Recorded facts
The coach-supplied team base plan; completion or final absence/non-completion; repetitions or segments and recorded results; supplied RPE, fatigue, and notes.
EXPLAINABLE COACH DECISION SUPPORT
Technical methodology
AccelPhase helps coaches make better training decisions. It turns training plans, athlete results, and feedback into structured, individualized recommendations, checks them against code-enforced rules, and gives the coach a clear basis for deciding what comes next.
01 / DECISION FLOW
From a team plan to the coach's next decision
- 01Team training objective
- 02Athlete history and current-session signals
- 03Data completeness and comparability checks
- 04Individual recommendation draft
- 05Code rules and threshold validation
- 06Coach approval, modification, or rejection
- 07Athlete execution
- 08Completion and feedback inform the next decision
02 / INFORMATION CATEGORIES
The source of every recommendation stays visible
Calculations
The system calculates differences between the plan and recorded results in comparable units and order.
Code rules
Code independently classifies minor or major changes; missing, stale, superseded, or incompatible status is never silently ignored.
AI interpretation
AI can draft an explanation only within structured facts and code rules; it cannot replace the coach's decision.
03 / TRAINING DECISION WALKTHROUGH
One base session, three coach-review outcomes
Recorded fact
Base plan: 60 m × 4, target 8.40 s, 4:00 recovery, normal load.
Calculation
Athlete B changes from 8.40 to 8.55 s and from 4:00 to 4:30.
Code rule
B remains minor because item count, order, type, purpose, and primary category remain unchanged and numeric deltas stay within thresholds.
AI interpretation
Athlete A: keep. Athlete B: draft small adjustment. Athlete C: insufficient evidence for individual review.
Coach decision
These review choices are unselected: the coach may review, modify, approve, or reject.
04 / CURRENT RULE VERSION
Code-owned change thresholds
- Adding, removing, reordering, or changing an item's training type, purpose, or primary category is a major adjustment.
- A total-work change over 15% is a major adjustment.
- A target speed or time change over 5% is a major adjustment.
- An intensity change over 5 percentage points is a major adjustment.
- A recovery change over 20% is a major adjustment.
- If AI calls a change minor but code classifies it as major, the draft is rejected and regenerated or falls back.
- Missing information is not treated as zero. Insufficient data lowers certainty and may require individual coach review.
05 / DATA USED
Supported inputs
Coach team base plan; completion or final absence/non-completion; repetitions/segments and recorded results; supplied RPE, fatigue, and notes; eligible comparable personal history; coach-confirmed persistent constraints; and missing, stale, superseded, or incompatible status.
DELIBERATELY EXCLUDED
Inputs we do not assume
This method does not assume weather, wind, altitude, terrain, or wearable readings are always known; they are deliberately excluded from required inputs here.
06 / STRUCTURED OUTPUT AND VALIDATION
Drafts must pass independent checks
Recommendations use predefined structured fields. Empty, invalid JSON, or schema-rejected output is rejected; an independent minor/major code check runs, and conflicting recommendations are rejected. The system distinguishes an original draft from a coach edit.
07 / FAILURE AND DEGRADED MODES
When it cannot decide, the system explains rather than guesses
Stale drafts cannot overwrite newer plans, results, status, or coach edits. During an AI outage, the page keeps a local factual summary and a missing-data explanation; the coach can retry or write feedback. Retries are idempotent. Safety feedback is excluded from bulk send and individually reviewed.
08 / PRIVACY, PERMISSIONS, AND AUDIT
Task-minimum data and traceable actions
The system uses task-minimum data and server-authorized team/athlete access; personal and team contexts are separate, authorized writes are idempotent, and significant draft, review, and send actions are traceable. Tests do not establish all-jurisdiction legal compliance.
Read the children’s privacy notice09 / EVIDENCE BOUNDARIES
Different evidence does not imply a stronger claim
Implemented behavior describes what code does. Local verification describes checks in the current environment. Production verification describes verified live behavior. Adoption describes whether real users use it. Training impact needs separate outcome evidence. This walkthrough explains a workflow, not a testimonial or performance-improvement claim.