One athlete profile, this week's training decision
Send athlete exports. Know what to adjust this week.
An AthDash Causal Driver Audit reviews one athlete's workouts, wellness, travel notes, and benchmark results from existing exports. It returns a plain-English Driver Card for this week's coaching decision: what is most likely worth changing, how strong the evidence is, and what is too thin to say yet.
Use it before the next check-in, workout change, or athlete conversation, with caveats and no-claim states already visible.
Alex Rivera · 10K build · next 7 days
Move heavy lower-body strength before 5pm or separate it from key running sessions.
Allowed wording: likely worth adjusting this week. Do not claim injury prevention or medical benefit.
Audits one athlete profile from existing exports
Start with the page that matches your coaching workflow.
AthDash is not a generic athlete dashboard. These pages explain the audit artifact, the export-first workflow, and the evidence packets an AI coaching agent can read before it writes.
Causal Driver Audit for coaches
Send athlete exports and get Driver Cards for what to adjust, what to watch, and what not to say yet.
Export-firstTraining data export audit
Use CSVs, logs, and wearable exports before promising a perfect integration or live connection.
AI backendAI coaching backend for athlete data
Give the agent evidence state, caveats, and refusal boundaries before it drafts coaching language.
When readiness scores are not enough
Readiness describes state. It does not automatically name the training lever.
ArtifactWhat is a Driver Card?
The coach-readable card that carries effect, interval, caveat, and decision license.
Agent inputWhat is an evidence packet?
The compact deterministic packet an AI coaching agent can read before it writes.
AthDash answers the question coaches actually ask: what should change now?
It reads the data you already have for one athlete, checks whether the pattern holds up, and returns a simple audit: what to adjust, what to watch, and what not to say yet. If there is not enough data, it says that instead of making up advice.
False patterns look persuasive
Track enough signals and something will line up with a good session. AthDash checks the pattern against your athlete's own history before it becomes coaching advice.
Scores do not tell you what to change
A readiness number can tell you today looks rough. It does not tell you whether the next move is sleep timing, training load, recovery, spacing, or doing nothing yet.
The dangerous part is confidence
A coach can live with "not enough evidence yet." The risk is a system that sounds certain from a few useful data points and turns a hunch into a rule.
The output is a weekly decision card, not another dashboard.
For each athlete, AthDash shows the likely adjustment, evidence strength, sample size, caveats, and how carefully to word the recommendation. A coach can use it in the next check-in; an AI coaching agent can read the same answer before drafting notes.
Aerobic decoupling → Time trial
lag 1-42 d / n=18 / FIT rides
For Alex, steadier long rides are tracking with stronger late-race pace this week. Adjust endurance work, but keep the fatigue caveat attached.
ADVISE: It is OK to say this likely helped Alex's late-race pace, as long as the fatigue caveat is included.
What should I change for my athlete this week?
AthDash points to the adjustment, shows how strong the evidence is, and keeps the uncertainty visible so the coach can act without guessing.
What can my agent safely say?
Before a coaching agent writes a check-in, it can read the same answer: suggested adjustment, effect, range, sample, caveats, and allowed wording.
The same answer travels everywhere.
Effect, range, sample, caveats, and allowed wording stay attached across the report, console, and AI-readable output for the same athlete.
The useful part: AthDash helps the coach make the call first. If an AI coaching agent drafts the check-in, it gets the same evidence and caveats.
Use the Driver Card before the next check-in.
Export the same effect, range, caveats, and sample.
Give an AI coaching agent the allowed wording.
The same input can help or hurt depending on the athlete's current state.
Most tools report one average effect. AthDash checks whether the answer changes under your athlete's real conditions, and only promotes that nuance when the evidence is strong enough.
If you use AI for check-ins, give it the card first.
AthDash can sit in the background as the stats layer for the athlete in question. Instead of asking an AI coaching agent to reread weeks of workouts and wellness logs, the agent reads the same Driver Card the coach sees: suggested adjustment, effect, range, sample, caveats, and allowed wording.
athlete="A.R.", actionable_only=true)
driver: "aerobic_decoupling"
outcome: "time_trial"
evidence: SUPPORTED (n=19, p<.001)
effect: +2.4% late-race pace
license: ADVISE
interval: [+0.78, +1.93]
caveat: "include fatigue flag"
› athdash_can_i_claim(
driver="ctl", outcome="ftp",
athlete="A.R.")
evidence: INSUFFICIENT (n=3)
license: DECLINE
The stats are digested first
Your coaching agent gets a small summary of the athlete's adjustment, effect, range, sample, caveats, and allowed wording. It spends time writing clearly, not re-parsing raw history.
It gets the numbers behind the sentence
Before the agent writes, it can read the effect size, uncertainty, stress-test status, and caveats for your athlete. That is more useful than asking a model to infer stats from raw exports.
The wording is still capped
Every finding carries a rung, DECLINE to ACT. The agent stays inside that permission instead of making the advice sound stronger than the evidence.
Every answer has a receipt
Each line the agent writes traces back to your athlete's digested data, range, and allowed wording. When the athlete asks why, the answer already exists.
The agent gets statistical depth without permission to invent performance advice.
Start with one athlete audit. Reuse the same answer anywhere.
The first surface is a Driver Card a coach can use in an athlete check-in. The same answer can feed a roster console or a read-only AI coaching tool without losing the effect, range, caveats, sample, or allowed wording.
Driver Card report
Turn one athlete's existing exports into a shareable audit: what likely matters this week, how strong the evidence is, and what not to say when the data is too thin.
