Inside the TYR Health Context Report

Health data is not health context

The blank chat box

A general AI starts with none of the facts that make a fitness or wellness question personal. The user has to remember what matters, summarize several apps, state a goal, explain constraints and identify gaps before the model can answer well. Most prompts leave something important out. TYR prepares the situation first.

Screenshots

A screenshot is easy to share but weak as evidence. It hides calculation definitions, compresses one time window into a dashboard and rarely shows the data that is absent. A stack of screens may mix incompatible units and periods while forcing the AI to extract values visually.

Raw data dumps

A raw export creates the opposite problem: thousands of rows with no priorities, supported baselines or missing-data rules. More data is not automatically more context. An AI can overvalue the most visible number, recompute a metric differently or connect records that only happen to overlap in time.

A health context report is neither a score nor a raw backup. It is a bounded, interpretable record with a defined window, recent detail, longer-term context and explicit rules for reasoning.

What an AI needs before it can reason

Goal and constraints

The same measurements mean different things for a person gaining muscle, preparing for a race or reducing fatigue. TYR states the goal, measurable target, timeline, user notes, injuries and constraints before inviting a recommendation.

Personal baselines

Population averages describe groups; they do not establish what is normal for one person. TYR places current values and supported deviations beside the user’s own baseline. Each baseline identifies its window, mean, standard deviation, sample size and coverage so an AI can see whether a change is unusual for that individual.

Coverage and confidence

A number without coverage can look more certain than it is. TYR identifies contributed and excluded days and marks evidence as complete, usable, partial or insufficient. Recent daily detail gives sequence; the longer window shows what is typical. Partial current days remain partial.

Missing-data semantics

Zero steps is an observation. No step record is an absence of evidence. Missing sleep, heart rate, HRV, nutrition, bodyweight or training can reflect wear time, battery, synchronization, permissions or incomplete logging. Converting absence to zero invents inactivity, fasting or nonadherence.

What health data needs to be AI-ready

• A defined time window
• A stated goal
• Personal baselines
• Recent deviations
• Coverage and confidence
• Missing-data semantics
• Training and muscle context
• Nutrition completeness
• Injuries and constraints
• Intervention history
• Units, dates and calculation definitions
• Clear medical and causal boundaries

What TYR assembles

The report opens with AI coach instructions, calculation conventions, the user’s goal and reliability notes. It then moves from baselines and current deviations into daily records and domain-specific detail. That order makes interpretation rules visible before the measurements.

Training and muscles

TYR identifies recent strength and cardio sessions, duration, working sets, strength volume, movement patterns and the workout or set detail recorded in the app. It maps exercise contributions into muscle exposure, fatigue zones and longer-term training-load context. Those estimates can reveal distribution and sudden changes; they do not measure tissue damage, personal effort or injury risk. General volume also cannot replace lift-specific weight, reps, effort and progression when the question is about one performance goal.

Nutrition

Nutrition summaries separate logged-day averages from calendar-day assumptions, label the current day as partial when appropriate and state coverage. Consistent logging increases confidence but does not guarantee complete intake or accurate portions. The AI is instructed to ask targeted questions before making strong claims from sparse records.

Sleep, recovery, activity and bodyweight

Sleep duration and timing, steps, daily average heart rate, resting heart rate, morning HRV and bodyweight are kept as distinct measures. TYR does not substitute daily average heart rate for resting heart rate or publish a formal HRV deviation without a compatible baseline. Wearable accuracy and wear time remain limitations. Short-term bodyweight changes are interpreted in light of water, glycogen, sodium and measurement timing.

Goals, injuries and interventions

The report carries the current goal, status, target date, measurable target, user notes, active injury context, unusual status flags and recorded interventions. These are context for safer reasoning, not evidence that an intervention caused an outcome.

Fitness assessments and performance context

Available assessments may include a Cooper test estimate, one-mile run, dead hang, plank, flexibility and body-profile measurements. Missing assessments remain unavailable. A single body measurement cannot establish a trend, and an estimated performance measure is not a diagnosis.

Why the resulting answer is different

Generic input versus structured context

“How can I improve my deadlift?” invites a generic program. A structured report can show whether training is consistent, protein is adequately logged, bodyweight is moving, sleep is short, workload is unusual and—just as important—whether the actual deadlift progression is missing. The answer can then focus on the decision the evidence supports instead of filling gaps with assumptions.

Strong conclusions versus weak conclusions

Coverage lets the AI say “sleep has been consistently short” with more confidence than “short sleep caused the stalled lift.” If the lift trend is absent, the honest result is insufficient evidence about progress. That limitation is useful: it identifies the next measurement that would improve the decision.

Observation → explanation → evidence → action → test

TYR’s coach instructions ask broad reviews to move from an observed pattern to a possible explanation, cite the supporting evidence, state confidence, suggest a small action and define what to watch over the next 7–14 days. The format encourages testable next steps instead of dramatic changes based on one unusual day.

The structure cannot guarantee a correct answer, but it removes common failure points: unlabeled windows, mixed units, invisible missingness, partial days treated as complete and population comparisons presented as personal facts.

Boundaries that make the context trustworthy

Correlation is not causation

A change that follows poor sleep or a nutrition intervention may be worth investigating. Timing alone does not establish why it happened. TYR instructs the AI to present plausible mechanisms as hypotheses unless repeated evidence supports more.

Missing is not zero

Unknown records stay unknown. TYR also reports calculation definitions, dates and units rather than asking the AI to infer them from raw rows.

General wellness, not diagnosis

A report can summarize recorded behavior, expose gaps, compare recent values with supported personal baselines and help form low-risk experiments. It cannot verify every log, diagnose or rule out a condition, determine the cause of a change, predict injury or replace a qualified healthcare professional.

User-controlled sharing

Generating a static report does not send it to an AI. The user chooses whether to copy or share it. TYR’s optional direct ChatGPT connection is separately authorized, read-only and revocable.

Read the complete real 90-day report and actual ChatGPT analysis to see how the format separates strong conclusions from missing evidence. Then read How TYR Works With ChatGPT for the six scoped tools and the difference between a static export and a direct connection.