Training Feedback Loops in 3 Signals: Reps, Recovery, and Adjustment

Training Feedback Loops in 3 Signals: Reps, Recovery, and Adjustment

AI coaching is only useful when it improves the next decision; the best loop is still the one that changes training based on what the athlete actually did, not what the dashboard hoped happened.

Justin Harris’s coaching notes on cardio show one clean mechanism: feedback-driven load adjustment. In the Skip Hill exchange, he says he’d rather not push harder much beyond 16 weeks of hard prep, and in the same thread he ties readiness to the athlete’s current body fat and dose context instead of a fixed calendar rule. That is the point: if AI coaching cannot convert real training output into the next training choice, it is decoration, not coaching.

The strongest version of the thesis is simple and falsifiable: AI helps coaching only when it shortens the distance between performance signal and program change; otherwise it mostly adds dashboards, summaries, and false certainty. Coaches don’t need more data in the abstract. They need tighter loops around what happened, what it means, and what to do next.

The loop starts with the signal, not the software

In real coaching, the signal is rarely “the athlete is good” or “the athlete is bad.” It is usually small, specific, and actionable: down 2.1 lb this week, a little more depleted than last week, stubborn lower back fat, sleep averaged 8:30, or caffeine intake that may be affecting recovery timing. Those are not glamorous metrics, but they are the ones that let a coach decide whether to hold, push, or trim.

That matters because AI systems tend to fail in two predictable ways:

1. They collect more inputs than the coach can actually use.
2. They produce confident summaries that do not change the next session.

A useful coaching loop does the opposite. It compresses noise into a decision. In the Alex Goracy transcript, Justin responds to a weekly change with a specific interpretation: lower back is the stubborn area, and the plan should be to get as lean as possible to improve the rebound. The reasoning is not “more data equals better.” It is “this pattern changes the next phase.” That is what feedback should do.

If an AI platform can’t help a coach distinguish between temporary fluctuation and real trend, it’s not a feedback loop. It’s a logbook.

Decision tradeoffs are the actual product

A lot of fitness technology sells the fantasy that better tracking removes tradeoffs. It doesn’t. It reveals them.

The Kahunas deep coaching material is blunt on this point in its PED education cases: when growth hormone worsens blood glucose, the coach has to choose between bodybuilding priority and health tradeoff. There is no magic algorithm that makes both objectives win at once. The mechanism is straightforward: more intervention can improve one outcome while degrading another.

That same logic applies to training performance feedback. If volume goes up, fatigue may rise before performance does. If calories drop, scale weight may fall before fullness and training output stabilize. If cardio increases, conditioning may improve while leg training quality takes a hit. AI coaching should not pretend these are bugs. They are the work.

The best coaching systems therefore need a tradeoff model, not just a prediction model. They should answer:

• What changed?
• What did it cost?
• What is the least risky next move?

That last question is where humans still matter most. The coach decides whether the current signal is worth paying for, and whether the expected improvement justifies the side effects.

Training feedback loops only work if the athlete can execute them

A perfect recommendation is useless if it can’t be carried out. One reason real coaches keep winning is that they do not separate plan design from compliance reality.

In the Harry Sims exchange, the client reports a routine with two 150 mg caffeine hits, a stable sleep schedule, and WHOOP data available for review. Justin’s response is not “add another metric.” It’s an example of how coaching inputs should be handled: identify the constraint, determine whether it matters, and then decide whether the information changes the plan.

That is exactly where AI can help, if it is used well. Not by replacing the coach’s judgment, but by making the feedback loop more legible:

• standardize check-ins,
• flag deviations from the normal pattern,
• summarize likely training consequences,
• and surface which variables actually moved.

But there is a catch: AI often makes adherence look cleaner than it is. A training app can show “completed,” “logged,” and “on plan,” while missing the things that determine whether the session mattered: bar speed, set quality, readiness, and whether the athlete had to reduce load to survive the workout. Coaches already know this. The problem is that technology often rewards the visible behavior rather than the relevant one.

So the useful question is not whether the athlete entered the data. The useful question is whether the data changed the next workout.

What a serious feedback loop should prioritize

For coaches, the highest-value training feedback loops are the ones that reduce avoidable uncertainty. Three categories matter most.

