Three AI reviewers, one doctor

The reviewers are AI personas, not clinicians, and this entry is not medical advice. Each persona read the full plan, my daily notes from July 26 to August 19, and the adaptive gate rules. The one recommendation all three made unprompted: get the in-person assessment the plan already schedules. Dr. Varun Reddy's review arrived the same day — it closes this entry.

Two data corrections came first

Before the reviews could mean anything, the data had to be honest. Two updates landed the same day. My Garmin history now goes back to mid-June, and every run has time-in-heart-rate-zone data — now charted on the dashboard. The zone boundaries are calibrated against an earlier VO₂ max test, so they're real. What they show: every run before August 16 spent 78–99% of its time above zone 2, while my notes confidently said "zone 2 most of it". My easy runs were not easy.

What all three agreed on

• The plan design is sound, but its starting assumption is wrong. The plan assumes I already tolerate about 35 km per week. The history doesn't show it.
• The heel is the controlling risk. Morning pain that fades as you warm up is the classic early pattern of plantar fascia or Achilles injury. The calf pain moving toward the Achilles belongs to the same loading chain.
• The August 19 run broke the plan's own rule. Pain at 3/10 and rising at the start of a run is a tissue amber. I ran anyway.
• Book the in-person assessment. Stop strides. Start the strength work the plan promises — my notes show stretching, never strength.

The 35 km assumption, against the data

One run a week, usually under an hour, at tempo effort or harder, with two multi-week gaps. Then Week 1 tripled the volume in one step. The heel, knee, and calf symptoms all appeared within days. The open question is no longer how fast to ramp toward 35 km — the ramp already happened. The question is how much of 30 km per week the heel can keep.

Where the gates failed

The sharpest finding wasn't about training at all. The adaptive gate system is well designed, and on August 19 it was describing an athlete who doesn't exist. The thresholds are fine; the data never reached them.

1. The gates ran without symptom data. My daily notes went into a personal notebook the gate engine can't read. Until the notes were imported on August 19, the engine saw no symptoms at all — the dashboard showed green for August 15, a day I skipped a run with a headache after too much wine.
2. So the circuit breaker couldn't fire. Four active issues produced one recorded amber all week. The mechanism built for exactly this situation was structurally unable to see it.
3. The rules can't see the heel's pattern. The gates check in-run pain, next-morning stiffness, and gait. Morning pain that disappears mid-run trips none of them — and that's the signature of early tendon injury.
4. Nothing checks intensity. The 80/20 split was enforced by nothing but my own perception, which the zone data shows was off by a full zone.
5. The plan and its automation disagree on whether the minimum viable long run is due by week 13 or week 14. And this dashboard said "no injuries" while the issue register held four.
6. Self-reports drift. My July 26 note says 12.5 km; Garmin recorded 14.2 km. A 13% error in the number I was most sure of says how far to trust my pain scores.

Where the AIs disagreed

How serious is this? The physiologist wanted the formal circuit-breaker review now. The coach wanted three runs a week at 60–70% volume until I'm examined. The podiatrist said nothing is race-threatening — one quiet week, then progress after two pain-free mornings. The split depends on whether four symptoms are four problems or one, and nobody can settle that without examining the leg.

Is the 26 km minimum long run still achievable? The coach said no — at my pace it takes 3:20 against the plan's own 3:15 cap; go time-based and rehearse run-walk. The physiologist said yes — a VO₂ max around 49 implies real headroom once volume is consistent. The zone data feeds both sides: if everything I ran was tempo, my aerobic base is thinner than the number suggests, but genuinely easy volume is also a stimulus I've never actually applied. November pace can't be known in August. The Week 11 half marathon decides, with the criteria written down in advance.

What failed on August 19 — the rule or me? The podiatrist blamed rule coverage; the coach blamed my self-negotiation. One incident, two readings, both plausibly true. Both fixes are cheap, so both go in.

Does the knee count toward escalation? The podiatrist pattern-matched it to a benign kneecap issue: monitor only. The physiologist counts any new symptom during a load spike. That's really a policy question — are the counters for sorting dangerous from benign, or for catching anything new?

