The agentic execution platform
for clinical trials.
A demonstration of the Damaros™ runtime on synthetic FHIR data. {{ captionHint }}
DMR-204 · EGFR-mutant NSCLC
NCT00000204 · Phase II · randomized 1:1 · hash aead45cf
{{ cur.name }}
{{ cur.meta }}{{ cur.prompt }}
Surfaces open protocols the site's evidence already covers. Aggregate only, no patient data.
{{ sn.open.summary }}
Drafts the amendment that unblocks a stuck criterion, routed for signature.
Surfaces quality signals early, while a correction is still cheap. Aggregate, never a verdict · not surveillance of people.
Every action on this trial is an immutable, hash-chained event. Luna reconstructs any finding from the chain in seconds and cites every row it used. It reads the record; it never writes to it.
Not a replacement.
One system for the stack.
Works with the software the site already runs.
Care is everywhere.
Clinical trials aren't.
Candidates surface. Enrollment is recorded. In between, the funnel goes dark.
higher trial enrollment at NCI-designated centers (21.6%) than at community programs (4.1%).
of US cancer patients are treated in community settings, about 80–85% across independent estimates.
of US counties had zero active cancer trials in 2022. Care is local; trial access often is not.
CRC turnover in FY2024 (15.5%). Sites are short on people; replacing one coordinator costs $50K–$60K.
Patients aren't scarce.
Proof is.
Site infrastructure a sponsor can verify.
A signed record of a probabilistic decision is not proof. Given the same clinical triage prompt ten times, a leading commercial LLM repeated itself only 49.3% of the time. Replay is reproducible by construction: same protocol, same evidence, same as-of, same verdict, every run.
Protocols now absorb a median 3.3 amendments each, up 60% since 2015; sponsors themselves call 45% of those avoidable. Trident recompiles each version into locked logic: re-screens run against the protocol that governs the site, not the one a coordinator remembers.
One run.
Protocol to proof.
Engine accelerates. Humans decide. Replay proves.
Trident compiles sponsor criteria into locked logic; one locked version runs byte-identical at every site.
Site-approved data arrives screening-ready as of the latest ingest, lineage and mapping intact.
Deterministic engine evaluates every compiled criterion; anything it can't evidence routes for review.
Staged options with cited evidence wait for one sign-off before the run closes.
Any run reconstructs end to end from protocol basis through final signature.
Four agents.
Zero verdicts.
Surface, prove, draft, flag. Never decide.
Surfaces open protocols the site's evidence already covers — protocols to sites, never patients to protocols.
Reconstructs any run from the chain, every row cited.
Compiles criteria into locked logic and drafts amendments.
Flags drift and deviations before they harden into findings.
Not a dashboard.
A decision you can defend.
State, evidence, and a signed replay bundle.
Epic / FHIR R4 · retrieved 06-20
Damaros™ helps
the site execute.
Site-hosted. Site-governed. PHI stays put.
Not a log you trust.
Proof you can replay.
The record a sponsor wants before selecting a site.
Every run is attributable.
PHI stays off the model.
Site-bounded controls, ready for audit.
No LLM touches patient data. The architecture, not a setting. PHI is bounded at the site and never enters a model path. Only a third of published LLM-vs-human clinical comparisons favor the model; most rest on non-real data.
We validate the execution chain. We do not claim clinical efficacy.