top-100 methodology + provenance review v1
thesis -> mechanism -> lever -> one step
- thesis: у нас уже есть не просто список biomarker-ов, а многослойная логика отбора; проблема не в отсутствии мысли, а в том, что provenance распределен по разным артефактам и не везде одинаково strong.
- mechanism: разводим 4 слоя происхождения решений:
guideline/evidence,expert/ops,founder/cohort fit,health-system fit, а затем смотрим, что из этого реально попало вv1и что изменилось вv2. - lever: это дает честную картину: где мы стоим на твердых official anchors, где на рабочей product logic, а где у нас явный research debt.
- one step: использовать этот review как board-prep перед
top100_biomarker_scorecard_v2.csvиtop100_biomarker_board_decisions_v2.md.
1) executive read
что у нас уже было
v1строился какcore-70 + gated-30для mostly healthy preventive cohort.- база логики была правильная:
- hard gates
- score model
- retest discipline
- preference for reproducibility over prestige
что стало ясно при пересмотре
- для
Healthи high-stress founder cohortv1был слишком generic. - цена в scoring мешала честно поднять полезные, но более сложные маркеры.
- endocrine / recovery / HPA / vascular mechanics / skin-hair layer были недоэксплицированы.
- часть
v2сейчас стоит на сильной product/system logic, но еще не полностью закрыта official-source provenance.
что получилось
v2теперь зафиксирован как отдельный canon:ai/data/biomarker_top100_selection_framework_v2.mdai/data/biomarker_top100_rerun_protocol_v2.mdai/data/system_age_contribution_model_v2.md- exact
top-100теперь относится только кorderable lab biomarkers. derivationsиcompanion modulesвынесены отдельно.- hpa best-read lock сейчас такой:
cortisol_amcortisol_pmspot_urinary_cortisol_creatinineurinary_metanephrines_fractionatedurinary_3_methoxytyramine24h urinary free cortisolоставлен только как fallback / confirmatory path.
2) where the current logic actually came from
A. guideline / official evidence layer
главные anchors лежат в:
- ai/data/sources.md
- ai/data/panel_evidence_matrix_v1.csv
что там уже strong:
- lipids / apoB / Lp(a) / CAC:
- C-101, C-105, C-122, C-131
- dysglycemia / glucose / HbA1c / OGTT:
- C-102
- kidney / eGFR / albuminuria:
- C-103, C-104
- thyroid caution / repeat logic:
- C-118
- vitamin D caution:
- C-119
- iron / ferritin:
- C-120
- liver fibrosis gate (FIB-4):
- C-121
- imaging boundaries:
- C-107, C-108, C-123, C-124, C-125, C-126, C-127, C-128
- vascular mechanics / PWV:
- C-133, C-134
- VO2max:
- C-129
B. expert / operational layer
главные anchors:
- ai/context/stanislav_skakun_biodata_call_2026_02_28_v1.md
- ai/data/first_panel_decision_packet_v1.md
что оттуда пришло:
- phased MVP instead of full-checkup from day 1
- baseline anchor:
- cbc
- expanded biochemistry
- hs-crp
- insulin / hba1c / glucose context
- tsh
- sex hormone block
- ultrasound-first and partner MRI/endoscopy
- reserve sample / reflex logic
- cIMT as adjunct, not standalone trigger
C. founder / cohort-fit layer
главный anchor:
- ai/context/founder_input_2026_02_22_25_v1.md
что оттуда пришло:
- target cohort = high-agency founders / operators
- raw data + API + provenance > black-box score
- core biomarkers ~100-130
- skin baseline + ecg/vitals + optional dexa/cgm
- "money is not primary"
- machine-readable longitudinal self-use as filter
D. health-system fit layer
главный anchor:
- ai/data/top100_blood_urine_health_audit_v1.md
что оттуда пришло:
- generic top-100 надо сверять с реальной Health truth-layer
- urine block оказался главным фактическим debt
- igf1 surfaced as serious add-candidate
- generic design было полезно, но не до конца закрывало endocrine/adrenal reading
E. equipment / companion-module doctrine
anchors:
- ai/data/mvp_equipment_philosophy_v1.md
- ai/data/vascular_intelligence_layer_v1.md
- ai/data/vascular_intelligence_ops_v1.md
что оттуда пришло:
- PWV как core vascular mechanics signal
- skin / dermoscopy / face / hair as machine-readable longitudinal companion layer
- hair as real endocrine / stress / nutrient-response surface, not cosmetic garnish