1. Performance trend

Did output improve, hold, or slip across the same movement or task? This includes rep quality, load tolerance, and repeated-session consistency. The goal is not a perfect PR every week. The goal is to detect whether the athlete is accumulating productive stress or just accumulating fatigue.

2. Recovery cost

What did the output cost? Sleep quality, appetite changes, morning readiness, local soreness, and overall depletion are all part of the cost function. Justin’s comments across the sources repeatedly imply that you cannot judge progress by one metric alone. Conditioning gains and rebound potential have to be weighed against how hard the body was pushed to get there.

3. Decision latency

How fast does the coach respond when the signal changes? A slow coach can be worse than a mediocre plan, because the athlete spends too long under the wrong stimulus. AI should shrink this latency by making trend changes visible sooner. But a fast response is only useful if it is grounded in the right interpretation.

That last piece is the real filter. If every small fluctuation triggers a program change, the system becomes reactive and unstable. If nothing ever changes, the feedback loop is dead. Coaching quality lives in between.

The tradeoff that matters most: more feedback versus better feedback

There is a temptation to assume that better AI means more frequent feedback. In practice, more frequent feedback can just mean more interference.

For physique athletes, especially, the hard part is not collecting all possible data. It is deciding which signals deserve action. Justin’s coaching style in the KB is consistent with that: use the signal that changes the next choice, ignore the ones that just create motion.

That principle scales well to AI coaching. The platform should not chase novelty. It should support:

• fewer but better check-in questions,
• clearer decision rules,
• and cleaner links between observed training performance and subsequent programming.

This is also where coach education matters. If the coach does not know what to do with the data, the AI is just a prettier interface. If the coach does know, the AI can save time by narrowing the search space and highlighting the most decision-relevant changes.

The bottom line for coaches

AI fitness coaching is not winning because it talks more. It wins when it improves the next call: keep pushing, hold steady, pull back, or change the stimulus. The evidence in the KB points to a practical standard: the best coaching systems are the ones that turn real performance signals into decisions without pretending the tradeoffs disappear.

If your tool gives you more metrics but fewer good adjustments, it is not a coaching upgrade. If it makes weekly performance, recovery cost, and decision timing easier to see, it may actually improve the job. That is the test.

Sources Used

• raw/_consumed/2026-05-31/kahunas-export/2026-05-31-w13-18m/clients/skip_hill___members-b7s-_d0oqqz4a1tadlvuztb0y_vzfqlr9c29tpakweg.json
• raw/_consumed/2026-05-31/kahunas-export/2026-05-31-w19-24m/clients/harry_sims___members-m-iwooxrve6qux4pdtrouhwbkhmvrxoandj1tgzve_m.json
• raw/_consumed/2026-05-31/kahunas-export/2026-05-31-w19-24m/transcripts/alex_goracy___members-m1vvgmmbnhazzipv5wqmwwhgjuyotawsompy4kzf6ri.md
• modules/03-knowledge/kahunas-coaching-deep-peds.md
• raw/_consumed/2026-05-26/troponiniq_kb.md

Sources Used

• /Users/justinharris/TroponinIQ/kb/supertrop/raw/_consumed/2026-05-26/troponiniq_kb.md
• /Users/justinharris/TroponinIQ/kb/supertrop/raw/_consumed/2026-05-31/kahunas-export/2026-05-31-w13-18m/clients/skip_hill___members-b7s-_d0oqqz4a1tadlvuztb0y_vzfqlr9c29tpakweg.json
• /Users/justinharris/TroponinIQ/kb/supertrop/raw/_consumed/2026-05-31/kahunas-export/2026-05-31-w19-24m/clients/harry_sims___members-m-iwooxrve6qux4pdtrouhwbkhmvrxoandj1tgzve_m.json
• /Users/justinharris/TroponinIQ/kb/supertrop/raw/_consumed/2026-05-31/kahunas-export/2026-05-31-w19-24m/transcripts/alex_goracy___members-m1vvgmmbnhazzipv5wqmwwhgjuyotawsompy4kzf6ri.md
• /Users/justinharris/TroponinIQ/kb/supertrop/modules/03-knowledge/kahunas-coaching-deep-peds.md