Research Tasks
Total: 14 tasks (RES-001..014)
Total Hours: 148-196h
Purpose: Build knowledge base for CornerMan AI agent
Status tracking: [ ] To Do | [~] In Progress | [x] Done
Branch Naming Convention
All branches: research/RES-XXX-short-description
Progress
| Phase |
Tasks |
Hours |
Done |
% |
| Phase 1: Foundation |
RES-007, 008, 014 |
34-44h |
0 |
0% |
| Phase 2: Core Sports Science |
RES-001, 002, 010 |
38-50h |
0 |
0% |
| Phase 3: Applied Knowledge |
RES-005, 006, 009, 011 |
42-58h |
0 |
0% |
| Phase 4: Enhancement |
RES-003, 004, 012, 013 |
34-44h |
0 |
0% |
| Total |
14 |
148-196h |
0 |
0% |
Phase 1: Foundation (Weeks 1-3)
| ID |
Task |
Branch |
Status |
Est. Hours |
Priority |
Key Output |
Done When |
| RES-007 |
Heart Rate Variability (HRV) Research |
research/RES-007-hrv-research |
[ ] |
14-18h |
High |
RMSSD interpretation, readiness algorithms, confounders |
Comprehensive HRV document covering: RMSSD vs SDNN, normal ranges by age/fitness, morning vs continuous measurement, confounders (alcohol, caffeine, illness), readiness algorithm design, recovery tracking protocols |
| RES-008 |
Sleep Science for Athletic Performance |
research/RES-008-sleep-science |
[ ] |
12-16h |
High |
Stage benchmarks, debt impact, napping protocols |
Sleep research document covering: stage durations (deep/light/REM/awake) benchmarks by age, sleep debt impact on performance, napping protocols for athletes, sleep hygiene recommendations, ring/band sleep tracking accuracy |
| RES-014 |
AI Health Coaching Safety & Liability |
research/RES-014-ai-safety |
[ ] |
8-10h |
Low |
What CornerMan CAN vs MUST NOT do; disclaimers |
Safety framework document covering: medical advice boundaries, required disclaimers, liability considerations, content CornerMan must refuse, hedging language requirements, emergency detection and referral protocols |
Phase 2: Core Sports Science (Weeks 3-5)
| ID |
Task |
Branch |
Status |
Est. Hours |
Priority |
Key Output |
Done When |
| RES-001 |
UFC Performance Metrics & Benchmarks |
research/RES-001-ufc-metrics |
[ ] |
12-16h |
High |
Normative ranges by weight class for VO2max, HRV, body comp |
UFC metrics database covering: VO2max ranges by weight class, resting HR norms, HRV baselines, body composition targets, grip strength, reaction time benchmarks |
| RES-002 |
UFC Training Methodologies & Periodization |
research/RES-002-ufc-training |
[ ] |
14-18h |
High |
Fight camp structure, session types, HR zone profiles |
Training methodology document covering: fight camp periodization (8-12 week structure), session types (sparring, pad work, grappling, S&C), HR zone distribution per session type, taper protocols, deload weeks |
| RES-010 |
Nutrition Science for Athletes |
research/RES-010-nutrition |
[ ] |
12-16h |
Medium |
Macro targets, meal timing, AI photo estimation accuracy |
Nutrition guide covering: macronutrient targets by sport/goal, meal timing around training, hydration protocols, supplement evidence ratings, AI food photo estimation accuracy benchmarks |
Phase 3: Applied Knowledge (Weeks 5-7)
| ID |
Task |
Branch |
Status |
Est. Hours |
Priority |
Key Output |
Done When |
| RES-005 |
Fight Preparation Analytics & HR Zones |
research/RES-005-fight-prep |
[ ] |
10-14h |
High |
MMA-specific HR zones, recovery windows, sleep needs |
Fight prep analytics covering: MMA-specific HR zone definitions, inter-round recovery analysis, pre-fight sleep requirements, training load management during camp, taper indicators from wearable data |
| RES-006 |
Weight Management & Cutting Science |
research/RES-006-weight-cutting |
[ ] |
12-16h |
High |