3) methodology evolution: v1 -> v2
| layer | v1 | v2 |
|---|---|---|
| cohort | generic preventive mostly healthy | high-stress / high-travel founders, men + women |
| score | bps v1 included cost_efficiency |
bps v2 removed price |
| count | biomarkers + derivations + non-lab items were too mixed | exact count = only orderable lab biomarkers |
| system-age | present but lighter endocrine weight | endocrine / recovery materially upgraded |
| hpa read | too flat / under-specified | dynamic read closed explicitly |
| vascular | blood + imaging boundary present | blood + PWV + carotid + CAC tied into one loop |
| skin/hair | present in equipment doctrine | now treated as required companion layer |
4) provenance map by cluster
| cluster | why it is in the model | primary provenance | status |
|---|---|---|---|
| cbc / hematology | baseline oxygen / marrow / inflammation context | framework logic + evidence matrix, but no clean hematology claim ids yet | medium / gap |
| cmp / chemistry / liver / electrolytes | intervention safety + organ context | evidence matrix, but chemistry anchors still partly gap |
medium / gap |
| glycemic core | strongest preventive longitudinal loop | C-102 + framework + system-age model |
strong |
| kidney / urine / uacr / cystatin | silent drift + serum-only miss protection | C-103, C-104 + audit |
strong |
| lipids / apoB / Lp(a) / CAC | causal vascular burden + reclassification | C-101, C-105, C-122, C-131 |
strong |
| thyroid | baseline sanity + repeat caution | C-118 |
strong for caution, medium for broad default use |
| iron / ferritin / TSAT | common false-action zone; needs context | C-120 + framework logic |
medium-strong |
| vitamin D | common panel item but not strong default-screening evidence | C-119 |
cautionary strong |
| liver fibrosis gate | cheap escalation logic | C-121 |
strong |
| imaging exclusions | avoid prestige/incidental theater | C-107, C-108, C-123, C-124, C-125, C-126, C-127, C-128 |
strong |
| VO2max | powerful fitness/mortality signal | C-129 + expert/founder tension |
strong signal, operationally mixed |
| vascular mechanics / PWV | closes vascular age loop beyond lipids alone | C-133, C-134 + vascular docs |
strong |
| skin / melanoma baseline | companion longitudinal surface layer | founder + equipment doctrine; repo has cautionary C-106, but not a strong routine-screening support anchor |
product-strong / evidence-gap |
| hair / trichoscopy baseline | companion endocrine/stress-visible layer | equipment doctrine + founder fit; no formal source registry anchor yet | product-strong / evidence-gap |
| endocrine / androgen closure | required for useful free-T and recovery read | founder fit + Health fit + framework v2; repo lacks strong direct claim anchors | important / evidence-gap |
| adrenal / HPA / SNS closure | critical for this cohort, absent from v1 logic | Health fit + framework v2; repo currently lacks strong source-registry anchors for spot urinary cortisol, urinary metanephrines, 3-methoxytyramine | important / evidence-gap |
| GlycA / holoTC / IGF-1 / apoE / NMR-LDL-P / sdLDL | promoted in v2 because they sharpen system-age or discordance reads | mainly v2 product/system logic; not yet fully backed by repo claim ids | medium / gap |
5) what we were really relying on
strong-evidence zones
- glycemic screening and OGTT reflex logic
- kidney / albuminuria logic
- lipid / apoB / Lp(a) / CAC hierarchy
- imaging exclusions and screening boundaries
- PWV as vascular mechanics signal
medium zones
- thyroid inclusion:
- strong caution evidence
- weaker routine-screening enthusiasm
- iron / ferritin / TSAT:
- useful, but context-dependent
- VO2max:
- strong signal, but packaging/utilization question remains
weak or missing-by-repo zones
- HPA / SNS best-read stack
- spot urinary cortisol / creatinine
- urinary metanephrines fractionated
- urinary 3-methoxytyramine
- IGF-1 in this cohort
- GlycA
- holoTC default policy
- apoE baseline-consent logic
- skin / hair companion layers as medical/protocol objects rather than premium UX objects
6) what v2 changed because of cohort fit
