Summary: four phases (Stabilize, Build, Durability, Taper) with an absorb week every fourth week; long run capped at 28 km or 3:15–3:30 on feet, whichever comes first; a midweek medium-long run from Week 7 so the long run never exceeds ~55% of the week; quality as the green-week default (strides, steady blocks, one short threshold touch, marathon-pace blocks from Week 12); a half-marathon tune-up in Week 11; a formal go/no-go in Week 13; and a dress rehearsal in Week 14. Gut training progresses to ≥60 g carbohydrate/hour by Weeks 12–14, with a 10–12 g/kg/day carb load in race week.

Change reason: two structured critiques of v2 — an exercise-physiology pass and a coaching pass — converged on the same structural flaw: a 32 km long run inside a 42 km week put two-thirds of the weekly load in one session. v3 fixes the ratio from both ends and commits to the quality work v2 left “optional”. Full reasoning in the change history.

Week by week

WeekPhaseStartsTargetLong runMidweek MLRNotes
W1Stabilize (retro)10 Aug31 km (as run)12.4 km (as run)—Adopted retroactively: run before the v3 document existed. Sessions recorded as run, not as planned.
W2Stabilize17 Aug35 km15 km (cap 1:50)—Baselines; capture reliability; zone anchors.
W3Stabilize24 Aug36 km16 km (cap 2:00)—Fuel practice on runs >75 min (30–40 g/h).
W4Duration31 Aug38 km18 km (cap 2:10)—Long run is sole progression. ★ Farthest-ever long run.
W5Absorb7 Sep32 km14 km (cap 1:40)—Review trends and zone anchors. Never make up missed distance.
W6Build14 Sep39 km19 km (cap 2:20)—Fuel/fluid/equipment rehearsal; sweat-rate test. ★ Farthest-ever long run.
W7Build21 Sep43 km20 km (cap 2:25)10 kmMidweek medium-long run begins — spreads load. ★ Farthest-ever long run.
W8Build28 Sep45 km22 km (cap 2:40)11 kmLog GI, skin, foot, late mechanics. Fuel ≥40 g/h on long run. ★ Farthest-ever long run.
W9Absorb5 Oct37 km16 km (cap 2:00)10 kmReview aerobic and durability trends; confirm HM logistics.
W10Durability12 Oct47 km24 km (cap 2:50)12 kmFuel every long run ≥50 g/h; settle race gear. ★ Farthest-ever long run.
W11Tune-up19 Oct45 km21.1 km12 kmHalf-marathon tune-up at controlled effort. Race-morning rehearsal; result sets the marathon pacing plan. ★ Half-marathon tune-up race.
W12Durability26 Oct51 km26 km (cap 3:05)13 km2×3 km at marathon effort inside the long run; fuel ≥60 g/h. ★ Farthest-ever long run.
W13Absorb2 Nov41 km19 km (cap 2:25)10 kmFix race-system failures. Go/no-go: choose time goal / completion / run-walk from evidence. ★ Week 13 go/no-go decision.
W14Peak9 Nov54 km28 km (cap 3:25)14 kmDress rehearsal: race breakfast, kit, anti-chafe, fuel ≥60 g/h, race start time. 4–6 km continuous at marathon effort inside the long run. Finish with reserves. ★ Dress rehearsal — the confidence anchor.
W15Consolidate16 Nov45 km23 km (cap 2:50)12 kmConfirm race plan; strength maintenance; nothing new.
W16Taper23 Nov34 km16 km (cap 2:00)9 kmVolume down, intensity maintained in small doses; protect sleep.
W17Race week30 Nov16 km——Freshness and logistics. Carb load Wed–Thu 10–12 g/kg/day. Race Fri 4 Dec. ★ Race day.

Week 17 target excludes the marathon itself. Week 1 is retroactive — recorded as run, not as planned. Long-run distances are envelopes with a hard time cap — the cap wins, and any shortfall is recorded, never made up.

The adaptive rules, in brief

Every day is classified green, amber, or red, and the status prescribes one action:

  • Green — execute as planned, including scheduled quality. Skipping quality on a green day is a recorded deviation.
  • Amber (life) — two poor nights, a corroborated HRV/resting-HR trend (7-day rolling average off the personal baseline for 3+ consecutive days plus a symptom), heat, travel, stress: run easy, shorten ~20%, no quality. That is the whole action.
  • Amber (tissue) — pain ~3/10 or rising, unusual focal stiffness, next-morning function not back within 48 h: reduce the provoking load 20–40% and recheck next morning.
  • Red — symptoms only (chest pain, fainting, gait-changing pain, dark urine, heat illness…): stop. No wearable score can clear a red, and no wearable score can create one.

Wearable-only deviations without symptoms default to an easy run, not a reduction. Three amber days in seven forces a structured review. A minimum viable peak — at least one fueled 26 km+ long run by end of Week 14 — guards against the quieter failure mode: arriving at race day under-trained because the controller kept saying “careful”.

Missed training has written rules too: 1–3 days — rejoin, never make up distance; 4–7 days — repeat the prior week at ~90%; 2+ weeks — drop back a phase and formally revisit the race goal.

The full framework — metric hierarchy, capture schema, fueling and heat protocols, clinician boundaries, and references — lives in the plan document Adaptive_Marathon_Plan_v3_17_Week.md in the project repository; the operational data (day-level sessions, gates, weekly decisions) lives in the project database and drives the dashboard.