StochStack
← Back to Ventures

Protocol OS Lite (One-Page, Mock-First)

Complete demo flow in a single page with fake but coherent outputs. No backend database required.

Flow Overview

Step 1: Select therapeutic area and preload defaults
Step 2: Medical Need Framing (what unmet need and clinical intent)
Step 3: Endpoint + Estimand Alignment (objective, variable, intercurrent strategy)
Step 4: Eligibility Impact Simulation (recruitment / sample-size / timeline effects)
Step 5: SoA Burden Quantification + Operational Risk Gates
Step 6: Agent Review, Accept plan, compile full chapter draft and export

1) Therapeutic Area + Defaults

Scenario key: oncology|II|Global

2) Brief Inputs (pre-filled defaults)

3) Evidence Snippets (pre-filled defaults)

4) Agent Runtime + Plan A/B

5) Agent Logs (Functional + Role)

No agent run yet. Click Generate A/B.

Feedback Loop: Weight Change

MedicalNeedFramingAgent

weight: 1.00

accept: 0 | reject: 0

reject rate: 0% | samples: 0

strategy: normal

EndpointEstimandAgent

weight: 1.00

accept: 0 | reject: 0

reject rate: 0% | samples: 0

strategy: normal

EligibilityImpactAgent

weight: 1.00

accept: 0 | reject: 0

reject rate: 0% | samples: 0

strategy: normal

SoABurdenAgent

weight: 1.00

accept: 0 | reject: 0

reject rate: 0% | samples: 0

strategy: normal

OperationalGateAgent

weight: 1.00

accept: 0 | reject: 0

reject rate: 0% | samples: 0

strategy: normal

MedicalReviewerAgent

weight: 1.00

accept: 0 | reject: 0

reject rate: 0% | samples: 0

strategy: normal

StatsReviewerAgent

weight: 1.00

accept: 0 | reject: 0

reject rate: 0% | samples: 0

strategy: normal

ClinOpsReviewerAgent

weight: 1.00

accept: 0 | reject: 0

reject rate: 0% | samples: 0

strategy: normal

RegReviewerAgent

weight: 1.00

accept: 0 | reject: 0

reject rate: 0% | samples: 0

strategy: normal

Feedback Loop: Next-Run Comparison

Need at least two runs to show comparison. Give feedback, then click Generate A/B again.

Two-Stage Drafting

After accepting a plan, generate synopsis first.

Synopsis Attribution Console

Generate synopsis first to view section mapping, diff, and impact attribution.

Chaptered Full Protocol Draft (SH/TransCelerate-style structure)

Generate synopsis and click Expand to Full Protocol.

Export Package

Current mock graph version: v1

Product Update Log

v1.0.0 · 2026-03-04 · Synopsis Attribution Console

  • - Added section-level mapping from synopsis blocks to responsible agents.
  • - Added before/after synopsis diff with source-agent attribution.
  • - Added per-agent impact contribution panel linked to current scenario.

v0.9.0 · 2026-03-04 · Fully dynamic feedback policy (3 layers)

  • - Sliding-window profile recomputation on latest N feedback events.
  • - Scenario-specific calibration by TA/Phase/Region.
  • - Auto strategy switch (normal/conservative/backup) when reject rate crosses threshold.

v0.8.0 · 2026-03-04 · Visible feedback loop for agent calibration

  • - Added per-agent Accept/Reject actions with reject reason tags.
  • - Added dynamic agent weight updates and persisted profile counters.
  • - Added next-run comparison panel to show score shifts after feedback.

v0.7.0 · 2026-03-03 · Two-stage drafting flow (Synopsis -> Full Protocol)

  • - Added explicit two-stage generation: first Synopsis, then Expand to Full Protocol.
  • - Export now requires both stages to be completed to reduce premature draft handoff.
  • - Traceability export now includes synopsis snapshot plus full chapter content.

v0.6.0 · 2026-03-03 · Decision-Centric upgrade with 5 key design checks

  • - Added Medical Need Framing, Estimand Alignment, Eligibility Impact Simulator, SoA Burden Quantification, and Operational Risk Gates.
  • - Added functional agents and role-review agents with runtime logs in-page.
  • - Added optional Tongyi Qwen reasoning adapter with automatic local fallback.

v0.5.0 · 2026-03-03 · Chapter template engine

  • - Upgraded chapter drafting to template placeholders and SH/TransCelerate-style language.

Ready. Protocol OS Lite runs entirely with mock data in one page.