01
AI now generates computational claims faster than anyone can review them.
TRAIRC Technologies
Independent quality control for AI in drug discovery.
LP-Gate checks whether the evidence behind a computational claim holds up, before money moves on it.
Context
Why independent review matters
Computational claims are moving capital and decisions faster than traditional review can keep up.
02
Most results are evaluated by the same team that built the model.
03
A strong score can hide what was never checked.
LP-Gate
What LP-Gate evaluates
Nine dimensions of evaluation quality. Each is evaluated or marked not evaluated.
Data integrity
Whether source data remains intact and fit for the claimed use.
Entity integrity
Whether compounds, assays, and labels stay correctly identified.
Leakage
Whether test information improperly influenced training or scoring.
Split quality
Whether train and test partitions support the claimed generalization.
Applicability domain
Whether predictions stay inside the space the model can support.
Provenance
Whether the evaluation trail can be reconstructed and audited.
Calibration
Whether confidence and scores match observed reliability.
Baselines
Whether claimed gains are measured against meaningful controls.
Robustness
Whether findings hold under reasonable perturbation and re-check.
Each dimension is evaluated or marked not evaluated. Missing evidence is never turned into a pass.
Output
The Validation Passport
What was checked, what wasn’t, and what couldn’t be evaluated — with evidence coverage reported.
A passport makes the evaluation transparent. Strong scores on checked sections do not erase missing evidence. Coverage is reported separately from outcome quality.
Each record states how it can be checked. Some can be verified in the browser against our published public key; scientific certificates are verified by TRAIRC. A valid signature confirms a record is authentic, not that a result is scientifically correct.
Incentives
Independence
Assurance only works when incentives are clean.
We don’t build or sell models
TRAIRC’s work is quality control for computational evidence, not competing model development.
Fees never depend on the verdict
Commercial terms are not tied to a favorable outcome of an assessment.
No strategic investment conflict
No strategic investment from companies whose claims we assess.
Calibration
Evidence so far
Public calibration work, including a signed correction when our own review found a problem.
TRAIRC published a public retrospective calibration of LP-Gate on public computational drug-discovery data, with a signed correction when an unauthorized deviation from precommitment was identified.
These results are an internal production calibration on public data. They are not independent external reproduction, prospective blind validation, or proof of general model correctness.
Audience
Who it’s for
Teams that need to know what the evidence actually supports before they act on it.
Pharma business development and licensing
Evaluate outside assets with clearer visibility into what was — and was not — checked.
Investors doing diligence
Separate strong evaluated claims from strong scores that hide missing evidence.
R&D teams receiving vendor results
Pressure-test partner and vendor computational packages before they shape program decisions.
Teams running AI agents on discovery work
Add an independent quality-control layer when automated systems accelerate claim generation.
Request a briefing
Methodology briefings, live demonstrations on synthetic data, and design-partner planning are available now.
Email george@trairc.comPlease don’t include confidential data in your message.