5-level truth hierarchy.
Every normalized claim is assigned a single level, H1 through H5, based on the strength of the human evidence behind that exact claim. The level is the ceiling on how confidently an account can state the claim before it starts losing points.
01Levels
Replicated systematic-review signal, high-quality evidence synthesis with moderate-to-high certainty, effect relevant to hard human outcomes. Confident language is appropriate.
Examples: LDL causal for ASCVD · Statins reduce CV events in secondary prevention · Exercise reduces mortality · VO2max predicts mortality · Grip strength predicts mortality.
Multiple decent human studies or one strong trial, with certainty limited by inconsistency, indirectness, or precision. Assertive-but-hedged language fits.
Examples: NMN raises NAD+ in humans · Caloric restriction improves healthspan markers · Fish oil has no CV benefit in general population.
Human evidence exists but is observational, underpowered, conflicting, poorly replicated, or only shows surrogate outcomes. Hedged language required.
Examples: Metformin extends human lifespan · Biological age is meaningfully reversible · Time-restricted eating has metabolic effect beyond calorie reduction · 30ml olive oil daily extends lifespan.
Support is primarily animal, cell, pathway, clinician anecdote, or testimonial. Uncertain language required; anecdotes are not proof.
Examples: NMN slows aging in humans · Cold exposure extends human healthspan · Plasma exchange rejuvenates healthy people.
No credible support, or stronger evidence points the other way. Asserting the claim is penalized heavily; refuting it is the evidence-aligned position.
Examples: Young plasma rejuvenates aging · Seed oils are toxic in humans · Follistatin gene therapy reduces aging · Blueprint protocol slows aging beyond confounders · Resveratrol clinically extends human healthspan.
02Why a 5-level ladder
A 0–1 continuous score suggests precision we don't have. A 5-level bucket matches how clinicians already think about evidence strength and maps cleanly to a confidence ceiling for the speaker: H5 tolerates assertive, H4 hedged-assertive, H3 hedged, H2 uncertain, H1 basically don't.
03Scope matters as much as level
Every evidence card has a scope block specifying who,
what intervention, and what outcome the level applies to, plus
a not_supported_for list. Extrapolating from a narrow scope to a
broad population is a scoping penalty (component B), even when the
evidence level itself is high.
Example: H4 evidence that NMN raises NAD+ in healthy adults does not support the claim that NMN slows aging in humans — that's a different claim family with a different (H2) level.
04Certainty modifiers
Each card carries GRADE-style modifiers — risk of bias, inconsistency, indirectness, imprecision, publication bias — to explain why the level is what it is. These don't affect the score directly in v1; they're a transparency field for reviewers.