The obvious way to help someone start a clinical protocol is to ask them questions. Primary endpoint? Eligibility thresholds? Sample size? It feels safe and collaborative — and it's exactly the wrong place to begin. Asking one question at a time just rebuilds the blank page one row at a time, and it forces decisions before there's any draft to react to.

We're building the opposite: hand Asthra a one-line indication, and it drafts the whole synopsis forward — then puts every design decision in front of you, cited, to keep or swap.

The inversion: from retrieve-and-ground to seed-then-expand

Until now, Asthra has worked the way most document automation does. You assemble a library — the investigator's brochure, the statistical analysis plan, prior study reports — and the system drafts by retrieving from and grounding in those sources. Rigorous, traceable, and exactly right when the source material already exists.

At the earliest stage of a program, it usually doesn't. You have an indication and an idea, not a library. So we inverted the model. Instead of retrieve-and-ground, the new flow is seed-then-expand: you start from a bare indication, and Asthra generates forward from there. The natural first artifact is the protocol synopsis — the concise, decision-dense summary that anchors the study — and that synopsis becomes the seed for the full protocol.

Registries as the evidence base — not the open web

When there's no internal library, Asthra draws on the public, citable record of what comparable trials have actually done. This is a deliberate complement to closed-system drafting, not a departure from it: your confidential sources still never leave your perimeter — the seed flow reaches outward only to authoritative public registries, precisely because there is no library yet, and every claim it pulls in arrives with a citation.

Two sources do most of the work, in order:

ClinicalTrials.gov first. For the parts of a synopsis that are genuine design commitments — endpoints, eligibility, sample sizes — nothing beats the registered record. Asthra searches trials in the same indication and phase and synthesizes the recurring, regulatorily-accepted choices rather than copying any single study.

PubMed second, for disease background, epidemiology, and standard of care.

Every claim is captured with its citation and rolled into the bibliography automatically. When a reviewer asks where an inclusion threshold or an endpoint came from, the answer is attached to the text — not reconstructed after the fact.

Auto-draft what's known; guide what's decided

Not every section should be written the same way. Sections that report established knowledge — background, rationale — are drafted from a light pass over the literature: a few high-signal references synthesized into a brief narrative. Sections that make design commitments — objectives and endpoints, eligibility, trial design, intervention, statistics — are handled as evidence-guided decisions: Asthra weighs what comparable trials did, chooses the best-supported option, drafts it, and shows its work.

The principle is one line: auto-draft where a section summarizes what the field already knows; guide where a section commits the trial to something consequential.

Draft-through, not block

So Asthra drafts the whole synopsis end-to-end, then surfaces every design commitment as a reviewable decision. Each carries the chosen value, a one-line rationale, its citations, and two or three cited alternatives from comparable trials. The writer reviews a concise panel and swaps any choice in a single action.

Crucially, swaps cascade. Change the primary endpoint and the sample-size justification and the analysis approach re-derive to match — because in a real protocol those things are not independent. Reviewing a confident draft and steering it is far faster, and far truer to how experienced writers work, than answering a checklist cold.

And intake is open-ended. Hand Asthra a single sentence — "a Phase 3 trial of an oral agent in adults with type 2 diabetes" — or a fuller picture with population, comparator, and target outcome. Whatever's missing, it fills with a conservative, cited assumption and flags it for review rather than blocking to demand it. The thinner the seed, the more it surfaces for your attention. Either way, you're never stuck at a form.

What it looks like

Give the flow this seed: Phase 3, type 2 diabetes, adult patients, oral investigational agent, placebo-controlled.

It returns a complete synopsis: a brief, cited background on disease burden and unmet need; a randomized, double-blind, placebo-controlled design; synthesized inclusion and exclusion criteria reflecting recent registered trials; an intervention and comparator section; a high-level statistical plan with an estimated sample size; and a schedule-of-activities overview.

Alongside the objectives sits a decision the writer can act on:

Primary endpoint — chosen: Change from baseline in HbA1c at Week 24

  • Rationale: the standard registrational endpoint for Phase 3 type 2 diabetes trials.
  • Evidence: multiple comparable registered Phase 3 studies.
  • Alternatives: Proportion achieving HbA1c < 7.0%; Time to initiation of rescue therapy — each with its own supporting trials.

Swap in the responder endpoint, and the sample-size assumptions and analysis approach update to match. The work shifts from authoring to deciding — which is where a medical writer's judgment belongs.

Standards at the core

The synopsis is structured to the ICH M11 protocol template — the harmonized format regulators across major regions have committed to — so the output maps cleanly onto the protocol it will become. Objectives are paired with endpoints in the spirit of the ICH E9(R1) estimand framework, so the scientific questions are stated precisely from the first draft.

Where this is going

The synopsis is step one — deliberately the seed for the next artifact. Once approved, it grounds the full protocol, where the same evidence engine drafts a literature-backed introduction and the body sections expand from the design the synopsis locked in. Indication to synopsis to full protocol: a cascade, each stage feeding the next.

The goal isn't to replace the regulatory writer. It's to hand them a confident, cited, standards-aligned starting point in minutes — and make every refinement fast, so the expertise goes into the decisions that matter, not into the blank page.