endo-protocols

Methodology — Case Study 001

Data source

All data was pulled from my personal WHOOP 4.0 account on 2026-04-23 using the whoop-ai-mcp open-source MCP server + the scripts/pull-whoop.mjs wrapper in this repository. The 45-day window captures 27 days of pre-detox baseline, the 14 days of Core Restore (with infrared sauna sessions throughout), and the immediate post-detox clean window.

Blood-work data comes from two Quest Diagnostics panels ordered through Parsley Health (January 20, 2026 baseline + April 23, 2026 post-cycle). See biomarker-results.md for the full directional summary.

⚠️ Concurrent medications and hormonal status

During the entire 45-day case-study window (and during the June 2025 7-day cycle), the author was not using:

• ❌ Hormonal contraception (combined OCP, progestin-only methods, hormonal IUDs, etc.)
• ❌ Hormonal endometriosis suppression (GnRH agonists, danazol, progestins, aromatase inhibitors)
• ❌ Hormone replacement therapy

Natural menstrual cycles were ongoing throughout. This was a deliberate root-cause exploration of how the body responds to liver-clearance support without a hormonally suppressed baseline.

Cycle phases were confirmed two ways:

1. Hormone testing — a January 2026 fasted-morning progesterone draw in the upper luteal-phase reference range anchors that timepoint to mid-luteal phase
2. WHOOP physiological signature — sharp HRV + recovery drop paired with elevated resting HR and skin-temperature peak appears on January 24, 2026 and April 22–23, 2026, consistent with ~28-day cycle interval

Cross-mapping to the protocol window:

Phase	Date range	Estimated cycle days	Cycle phase
Pre-detox baseline	Mar 9 → Apr 4	Full cycle cross-section	Mixed
14-day Core Restore	Apr 5 → Apr 18	~cycle day 14 → 27	Ovulation + full luteal
Post-detox / Lab 2 draw	Apr 19 → Apr 23	~cycle day 28 → menses transition	Late luteal / pre-menses

Why this is methodologically important: baseline HRV is typically lower in luteal phase than follicular. The 14-day cycle’s HRV improvement happened during the cycle window where HRV would naturally be suppressed — meaning the protocol effect is, if anything, understated relative to a cycle-phase-controlled comparison.

See research/birth-control-context.md for full scientific rationale of why this choice matters for data interpretation, and why the protocol may behave differently for readers on hormonal contraception.

Tools used

• WHOOP 4.0 strap (worn 24/7)
• WHOOP API v2 via OAuth2 + PKCE
• whoop-ai-mcp as the auth + API client wrapper
• Node 20+ to run the analysis scripts

Endpoints used

• /v2/recovery — daily recovery score, HRV (rmssd), resting heart rate, SpO2, skin temperature
• /v2/activity/sleep — sleep stage breakdown, duration, performance, respiratory rate
• /v2/activity/workout — individual workouts with strain and HR zones
• /v2/cycle — daily physiological cycle with aggregate strain

Phase definitions

See phases.json. Phases are date ranges I assigned based on my own calendar:

Phase	Range	Days	Definition
Pre-detox baseline	2026-03-09 → 2026-04-04	27	Normal eating, normal training, normal life
Core Restore	2026-04-05 → 2026-04-18	14	On-protocol: elimination diet + supplements + 2 shakes/day + infrared sauna sessions throughout
Post-detox clean	2026-04-19	1	Immediate rest day following the 14-day cycle

Statistical approach

I did not run significance tests. The sample size is too small, phases are unequal length, and the variables are highly autocorrelated day-to-day. Running a t-test would be dressing up a case study as a trial. Instead, I report:

• Medians per phase (less sensitive to single bad nights than means)
• Percent change for HRV specifically (log-distributed, percent is more interpretable than absolute ms)
• Direction and magnitude, framed as observations, not conclusions

Confounders are listed explicitly in the main case study post.

What I didn’t do (on purpose)

• Withhold data. All 45 days of daily summaries are in data/daily-summary.csv. If you want to run your own analysis, the CSV is the source.
• Cherry-pick windows. The phase boundaries are my calendar dates, not the dates that make the effect look biggest.
• Subtract training load. I considered normalizing HRV against cycle strain, but the strain variation during the window is within a small enough range that the correction wouldn’t change the direction of the effect. If you’re a more elite athlete reading this and want the strain-adjusted analysis, the raw data is public — please DIY and publish your version.

Data pipeline reproducibility

# 1. Pull raw data (outputs ~/Documents/Dev Projects/diet-signal/whoop-raw.json by default)
node scripts/pull-whoop.mjs --days 45 --out /tmp/whoop-raw.json

# 2. Export anonymized daily summary
node scripts/export-csv.mjs \
  --in /tmp/whoop-raw.json \
  --out case-studies/001-core-restore/data/daily-summary.csv

# 3. Generate phase comparison table + daily detail
node scripts/analyze-phases.mjs \
  --data /tmp/whoop-raw.json \
  --config case-studies/001-core-restore/phases.json \
  --detail 14

The /tmp/whoop-raw.json file contains user_id and email — it must not be committed. Only the anonymized CSV ships in this repo.

What would make future case studies better

• Longer baseline windows (60+ days) to reduce sensitivity to any single outlier
• Training-load matching between phases — if strain spikes during a test phase, isolate that
• Menstrual cycle phase logging so HRV is compared across similar cycle stages
• Concurrent symptom diary (already addressed via templates/reintro-tracker.md)
• Bloodwork bookends — Parsley runs labs periodically; pairing liver enzymes and inflammatory markers (hs-CRP, ferritin, homocysteine) with WHOOP data before/after would add a mechanistic layer

This site is open source. Improve this page.