Built from the McPhases women’s health cohort.
Evidence Brief
Women’s Health Evidence.
PI was evaluated on the McPhases Women’s Health dataset. This page summarizes what the latest cohort artifacts support, what PI demonstrated, and what it does not claim.
This brief reflects the latest women’s health Meta refresh dated April 21, 2026. It is an evidence summary, not a clinical validation claim.
At a glance
What this evidence covers.
PI processed real, time-based women’s health histories and kept early ideas, refined structure, and confirmed findings separate. It also produced a cohort-level map of recurring structure across the dataset.
Separate episode-level representations were preserved.
Confirmed cohort structure after gating and authority rules.
The current public summary is based on the latest refresh.
What PI demonstrated
What the run shows in plain English.
The value here is not a single accuracy number. It is that PI held structure, limits, and evidence quality together while working through changing cases.
Different stages stayed separate
PI did not collapse early hypotheses, structural candidates, and confirmed findings into one undifferentiated answer.
Continuity was carried across time
Earlier confirmed structure could persist, shift, or be reconfirmed later instead of being flattened into a single snapshot.
Action stayed governed after selection
PI could release a selected action lane in one case and abstain in another, instead of forcing the same answer everywhere.
Continuity view
How findings behaved across episodes.
- 22 participants had one authoritative episode.
- 10 participants kept earlier confirmed structure because the latest episode did not meaningfully reassess it.
- 8 participants showed a later structural shift without direct reconfirmation of the earlier picture.
- 2 participants showed direct reconfirmation across episodes.
Why this matters
PI keeps honesty when the story changes.
Real health histories rarely move in a straight line. This run shows that PI can keep earlier confirmed findings on record, distinguish new shifts from true reconfirmation, and avoid pretending every episode says the same thing.
Representative cases
Two examples from the latest run.
These anonymized examples show both sides of the women’s health run. One case produced a selected action lane with minimum-effective and optimal range planning, while another kept the stronger action lane gated.
Case A
Explanation and a selected action lane stayed separate.
90-day history
127
Confirmed edges
40
Confirmed interactions
67
Confirmed chains
- 231 nodes and 818 edges met sufficiency for Stage 3 testing.
- The focal target was cramps severity.
- A stable explanatory signal was available on that focal target.
- The same participant also produced one selected action lane on glucose mean through exercise minutes.
- That action lane carried one minimum-effective and optimal range plan rather than a free-form suggestion.
Case B
Explanation held even when action stayed gated.
90-day history
138
Confirmed edges
21
Confirmed interactions
92
Confirmed chains
- 209 nodes and 775 edges met sufficiency for Stage 3 testing.
- The focal target was again cramps severity.
- A confirmed explanatory signal was available for that focal target.
- Recommendation was again withheld because the stronger action lane did not survive governance.
- An actionable-with-caution item remained available on a different target.
Cohort map
What the cohort-level map showed.
Above the individual cases, PI also built a cohort-level map of recurring structure across participants. This is where bridges, vulnerabilities, hidden levers, and recurring phenotypes became visible.
Recurring bridges
- Fatigue was a recurring bridge across 35 participants.
- Stress self-report appeared as a bridge in 30 participants.
- Cramps severity appeared as a bridge in 28 participants.
- Indigestion and bloating also appeared repeatedly.
Common control levers
- Distance traveled appeared most often among control levers.
- Vigorous activity appeared repeatedly across the cohort.
- Exercise minutes also appeared as a recurrent lever.
Shared systems showed up across many participants.
The latest cohort artifacts surfaced a symptom-symptom regulation system in 42 users and a physiology-symptom regulation system in 40 users. These are emerging shared systems, not settled population claims.
Two recurring phenotype groupings appeared.
The latest Meta refresh surfaced an Autonomic-Metabolic Regulatory phenotype across 38 users and a General Multi-System phenotype across 22 users.
Some links repeated across users.
One recurring link was a glucose instability state connected to cramps severity across 20 users. In the latest artifacts it remains a prototype signal, not a settled clinical fact.
Governance
What passed the integrity checks.
- Stage separation remained intact.
- Privacy thresholding was enforced.
- No model-training artifacts were present in the Meta layer.
- Publication thresholds were configured before reporting.
- Lineage completeness remained above threshold.
Claim boundaries
What this page does not claim.
- This is not a diagnosis or treatment recommendation page.
- This is not a clinical validation claim.
- This does not show publication-grade mediation evidence yet.
- The latest validation summary explicitly reported no publication-eligible mediating signals.
- The latest Meta refresh summarized the current artifact set; it was not a brand-new incoming-data rerun.
Discuss the evidence
Talk through the women’s health findings.
The best discussion is usually about scope and fit: what PI demonstrated here, what it did not claim, and where partner or research work should go next.