recovery 2:57 → last interval -3.0); rows[0].lap_press = {ended_by: lap, elapsed_s: 961, estimate_s: 1020, …}

The one remaining clear deviation is rep 1 at +5.6 % above the target range, which is correct.

Offline, same plan, warm-up ended by lap after 12:00 (compact):

before: [plan 1] warmup 17:00 @ 110-175 W: 12:00, … | 5:00 shorter than planned      (+ "started 5:00 earlier than the plan timeline" on every later step in standard)
after:  [plan 1] warmup until lap press @ 110-175 W: ended by lap after 12:00 (planning estimate 17:00), …   steps with clear deviations 0

Offline, warm-up ended by lap after 25:00:

before: [plan 1] warmup 17:00 @ 110-175 W: 17:00, …; +8:00 extra time inside the plan
after:  [plan 1] warmup until lap press @ 110-175 W: ended by lap after 25:00 (planning estimate 17:00), …

2. Health markers: note only while current (R39-10)

The note from a SICK/INJURED entry's description is shown only while the entry covers the tool's
day, or ended at most 2 days before it. Later in the 14-day window the line has only category,
name and dates (JSON note null). The note limit stays at 120 characters (NOTE_CHARS, confirmed
and tested). Live, read-only, get_recovery_snapshot(date_str=…) for the entry of 2024-12-22 to
2024-12-27:

2024-12-29 (ended 2 days before), before and after:
- SICK 2024-12-22 (Sun, 662 days ago) to 2024-12-27 (Fri, 657 days ago): "Infekt ~39,2°C Fieber ...", note "27.12. - Fieberfrei"
2025-01-03 (ended 7 days before), before:
- SICK 2024-12-22 (Sun, 662 days ago) to 2024-12-27 (Fri, 657 days ago): "Infekt ~39,2°C Fieber ...", note "27.12. - Fieberfrei"
2025-01-03, after:
- SICK 2024-12-22 (Sun, 662 days ago) to 2024-12-27 (Fri, 657 days ago): "Infekt ~39,2°C Fieber ..."

3. get_event_by_id: last day of multi-day events (R39-13)

This reuses event_day_span from #39. Live, read-only, SICK event 49921901:

before: End: 2024-12-28T00:00:00 (local)                         JSON: no last_day
after:  End: 2024-12-28T00:00:00 (local; last day 2024-12-27 (Fri))   JSON: "last_day": "2024-12-27"

Single-day events (e.g. the workout 144186380, End: 2026-10-14T00:00:00 (local)) are unchanged in
text and JSON, and so are events without an end. An end after midnight (22:00 → 02:00) names the
next day.

4. Float noise in text outputs (R39-14)

format_value (the "as stored" helper behind _stored/_val) now shows floats with 4 significant
digits instead of up to 6 decimals. Whole numbers are never rounded. Known metrics get their usual
precision through a new digits argument:

Output	Precision
activity summary	power/HR 0; intensity, decoupling, load, CTL/ATL, TRIMP, L/R and wind shares 1; VI, Power:HR, EF, stride 2; speeds 3
interval and group blocks	load 1, torque 1, respiration 1, W/kg 2, variability 3
thresholds	eFTP/CP/W'/Pmax 0
wellness	CTL/ATL/ramp 1, eFTP 0, weight 2
recovery snapshot	CTL/ATL 1; 7-day and baseline mean/median/SD 1

Also covered: training-summary form and ramp, get_event_by_id other fields, report device values.
A stored non-zero is never shown as 0; it keeps one significant digit, e.g. a gradient of 0.001984 % is
shown as 0.002%. JSON values are unchanged: a structural comparison of the live JSON outputs before
and after shows only the new lap-press keys and the corrected start_offset_s.

Live before → after (i196751426 and today's snapshot):

Interval 1:  Distance: 8150.6797 meters → 8150.7 meters | Training Load: 12.480418 → 12.5 | Torque: Avg 18.373035, Min 0.540526, Max 61.539913 → Avg 18.4, Min 0.5, Max 61.5
             Decoupling: -1.340999 → -1.3 | Respiration: 28.267124 → 28.3 breaths/min | Variability: 0.039962 → 0.04
Activity:    Intensity: 91.02564 → 91 | Power:HR Ratio: 1.2857143 → 1.29 | Variability Index: 1.1833333 → 1.18 | Fitness (CTL): 67.02364 → 67 | Efficiency Factor: 1.5214286 → 1.52
Custom:      Recovery Time [RecoveryTime]: 25.666667 h → 25.67 h | VO2 Max [VO2MaxGarmin]: 50.969925 → 50.97 | EPOC 165.7373 ml/kg → 165.7 ml/kg
Snapshot:    hrv: … 7d mean 39.714286 | 42d baseline mean 38.333333, median 40, sd 5.308377 → 7d mean 39.7 | 42d baseline mean 38.3, median 40, sd 5.3
             CTL 65.44669 | ATL 67.85514 → CTL 65.4 | ATL 67.9
Event:       Other fields: icu_atl=76.81097, ss_p_max=0.47685498 → icu_atl=76.81, ss_p_max=0.4769

Floats with 4 or more decimals across about 67 live text outputs (44 read tools): 727 before, 11
after. The remaining ones are intended:

• values below 1 at 4 significant digits (Stride 0.9177, ss_p_max=0.4769, Elapsed Time 0.9775 hr);
• the threshold pace in m/s, 4 decimals as in get_sport_settings;
• a near-zero gradient kept non-zero (-0.00002%).

