Evidence model

The refusals are part of the product, not gaps in it. When Livemore Bio cannot answer from admitted evidence, it names the missing measurement instead of producing a plausible-looking number.

Result fields

Field	What it tells you
`evidence_state`	What kind of result it is, for example `MODELED_ONLY`, `MODELED_MECHANISTIC` or `MODELED_STEADY_STATE`
`qualification_state`	Whether it has been qualified against independent measurements. Nothing is qualified today: results are `NOT_QUALIFIED`, and research comparisons wait at `REVIEW_REQUIRED`
`applicability` · `interpretation_limits`	Where the protocol defines them: the physiology and context the model covers, and what it does not cover
`controls`	Checks that run on every execution: zero-dose equals untreated, mass balance, conservation
`claims_allowed` · `claims_prohibited`	Where the protocol defines them: which statements the result can and cannot support
`calibration_authorized`	Always false until an independent evidence review authorizes calibration. No model calibrates itself
provenance	Input and source SHA-256 digests, the pinned model commit, the solver and its version, and the citation behind each parameter

Validation levels

Every claim states its validation level.

A visual demonstration is not scientific validation.
A simulation is not clinical validation.
A wet-lab signal is not human efficacy.

Telemetry from the LIFE Patch updates observed state only and never silently recalibrates a model. Calibration stays unauthorized until an independent evidence review allows it: no model calibrates itself.

Not built yet

The physical LIFE Patch. The B0 EVT design target has known electrical release blockers.
The Cendos wet lab.
Person-calibrated or context-validated models.
Production PHI handling.

How to evaluate Livemore Bio

Try to make it lie. Ask for a dose nobody measured, a compound with no pharmacokinetic source, or a twin with no admitted model. Each should refuse with a reason code and name the missing measurement.
Read the controls block. A zero-dose arm must reproduce the untreated arm exactly, and conservation identities must hold. These run on every execution, not on request.
Read evidence_state and qualification_state. Every result is a modeled state, and nothing is qualified. Nothing here claims to be validated.
Check the provenance. Every number carries its pinned sources, solver and citations.