Safe protocols, body comp targets, safety guardrails |
Weight management document covering: safe weight loss rates, water cut protocols and risks, body composition monitoring, rehydration protocols, safety guardrails for CornerMan (never recommend dehydration cutting, flag dangerous rates) |
| RES-009 |
Recovery Science & Modalities |
research/RES-009-recovery |
[ ] |
10-14h |
Medium |
Evidence-ranked recovery strategies (CWI, foam rolling, etc.) |
Recovery guide covering: cold water immersion evidence, compression garments, foam rolling, active recovery, massage, sleep optimization, HRV-guided recovery, evidence quality ratings for each modality |
| RES-011 |
GPS Activity & Training Load Modeling |
research/RES-011-training-load |
[ ] |
10-14h |
Medium |
TRIMP, ACWR, race prediction formulas |
Training load document covering: TRIMP calculation methods, acute:chronic workload ratio (ACWR), injury risk thresholds (>1.5 flag), race prediction formulas (Riegel), GPS metric interpretation (pace zones, elevation gain) |
Phase 4: Enhancement (Weeks 7-9)
| ID |
Task |
Branch |
Status |
Est. Hours |
Priority |
Key Output |
Done When |
| RES-003 |
UFC Performance Institute Research |
research/RES-003-ufc-pi |
[ ] |
8-12h |
Medium |
Published papers, athlete normative database |
Summary of UFC PI publications covering: their normative athlete database, testing protocols, published research papers, performance benchmarks they've established |
| RES-004 |
Combat Sports Wearable Technology |
research/RES-004-wearables |
[ ] |
8-10h |
Medium |
Which devices fighters use, accuracy limitations |
Wearable tech review covering: devices used by UFC fighters, accuracy studies for HR/HRV/SpO2 in sport context, ring vs watch vs band comparison, limitations during combat sports (motion artifacts) |
| RES-012 |
Readiness & Strain Scoring Algorithms |
research/RES-012-scoring |
[ ] |
8-12h |
Low |
How WHOOP/Oura/Garmin score; build Sellwyse's own |
Scoring algorithm document covering: WHOOP strain (0-21) methodology, Oura readiness score components, Garmin Body Battery algorithm, design Sellwyse's own readiness + strain scoring using available ring/band data |
| RES-013 |
Anomaly Detection in Health Data |
research/RES-013-anomaly |
[ ] |
8-10h |
Low |
RHR elevation, SpO2 dips, overtraining signals |
Anomaly detection guide covering: resting HR elevation patterns (illness, overtraining), SpO2 dip thresholds, HRV suppression detection, overtraining syndrome markers, when to flag vs when to alert |
How Research Gets Used
| Storage Method |
What Goes There |
Examples |
| System Prompts (hardcoded) |
Safety guardrails, disclaimers, scope limits |
"Never recommend dehydration weight cutting" |
| RAG Knowledge Base (pgvector) |
Research summaries, citations, benchmarks |
Paper findings, normative tables |
| Backend Rules (Rust code) |
Threshold alerts, formulas, safety limits |
TRIMP calc, SpO2 <94% alert, ACWR >1.5 flag |
| Lookup Tables (PostgreSQL/JSON) |
Normative ranges, weight class data, zone defs |
UFC benchmarks by weight class |
| Few-Shot Examples (future) |
Ideal CornerMan responses |
Curated from real usage logs |
Key Safety Guardrails for CornerMan AI
MUST embed in system prompt from day 1:
- Never diagnose medical conditions
- Never recommend dehydration-based weight cutting
- Never recommend losing >1.5% body weight/week
- Always flag body fat below healthy minimums (male <5%, female <12%)
- Always include: "This is not medical advice. Consult a healthcare professional."
- Use hedging: "Research suggests..." not "You must..."
- Cite evidence quality when making claims