Review of this PR (R40) — second commit

• R40-1 (first/last lap-press step took in riding before/after the workout):
  • The press closes exactly one lap. So a lap-press step is its closing lap, preceded only by full
    device auto-laps: laps of the auto-lap length, at most 0.5 % short of the setting (a press lap
    usually is shorter).
  • Without a detected layout, a usual auto-lap distance from 200 m to 20 km counts instead.
  • The step never reaches across a press, a step lap or the lap that starts the workout.
  • With lap-press steps, the riding before or after the plan costs its fixed part once per block.
    A long ride home in many auto-laps is therefore not moved into the plan.
  • The first step takes all full auto-laps back to the workout start. A lap well off its target at
    the plan edges counts as outside the plan.
  • Reviewer cases R4/R5 and a 60 min ride home with 1 km auto-laps at three speeds are now
    correct: "Additional training after the plan: 15:00 / 60:00", "Riding before the plan: 20:00".
• R40-2 (press lap of about the auto-lap length made the DP skip a step):
  • Ending on a full auto-lap now costs 0.4, below GAP_COST.
  • It costs 1.0 only when a shorter lap that fits the target follows (the press came later).
  • R1 (975/991 m), R3 (986 m) and seeds 195/210 are matched with 0 deviations.
  • Such a row says "ended at a full device auto-lap (no shorter lap closed by a press; …)".
• Lap trigger: Intervals.icu does not expose the lap trigger. The interval objects and the activity
  have no such field. The original FIT has lap_trigger (manual/time/session_end) and
  wkt_step_index (verified read-only for i196751426). Using them would need an extra file download
  per analysis and a FIT decoder, so this is left as a possible follow-up. The rule above uses the
  lap lengths instead.
• R40-3: a long warm-up whose auto-laps are too few for a layout is matched as a whole.
• R40-6 to R40-9:
  • A non-zero value is never shown as 0, also below 1e-6.
  • Form and ramp are never -0.0.
  • Snapshot baselines use 1 decimal only for the known metrics, 4 significant digits otherwise.
  • "Work: 632 kJ" replaces "Kilojoules: 631775", which was in J.
  • With a pause, the text names the elapsed and the moving time.
  • The plan total includes distance estimates.
  • The report says "length set by the lap press, not judged; targets judged".
  • JSON note_withheld is added.
• R40-10: with very many laps, the pairwise merging never merges across a lap that may close a
  lap-press step.
• R40-4/5/10: listed as known limitations in the execution guide (no press at all, double press in
  consecutive lap-press steps or at the plan edges, workout start exactly on an auto-lap at the
  warm-up's intensity, more than about 600 laps).

Measurements (offline, the reviewer's scripts):

Check	Result
Reviewer's 400-workout lap-press fuzz (r40/lp_fuzz3.py)	bad 5 (main 184); worse than main 2, better 181
Same fuzz, riding before/after at a 3.7 % different speed	bad 4 (main 187); worse than main 0, better 185
Reviewer's 95 lap-press scenarios	worse than main only the documented R40-4/R40-5 cases
Timed-only plans (r26 scripts, 2,307 analyze calls)	JSON and text identical to main, 0 exceptions, same runtime

The 2 worse cases in the original fuzz (seeds 77 and 226) share one cause. The synthetic riding
before the workout covers an exact multiple of the auto-lap at the warm-up's intensity, so no lap
marks the workout start.

Live, read-only, i196751426: unchanged.

[plan 1] warmup until lap press @ 110-175 W: ended by lap after 16:01 (planning estimate 17:00), 157 W, HR 126/143, 73% in target, in range (110% of target)

The report finding now ends "…; lap-press steps (length set by the lap press, not judged; targets
judged): plan 1 warmup ended by lap after 16:01 (planning estimate 17:00)".

Checks

• ruff check .: all checks passed.
• mypy src tests: no issues in 130 files.
• pytest: 1154 passed (main 1129), including the catalogue budgets.
• pylint gate (--disable=C0301 --fail-under=0 --fail-on=E,F): exit 0, and no new messages compared
  with main.
• Live checks were read-only. Every non-GET request was refused at the httpx transport (sync and
  async), and all requests were GET. Request counts per tool are unchanged.
• Nothing was written to Intervals.icu. The production checkout and the services were not touched.

🤖 Generated with Claude Code