Rates: Phase III $55,716 and Phase II $23,737 per day of direct conduct cost (Tufts CSDD, August 2024); Phase I and device studies are priced at the Phase II rate as a floor. Sponsor revenue protected uses about $500,000 per day for Phase III only (Smith, DiMasi and Getz, 2024). Query cost ranges from $28 to $225 depending on the check (meta-analysis of 20 Phase III trials, Society for Clinical Data Management, 2020). Figures on this page use a synthetic portfolio.

Only 1.7% of data points are changed by the queries raised against them (meta-analysis of 20 Phase III trials, Society for Clinical Data Management, 2020), and source data verification corrects about 1.1% of case report form data while costing up to a quarter of a trial's budget (TransCelerate, 2014). The failures that matter are the ones above, and they are rarely in the sample.

Nightingale keeps a 21 CFR Part 11 audit trail on every value in its own record, so a REDCap study can move to a compliant environment without rebuilding one. Part 11 is a property of a system and its procedures; the audit trail is the part that is hardest to reconstruct afterwards. System names come from USC's public resource pages; each connection is confirmed with the institution's own team before it is used.

None of the figures above is a Nightingale result. They are the size of the work today. A four-week pilot measures what changes on one study, with the study team signing off each number.

Scope is first visit to database lock. Site selection, feasibility, screening and referral stay with the tools already doing them: TriNetX, All of Us and LADR for cohort discovery, and the Ryght Research Network for site matching and referral. Week 1 uses documents only; no patient data touches anything.

Studies are real and public; every operational figure is simulated.

Four of these six studies run with no interface at all. The engine is the product; the console is optional, and it is chosen study by study. Sponsors and contract research organisations that already have a system of record normally take the embedded route and never see a Nightingale screen. Sites and academic coordinating centres usually want the console, because observability across their own portfolio is the thing they are missing.

Query rate is queries raised per 100 case report form pages. The study-level average before deployment was 17.1.

Handwritten visit notes and scanned consent are read with optical character recognition, proposed as structured values, and held for human confirmation before any one of them enters the pipeline. Nothing reaches a study database on an unconfirmed read.

Sponsor expectation comes from the output specification, the protocol floor from the statistical analysis plan, and site commitments from each site's activation plan. Projections are 2,500 simulated runs of the weeks left before the cut, using each site's own completion rate by visit type (a beta-binomial model by site and visit type).

The required output set comes from the sponsor's data specification and the protocol's stated endpoints, not from what happens to be in the study database. That is why the gap can be non-zero on a study whose case report forms are 100 percent complete.

Every stage is a fixed rule approved in the rulebook. The same input produces the same output on every run, and the run is reproducible on demand.

Every stage above is a fixed rule approved in the rulebook. The same input produces the same output on every run, and the run is reproducible on demand. No value is inferred at run time.

Entries are append-only and cryptographically chained. An entry cannot be edited or removed, only superseded by a later entry that names it. This is what 21 CFR Part 11 requires of an electronic record, and it is written at the moment of creation rather than reconstructed at lock.

The same record states what would have had to be collected, or whom would have had to be enrolled, for a shortfall to have been avoided. That statement is written from the gap analysis history, not reconstructed afterwards.