Digital Labs
Session, capture and review in one thread, on iPhone, iPad and Mac.
What it is
A native, local-first laboratory work companion. The unit of work is a bench session: planned work, a version-pinned protocol, steps, timers, materials, instrument context, captures, observations, deviations and review state, held together so an interrupted run can be picked up again.
The camera is treated as a laboratory instrument rather than an attachment picker. A capture selects the task and the sample, shows which model package would run and where it would run, and records every outcome — including an abstain, a capture-gate failure, and the case where no approved model exists.
Who it is for
Bench scientists and technicians moving between bench, instrument and desk, often with gloved hands and poor lighting. Lab managers, research leads, data stewards and quality reviewers who need provenance and a review boundary. Automation engineers who need the record to leave the app intact.
How it is governed
A model result is a proposal. Review shows the confidence beside the package threshold and the abstain band, and a human accepts, corrects or rejects it; accept and correct write a typed scientific value, reject writes none. No threshold value permits automatic acceptance. A model package is refused if its bytes do not hash to the digest its manifest declares, if the file it names is missing, or if its effective window has passed — each with a stated reason. Counts carry their countable window; derived concentrations carry their factors.
What it does not do
It is not a laboratory information-management system, an electronic laboratory notebook, a validated record system, or an autonomous decision engine. It does not control instruments, make a clinical diagnosis, or approve science on its own. It makes no compliance claim: intended use, technical controls, validation evidence and operating procedures have to be approved together before anyone may say such a thing.
Status
In development. The foundation slice runs with a deterministic preview vision provider, clearly labelled as simulated, so provenance and mandatory human review can be exercised before a camera and an approved model are connected.
Related
VisionML produces model packages of the kind the client verifies. VisionAI shares the discipline of returning a measurement with every verdict. The Digital Labs platform — the API and the review web application — is built separately and is not part of this site.