QUANDRA is decision support for biotech teams. Every response is re-derived symbolically, unit-checked, statistically recomputed, citation-grounded and adversarially attacked. When the checks disagree, it refuses instead of guessing.
6 free lab & biostat calculators — no account neededA language model proposes. Deterministic code disposes. Every channel runs outside the model and can veto the answer on its own.
Closed-form CAS re-derivation of every equation the model writes.
Dimensional consistency across mg, mg/kg, µM, IU and time bases.
Deterministic recompute of power, CI, TOST, survival and Bayesian assurance.
Claims matched against a pinned 21 CFR / ICH / FDA corpus or refused.
A second model attacks the answer before you ever see it.
REFUSAL — when channels disagree, the unit math fails, or a citation cannot be grounded in the pinned corpus, the system returns a reasoned refusal with a code. Every refusal is logged.
Patient parameters in, verified dose out — with the label citation and the safe envelope stated.
Trial design in, required n out. Flags underpowered arms and missing multiplicity correction.
LOD/LOQ, linearity, precision and robustness per ICH Q2(R2).
Reads a protocol and returns the statistical and regulatory holes in it.
Cpk/Ppk, content uniformity and dissolution f2 with confidence intervals.
Every answer we declined, with reason codes. The compliance story lives here.
2,000 curated worked examples and a pinned reference corpus you can diff.
Retrieval over regulatory text with chunk-level provenance.
Ad-hoc scientific questions through the same tribunal. Streamed live.
ICH Q1A/Q1E shelf-life regression with batch poolability F-test.
PRR / ROR / EBGM signal detection on disproportionality tables.
Bring the question your team argues about. Watch the tribunal either prove it or refuse it in the open.