Then Dr. Reddy weighed in

Dr. Varun Reddy — the project's sports-medicine collaborator — reviewed my training history and current state the same day. He reviewed independently and didn't read the AI review, so the comparison in the next section is mine, not his. His comments, paraphrased:

• The ramp was the mistake. Two weeks of no running, then about 40 km in seven days — not optimal even at slow paces. Fall back to 20–30 km per week. (He's right about the number, too: the rolling seven days ending August 19 total about 43 km, worse than the 30.8 km calendar week.)
• Then rebuild at 15% per week, from 20 km to a peak of 50 km per week — about 20, 23, 26, 30, 35, 40, 46, 50 — reaching the peak around mid-October if nothing interrupts. The plan's written 54 km peak is not a target.
• Biomechanics first. The risk of biomechanical injury is much higher than the risk of cardiovascular injury. The engine isn't the constraint; speed can be adjusted later. Don't worry about it now.
• Strength, second priority: hip and core training, dedicated calf and knee work. Ignore upper body. On calf work, in his words: “As for calf loading the drop heel raises are specifically a loading stretch.”
• Pace: the zone 2 runs are too fast. Be prepared for boredom.
• Book a podiatrist assessment.
• Stretch every day: towel stretch, toe extension, calf stretch on a step, calf stretch against a wall.
• Run only on flat surfaces.

One override, documented: I don't like tracks or treadmills, and I prefer park connectors. The compromise is that I walk the slopes instead of running them.

Dr. Reddy against the AI panel

Where they agree — and it's most of it: the ramp was too sharp, tissue risk outranks cardiovascular risk, the zone 2 runs are too fast, flat routes, see a podiatrist. His 20–30 km band lands inside the AI range (the coach's cut works out to about 18–22 km; the AI podiatrist's hold-at-Week-1 is about 31 km). He also rejected the same comfort I'd been taking: slow doesn't make volume safe.

What he added that all three AIs missed:

• Hip and core strength. The AIs prescribed calf and foot work only. Hip strength is standard care for exactly the knee pattern in my notes, and none of the three personas mentioned it.
• Dedicated knee work, and explicit de-prioritization — ignore upper body.
• A concrete daily stretching protocol, the classic plantar fascia and Achilles program, matched to the heel's morning-pain pattern.
• "Be prepared for boredom." None of the AIs addressed the adherence psychology of genuinely easy running — the thing most likely to break the pace discipline in practice.

Where the emphasis differs. The AI panel treated my intensity problem as a first-order failure; Dr. Reddy agrees the runs are too fast but frames speed as a later concern — the cardiovascular system isn't at risk, so pace matters mainly through impact load. The AI podiatrist pushed loading over stretching — but the protocol he prescribes turns out to include loading: drop heel raises are “specifically a loading stretch”. That narrows the gap and sharpens one real conflict, because the AI podiatrist advised against drop (deficit) work while the heel is reactive. The clearest single question for the in-person visit. Set against the panel's severity dispute, his position lands in the middle: no alarm, no all-clear, a concrete volume band. His rebuild is also steeper than the AI trajectory — the AI reviewers held volume and slid the build later, which lowers the achievable peak, while he prescribes 15% per week to a 50 km peak, just under the plan's written 54 km and above the traditional 10% guideline. And his review doesn't address the race-goal arithmetic — whether 26 km by Week 14 survives stays open until the Week 11 tune-up.

What changes now

1. Week 2 is formally a reduce week: 20–30 km, recorded in the decision log.
2. Volume then rebuilds at 15% per week from 20 km toward a 50 km peak held in Weeks 9–10, then descends to race day. This is now plan v4, recorded in the change history.
3. Podiatrist assessment to be booked this week.
4. Daily stretching protocol starts now; hip, core, calf, and knee strength enter the schedule. No strides. Slopes are walked, not run.
5. Two gate rules get added: repeated same-site morning pain is a tissue amber, and a pre-run amber can't be overridden at the door.
6. The week 13/14 inconsistency and this dashboard's "no injuries" text get fixed.