Roster console
Run the same per-athlete evidence gate across a roster and see which relationships are supported, exploratory, borrowed, or declined for each athlete.
Agent grounding
Expose the same findings through read-only evidence packets, so a coaching agent reads your athlete's stats before it writes advice.
How AthDash decides whether to say yes, maybe, or no.
Five checks, in order, every time. This is why AthDash can recommend an adjustment, flag it as a hypothesis, or say the data is too thin.
Ingest
sourcesAccept the athlete exports coaches already have, including wellness, sessions, FIT files, and benchmarks, then align them to one athlete timeline.
Model
training questionDeclare the possible input, the outcome, the timing window, and the obvious confounders for your athlete. AthDash tests the question; it does not treat a pattern as truth.
Stress test
placebo + uncertaintyCheck each estimate against confounders, small-sample uncertainty, and placebo shifts. Fragile relationships are downgraded or withheld instead of promoted.
Gate
honestyReturn SUPPORTED, EXPLORATORY, or INSUFFICIENT, then attach how strongly the coach or agent may word the recommendation.
Deliver
with uncertaintyShip the finding as your athlete's Driver Card, console payload, or AI-readable evidence packet, with the effect, range, caveats, and allowed wording still attached.
The catalog comes before the model.
AthDash does not ask an AI coaching agent to discover a training rule from raw exports. The public operating model is stricter: a signal catalog decides what can be trusted, a registry frames the athlete-specific question, and the engine returns only the claim the evidence licenses.
Same athlete data, same catalog, same engine path: the gates, estimates, intervals, evidence states, and wording permissions stay reproducible.
Every candidate signal starts with an eligibility rule.
Personal baselines come first. Device-class differences stay visible. Attractive shortcuts are rejected before they reach the audit.
The training question has to be declared.
The system names the driver, outcome, lag window, aggregation, direction, and likely confounders before estimating anything. A vague dashboard pattern never becomes a Driver Card by itself.
Adjustment is derived, not improvised.
AthDash uses the declared relationship and graph structure to decide what must be adjusted for, what should not be adjusted away, and where small samples or placebo shifts should downgrade the claim.
The output carries its own permission.
SUPPORTED can become advice, EXPLORATORY stays a hypothesis, and INSUFFICIENT becomes a clean no-claim state. The same license travels into the Driver Card, roster payload, and AI-readable packet.
Every training claim has to survive the audit.
AthDash does not prove a claim by storing more data. It licenses a claim only when the athlete's own outcomes survive adjustment, interval width, placebo checks, and the no-claim gate.
The engine tries to break the finding first.
Each Driver Card starts with real coach exports and repeatable benchmark outcomes. Then the registry frames one driver-to-outcome question, the causal graph derives the adjustment set, and the estimator checks whether the interval and placebo test still permit advice.
state: supported / n: 18 / ci: +0.9..+3.8 / placebo: clean / caveat: fatigue flag
Backdoor set derived from the causal graph, not chosen after seeing the result.
Effect is shown with its 95% range. Width stays visible before the coach acts.
Lag-shifted control can demote the finding when the pattern is fragile.
Say the adjustment likely helped, but keep the fatigue caveat attached. If the interval crossed zero, this would become HYPOTHESIZE or WITHHOLD.
HAC CI coverage at n=12 in the validation harness after small-sample t inference.
Max Type-I rate across null validation cells. Weak null patterns stay exploratory.
Insufficient edges carry no effect, and exploratory reasons state the real failure mode.
What you get in a founding audit.
The paid founding offer is a focused audit for coaches and trainers with real exports from the athlete in front of them. You send the athlete history; AthDash returns Driver Cards for the current training decision: what to adjust, what to watch, and what not to say yet.
One athlete's existing exports, benchmark history, and notes on obvious confounders.
Plain-English Driver Cards for your athlete's current decision, with effect, likely range, evidence state, caveats, allowed wording, and next-data guidance.
No effect from insufficient data, no personalized claim from a borrowed prior, no medical diagnosis, and no claim that outruns your athlete's own history.
- Export-based ingestion
- Shareable Driver Card reports
- Confounder and caveat review
- No data resale
- Roster console payloads
- Supported, exploratory, borrowed, or declined states
- Compact evidence packets
- Priority support
- Production billing
- Team permissions
- Managed export-source setup
- Department support
Questions, answered honestly.
What does AthDash actually do?+
Where does the data come from?+
Is this a medical device?+
How is this different from a readiness score?+
Can I export the answer?+
What happens when there isn't enough data?+
Is AthDash causal or correlational?+
How does this help an AI coaching agent?+
What is an effect modifier?+
Does it replace a coach?+
Glossary
Short definitions for the words that show up in Driver Cards, roster payloads, and AI-readable packets.
- evidence packet
- The small AI-readable version of one athlete's Driver Card: suggested adjustment, outcome, evidence state, effect, range, sample, caveats, and directive.
- claim gate
- The background check before a sentence becomes advice. If athdash_can_i_claim returns WITHHOLD or DECLINE, the agent cannot claim the relationship.
- allowed wording
- How far the agent may go, from DECLINE to ACT. The estimate carries its own permission instead of relying on the model to self-police.
- read-only tool
- The agent can query AthDash, but it cannot mutate athlete records, rewrite evidence, or manufacture a new effect through the local MCP surface.
- borrowed prior
- A cohort signal for cold-start athletes. It may be offered only as borrowed evidence, never as if it were proven for that athlete.
- decline
- A valid answer, not a failure. When the data is too thin, the safest product behavior is to stop the advice before it reaches the athlete.
Make this week's training decision from your athlete's evidence.
Start here: what should I change for my athlete this week?