Evidence
that stands.

From draft protocol to locked data. Simulate in Antevra. Run in Actora.

I am here because I…

Antevra checks what a protocol can support before you commit to it. Actora runs the study on a record that can be inspected.

Talk to us (opens in a new tab)
The Astralith mark: four stones set square, records that are only ever added to; the star where they meet is the decision a person signs.

Two moments decide whether a study can be relied on: when the protocol is written, and while the study runs. Antevra works in the first. Actora works in the second.

Know before.
Prove during.

Most trial software starts when the study starts. The decisions that decide whether a study can be analyzed are made earlier, on documents. Antevra works before the trial. Actora works during it.

  1. AntevraBefore the trial
  2. Draft protocolExtract a typed study design, every value cited to the protocol text.
  3. Readiness reviewFacts, assumptions, evidence status and owned decisions. Rules can lower a status, never raise it.
  4. Synthetic dataAn SDTM-shaped dataset with a validation report and a traceability chain. No patient data.
  5. Protocol finalA gated Readiness Pack. Nothing exports until a person has decided; there is no override.
  1. ActoraDuring the trial · BE/BA
  2. ScreeningVolunteer to PI eligibility decision.
  3. EnrollmentConsent, randomization, withdrawal and replacement.
  4. Dosing and samplingEvery event captured in its window, and every sample tracked from draw to storage.
  5. ReviewQueries and sign-off by someone other than the person who recorded the data.
  6. Data lockE-signed, then standards export.

The two products are designed to connect through versioned data contracts. That connection is on the roadmap and not built yet.

One protocol fact,
followed to lock

Fact F-07 · draft“Blood samples at 0.5 h post-dose.”Protocol v0.3, §9.2 · no owner yet

First it lands as a sentence in a draft.

Fact F-07 · supportedCited to protocol §9.2 and SAP §6.1Owner: biostatistics · decision recorded

Then Antevra cites it and a person owns it.

SUBJ-0142 · P1 · T+0.50 hDrawn at +0.51 h · deviation +36 sRecorded by CRC · server time

Then Actora records it at the bedside.

STUDY-A · lockedE-signed data lock · Dataset-JSON v1.1Audit chain intact

Last, it is locked and signed.

Try to change
a signed record.

This is how Actora behaves, on a synthetic study. Signed clinical rows accept only appends; the database rejects the edit, and the attempt is written to the audit chain.

The database rejects a clinical write without its audit row. Each write adds a layer, chained to the one below; no layer can be pulled out.

Vitals · SUBJ-0142 · Period 1 · T+0.50 hSigned by the PI (demo)

Two products

Antevra Reviews

Facts and assumptions from the protocol, SAP and labels, each cited to a source revision.

How Reviews works

Antevra Simulations

An SDTM-shaped synthetic dataset from the protocol’s own design, with a validation report. No patient data.

How Simulations works

Actora Clinic

Check-in, dosing and every event captured in its window, with deviation seconds.

See the Clinic app

Actora Release

E-signed data lock, then CDISC Dataset-JSON, ODM v2.0 and casebook PDFs.

What Release exports

Shared rules

Two products, built to the same rules. Nothing is overwritten . Every action is attributed to a named person, with server time. A person signs the decisions that matter, and no administrator can skip them. A new protocol is configuration, never a code branch.

Why the protocol
window matters

76%

of protocols had at least one amendment, up from 57%. Tufts CSDD, 2024

23%

of amendments were judged avoidable. Tufts CSDD, 2024

260 days

on average to implement one amendment. Tufts CSDD, 2024

$535k

median direct cost per substantial Phase III amendment. Getz et al., 2016

E6(R3)

GCP in effect in the EU from 23 July 2025, raising data-governance expectations. Swissmedic

Software proposes. Rules decide. A person signs.

AI does the reading and the searching, so your experts spend their time on judgment. Follow one finding through all three steps.

  1. Step 1 · The model proposes

    Washout of at least 5 half-lives

    "…a washout of at least five half-lives…" SAP §4.1 · synthetic

    Proposed: Supported
  2. Step 2 · Rules decide

    A rule lowers it, never raises it

    SupportedPartial

    Rule: source authority. This source cannot support a final claim on its own. The rule is named, so anyone can inspect it.

  3. Step 3 · A person signs

    Decided by a named owner

    Clinical pharmacology lead · accepted with note

    Nothing leaves the system until a person has decided.

  • You choose the modelEvery model call goes through one place, including a model you host yourself.
  • Every stage names its modelA stage that did not run says so and names no model.
  • No training on your dataYour documents train nothing.
  • No accuracy figureNone has been measured, so none is quoted. Every claim is cited and decided by a person instead.
What is true today, and what we do not claim
AntevraActora
Where AI is usedExtraction, review findings, repair proposals, anomaly detectionAn in-app assistant whose help answers cite approved help topics
ControlThe model proposes; rules can only downgrade a status; a person decides; export is gatedAnswers cite approved help; a confidentiality guard; the data mode is paused until a live evaluation
ModelsOne seam with switchable backends, including a model you host yourselfProvider-neutral design
ProvenanceEvery stage records the model that answered; a skipped stage names noneEach turn can be replayed
Not claimedNo accuracy figure; no training on customer dataAI duplicate-volunteer matching is built but switched off

A guiding star,
set in stone.

Rakesh Muppa, founder of Astralith

The name

Latin astra, stars, and Greek lithos, stone. Guidance you can build on, and a record that holds its weight.

Where we stand

Antevra is built and demonstrable. Actora is built from the first volunteer to the e-signed data lock and open to three design partners.

Meet the founder

Two ways to start

For sponsors and R&D A protocol preflight Bring one protocol. See its register of facts and assumptions and a synthetic dataset. Book 30 minutes (opens in a new tab) For BE/BA CROs, three places A design-partner pilot One real BE/BA study, configured together, in exchange for weekly feedback. See the offer
Bring a protocol. We will show you both halves. A 30-minute call about your study and whether Antevra, Actora or neither fits. Talk (opens in a new tab)