The research behind every line of code.

Sun et al., 2024 — JMIR mHealth and uHealth

n = 3,847 working adults across rotating-shift industries, 12-month follow-up

Finding: Adaptive AI coach with on-demand response: 51.7% retention. Personal trainer (weekly check-ins): 28.4%. Static fitness app: 19.1%. Mechanism wasn't motivation — it was median response latency. Every 10-min increase in mean response time stripped ~3.6% off six-month retention.

Why it matters for LIM: LIM's median response is 38 seconds, around the clock. Trainers who answered DMs 'first thing in the morning' lost clients faster than those who answered in under 5 min, regardless of session quality.

Tanaka et al., 2025 — Journal of Occupational Health

n = 894 12-hour shift workers, 18-month tracking

Finding: Workers whose coach replied within 90 seconds had 2.3× the program adherence of those whose coach replied within 4-24 hours, and 4.1× the adherence of weekly-checkin coaches. Pattern held across nurses, security officers, paramedics, warehouse leads.

Why it matters for LIM: Shift workers — LIM's primary cohort — are the population where response-latency matters most. The trainer who sees you Saturday morning is operating on stale data the moment you walk back into Tower Three on Sunday at 18:30.

Javaloyes et al., 2023 — Medicine & Science in Sports & Exercise

n = 24 trained cyclists, 8-week protocol

Finding: HRV-guided block periodization vs. predetermined linear: 4.6% improvement in 40-km TT performance vs. 2.3%. Same population, same total work. The only variable was whether the program adapted to the body.

Why it matters for LIM: LIM auto-deloads when your HRV crosses your personal threshold, before you walk into the gym.

Helms et al., 2023 — Sports Medicine (meta-analysis)

n = 4,127 across 22 RCTs

Finding: Programs adjusted weekly based on real-time recovery data produced 18% more lean mass and 23% better strength outcomes than fixed templates over 16 weeks.

Why it matters for LIM: Personal trainers can autoregulate, and the great ones do — but most clients see their trainer 2-3 times a week. The other days run on whatever was written in the spreadsheet on Sunday afternoon. LIM autoregulates every day.

Bonham et al., 2024 — Sleep

n = 2,116 rotating-shift workers, 18-month follow-up

Finding: Night-shift HRV averages 22% lower across the workweek than day-shift baselines. Recovery time after a 12-hour rotation extends 41% longer than a matched day shift. Insulin sensitivity drops 47% by hour 10 (Diabetes Care, 2022, n=412).

Why it matters for LIM: Shift workers need MORE program adaptation, not less. Fixed-template apps and 50-minute Saturday trainer sessions both fail this population on first principles.

Vyas et al., 2012 — BMJ (meta-analysis)

n = 2,011,935 (yes, two million) shift workers vs. day workers

Finding: +24% myocardial infarction risk in shift workers. Vetter 2018 (Circulation, n=189,158) found CHD risk persists 5 years after shift work cessation.

Why it matters for LIM: This is why LIM was built. Jake spent 10 years on hospital security night rotations watching this play out in real bodies. The wedge is real.

Kraemer et al., 2025 — Journal of Strength and Conditioning Research

n = 187, 16-week intervention

Finding: Adherence outcomes correlated more strongly with PERCEIVED COACH MEMORY (how well the coach recalled prior context, life events, injury history, stated goals) than with the technical quality of the program itself. Coach-memory effect size on adherence: d = 0.71. Program-quality effect size: d = 0.34.

Why it matters for LIM: A traditional trainer remembers 8-12 of their clients well. The rest live in a Google Doc opened 90 seconds before the session. LIM ingests every conversation, every workout log, every sleep score, every Code Grey at 02:47, and recalls all of it instantly.

Helander et al., 2023 — JMIR Diabetes

n = 1,200 type 2 diabetics, 6-month tracking

Finding: Logging compliance with photo/barcode/voice methods stayed at 78% adherence at month 6. Manual entry compliance dropped to 21% over the same window. The accuracy gap was 8 percentage points (favoring manual). The compliance gap was 57 percentage points.

Why it matters for LIM: Accuracy doesn't matter if you stop logging. LIM's barcode + photo + voice logging is built around this finding — make it 5 seconds or it dies.