Calibrated vs cited

Lab puts every numeric value into one of three buckets: cited, calibrated, or uncited. The pill colors on every row tell you which.

Cited (green)

The number itself comes from peer-reviewed literature. Examples:

• Caffeine half-life ~5 hours — primary source: Arnaud 2011
• CYP2D6 poor-metabolizer activity ~0.22× — primary source: CPIC 2020 guideline
• MT1 receptor binding affinity ~80 pM — primary source: Audinot 2003

For these values, the published study reports the number, and Lab uses it directly.

Calibrated (blue)

The number is a Lab calibration — an engine-tuned scalar that's there to make the simulation behave realistically, but isn't itself reported in any single paper.

This sounds suspect at first. Why would Lab make up numbers? Because some claims have no clinical-PD endpoint to anchor against. Examples:

• "Sleep raises growth hormone by 8 engine units" — no paper measures GH in "engine units"; literature establishes the direction (deep sleep triggers a GH pulse) and the order of magnitude (~70% of daily GH released during sleep), and Lab calibrates the scalar that reproduces that behavior.
• "Meditation lowers cortisol by 5 engine units" — Pascoe 2017 meta-analysis confirms the effect; the magnitude is calibrated to match the reduction reported across 45 RCTs.

Every calibration carries:

• Rationale — what literature backs the direction and range (often shown as a green pill next to the blue one — those green pills are the supporting cites)
• Owner — who tuned it
• Last reviewed — ISO date; reviews stale at 1 year, fail at 2 years

The blue pill marks the bucket. The green pills next to it tell you the calibration isn't arbitrary.

Uncited (red ⚠)

The number is a working estimate with no source on file yet. Lab tracks every red pill; the gate that runs in CI prevents new ones from being added without justification, and existing ones are tracked as debt to clear.

If you see a red pill on a value you care about, treat the prediction as approximate. We're working through the backlog systematically.

Why this matters

Most simulation tools either:

• Hide their internal scalars (you have no idea what's tuned vs cited), or
• Pretend everything is cited (and bury the tuned scalars in opaque "model parameters")

Lab does neither. Every value's bucket is one glance away. If you build a treatment plan on a row of green pills, you're standing on literature. If you build on blue, you're standing on Lab's best calibration. If you build on red, you're guessing — and Lab tells you so.