biomarker top-100 rerun protocol v2
thesis -> mechanism -> lever -> one step
- thesis: следующий прогон должен не "слегка подкрутить v1", а честно пересобрать top-100 под
Healthsystem-age logic и high-stress founder cohort. - mechanism: сохраняем v1 как reference, прогоняем расширенный universe через
bps v2, отделяем order-markers от derivations и companion modules, затем lock exact 100 только после cohort-fit and system-fit review. - lever: это делает новый список воспроизводимым и защищает от смешения крови, расчётов, imaging и surface data в одну неясную корзину.
- one step: использовать этот протокол как board-prep и выдать
framework v2 + scorecard v2 + board decisions v2.
1) preserve what already exists
do not overwrite these artifacts:
- ai/data/biomarker_top100_selection_framework_v1.md
- ai/data/biomarker_system_cluster_scorecard_v1.csv
- ai/data/first_panel_decision_packet_v1.md
- ai/data/top100_blood_urine_health_audit_v1.md
- ai/data/top_framework_v2_high_stress_achievers.html
v1 remains:
- historical baseline
- comparison set
- source of previous deltas
2) new inputs for the rerun
the new universe must include:
- all v1 order-markers
- all markers promoted in biomarker_top100_selection_framework_v2.md
- Health-driven gaps for endocrine / adrenal / system-age closure
- vascular mechanics signals already formalized in:
- ai/data/vascular_intelligence_layer_v1.md
- ai/data/vascular_intelligence_ops_v1.md
- surface-layer companions already formalized in:
- ai/data/mvp_equipment_philosophy_v1.md
3) normalize the candidate universe
every candidate must be tagged into exactly one class:
- orderable_biomarker
- derivation
- companion_module
- research_only
rules:
- only orderable_biomarker candidates compete for top-100 slots.
- derivation must still have explicit formula and source-marker dependency.
- companion_module must still have cadence and data-contract fields.
4) fields required for every candidate
the v2 scorecard row must include:
- candidate_id
- name
- class
- system_cluster
- sex_scope (shared | male | female | cycle-aware female | one-time)
- evidence
- actionability
- reproducibility
- longitudinal_signal
- operational_fit
- bps_v2
- tier_v2
- noise_level
- role_tag
- preanalytic_requirements
- discordance_rule
- retest_rule
- cadence
- system_age_targets
- claim_ids_or_gap
5) scoring procedure
step order:
1. run hard-gate filter.
2. score via bps_v2.
3. tag system_age_targets.
4. tag role_tag.
5. apply cohort-fit review:
- high stress
- high travel
- men + women
- entrepreneur / operator recovery patterns
6. run derivation cleanup:
- remove calculations from the order list.
7. run companion split:
- move skin / hair / vascular mechanics / body composition / ECG / VO2max out of the 100-count.
8. only then freeze core / overlay / gated / hold.
6) exact output shape to produce
the rerun should not output a single flat dump.
it must output:
1. shared_core_v2
2. male_overlay_core_v2
3. female_overlay_core_v2
4. gated_v2
5. derivations_v2
6. companion_modules_v2
counting rule:
- shared_core + overlay_core should be interpretable as the actual default baseline for each sex.
- exact 100 should be counted on orderable biomarkers only.
7) compulsory closure checks before lock
A. endocrine / androgen closure
must pass:
- total_T + SHBG + albumin -> free_T_calc
- LH + FSH + estradiol
- male DHT
- female progesterone + AMH
B. adrenal / HPA / SNS closure
must pass:
- cortisol_am
- cortisol_pm
- spot_urinary_cortisol_creatinine as default integrated urinary read
- urinary metanephrines
- DHEA-S
- IGF-1
practical rule:
- 24h_urinary_free_cortisol is fallback / confirmatory only, not baseline default, because collection compliance is too fragile for routine longitudinal use in this cohort.
C. vascular age closure
must pass:
- blood lipid/atherogenic layer
- blood pressure
- PWV as mechanics layer
- plaque-first carotid protocol
- CAC gating rule
D. surface age closure
must pass as companion modules:
- total body skin baseline with dermoscopy path
- face skin imaging
- trichoscopy / hair baseline
8) system-age acceptance criteria
the rerun fails if any major system is underpowered.
minimum criteria:
- each core system has at least 3 independent inputs or 1 robust composite plus 2 adjacents.
- endocrine / recovery is not based on single-point cortisol alone.
- vascular age is not based on lipids alone.
- visible-aging layers are explicitly attached even if not inside the 100-count.
- every system has a cadence policy.
9) what to do with women / cycle complexity
do not force fake symmetry.
instead:
- keep shared core.
- allow female overlay core.
- make cycle-aware fields explicit.
- add separate note for peri/post-menopause where cycle timing breaks.
10) research pass required before board-lock
before freezing the list, refresh these uncertainty zones:
- glyca method availability and comparability
- holotc availability / default vs reflex policy
- whether spot urinary cortisol/creatinine can be standardized across partner labs well enough to fully replace 24h UFC in default flow
- apoE consent boundary
- IGF-1 assay standardization
- PWV acquisition SOP / cadence tie-in to vascular loop
- skin / dermoscopy / hair data export structure
11) deliverables for this rerun
ai/data/biomarker_top100_selection_framework_v2.mdai/data/system_age_contribution_model_v2.mdai/data/top100_biomarker_scorecard_v2.csvai/data/top100_biomarker_board_decisions_v2.md- optional:
ai/data/top100_companion_modules_v2.md
12) default decision principles for the board
- prefer more signal if it improves longitudinal truth, even if it makes the baseline larger.
- do not prefer more signal if it is only prestige or incidental-finding theater.
- price is not a tie-breaker.
- method transparency is a tie-breaker.
- system-age leverage is a tie-breaker.
- dynamic-read closure beats single-marker storytelling.
13) failure modes to block
- counting derivations as biomarkers
- mixing blood biomarkers with phenotyping modules in one flat top-100 number
- keeping
cortisol_amwithout dynamic HPA closure - keeping testosterone without
SHBG + albumin - calling vascular age closed without
PWV - dropping skin / melanoma / hair because they are "not labs"
- leaving cycle-aware female logic implicit