explicit additions
free_t3fasting_c_peptideigf_1cortisol_pmspot_urinary_cortisol_creatinineurinary_metanephrines_fractionatedurinary_3_methoxytyramineglycaholotranscobalaminapoe_genotypeapoa1nmr_ldl_particle_numbersmall_dense_ldl
explicit closure rules
- testosterone is not interpreted without:
total_testosteroneshbgalbumin- derived
free_testosterone - adrenal/HPA is not interpreted from morning cortisol alone
- vascular age is not treated as closed without
PWV
explicit counting cleanup
- out of exact 100:
free_t_calchoma_iruacrfib4non_hdlapob_apoa1tsat- kept outside count as companion modules:
PWV- carotid protocol
- skin mapping / dermoscopy
- face skin analysis
- hair diagnostics
- ECG
- DEXA
- VO2max
7) the HPA block we locked now
best-read block:
- cortisol_am
- cortisol_pm
- spot_urinary_cortisol_creatinine
- urinary_metanephrines_fractionated
- urinary_3_methoxytyramine
why this is the chosen shape:
- am + pm gives diurnal shape
- spot urinary cortisol gives a more integrated read without 24h collection fragility
- urinary metanephrines separate sympathetic / catecholamine load from pure cortisol story
- 3-methoxytyramine adds extra catecholamine-metabolite nuance when we want the fullest read
what we explicitly rejected as default:
- 24h urinary free cortisol
why:
- too much collection friction
- weak routine longitudinal compliance for this cohort
- bad default if the real product is repeatable tracking
8) what came out of the review
the good
- there was already a real selection engine, not a random lab wishlist.
- the repo already had good anchors for cardio-metabolic, kidney, vascular boundaries, and imaging exclusions.
- the move to
v2is not a vibe-shift; it is a coherent response to the actual intended cohort andHealthsystem-age use case.
the weak
- provenance is uneven.
- some
v2promotions are currently justified more by product/system logic than by explicit claim ids in the repo. - HPA / SNS / endocrine closure is directionally right, but still under-cited.
- skin/hair are clearly part of the product, but still under-anchored as research-backed longitudinal modules.
the actual result now
- we have a cleaner doctrine.
- we do not yet have the final exact v2 locked 100.
- the next artifact must be a scorecard + board decision pass, not another prose-only brainstorm.
9) what must be added next
highest-priority provenance debt
- HPA / SNS evidence anchors:
- spot urinary cortisol / creatinine
- urinary metanephrines fractionated
- urinary 3-methoxytyramine - endocrine / recovery anchors:
- IGF-1
- free T interpretation logic in preventive cohort
- DHT / AMH / progesterone timing rationale - v2 promoted markers:
- GlycA
- holoTC
- NMR-LDL-P
- small dense LDL
- apoE baseline-consent framing - companion-module provenance:
- skin mapping / dermoscopy as longitudinal protocol object
- hair diagnostics as endocrine/stress-response longitudinal object
operational debt
- exact cadence by cluster
- exact role-tag map (
guardrail / response / engagement) - exact top-100 count after removing derivations and companion modules
10) final read
- if the question is
did we have a method?-> yes. - if the question is
was that method evenly provenanced?-> no. - if the question is
does v2 improve the actual self-use / Health / founder-cohort fit?-> yes, materially. - if the question is
are we done?-> no; the next real step istop100_biomarker_scorecard_v2.csvandtop100_biomarker_board_decisions_v2.md.
source map inside repo
- methodology baseline:
ai/data/biomarker_top100_selection_framework_v1.md- methodology rerun:
ai/data/biomarker_top100_selection_framework_v2.mdai/data/biomarker_top100_rerun_protocol_v2.md- evidence registry:
ai/data/sources.mdai/data/panel_evidence_matrix_v1.csv- cohort / expert provenance:
ai/context/founder_input_2026_02_22_25_v1.mdai/context/stanislav_skakun_biodata_call_2026_02_28_v1.md- system fit:
ai/data/top100_blood_urine_health_audit_v1.md- companion modules:
ai/data/vascular_intelligence_layer_v1.mdai/data/vascular_intelligence_ops_v1.mdai/data/mvp_equipment_philosophy_v1.md