How Signals turns a question into research.
Start with a brief, compare model output, check the evidence, and keep the work ready for the next decision.
What stays with a signal.
A Signal keeps the claim, structure, model agreement, and evidence together. Reports can point back to the work behind the claim.
Claim
The thing being watched is stated plainly, so the reader can challenge the finding before trusting the system.
Structure
Scores and categories make the Signal comparable across a Session instead of leaving it as prose.
Convergence
Independent model runs stay visible, showing where the finding came from and where agreement formed.
Evidence
Sources, freshness, verdicts, and grounding keep the trust state inspectable after the report is written.
What is changing across AI-driven drug discovery platforms, GLP-1 follow-ons, and the shifting economics of clinical trials?
Trial sponsors utilize historical patient data to replace traditional placebo groups in metabolic studies. Signals immediate cost reductions in trial execution and improved patient recruitment metrics.
“Synthetic control arm adoption”
GPT-5.4
“Synthetic Control Arm Adoption”
Gemini 3.1-Pro-Preview
Converged across 2 of 32 frontier models on “Biotech platform shifts.”
What the method gives your team.
One brief runs across several models and is checked against sources.
Models work without seeing one another's answers. People review the record before accepting a first pass.
Repeated signals show agreement. Standalone signals stay visible as possible outliers or early warnings.
Scan checkpoints
- 01
Brief set
Question and boundaries recorded.
- 02
Candidates collected
Independent model passes complete.
- 03
Agreement grouped
Repeats merge; outliers stay visible.
- 04
Sources judged
Every source carries a verdict.
- 05
Trust recorded
Each Signal has a grounding state.
- 06
Signal set ready
The radar is ready to inspect and reuse.
Start with a brief your team can agree on.
A Session opens with a Briefing: the objectives, scope, and constraints that guide the work. An assistant helps you draft it, then it stays attached to everything the Session produces.
The brief keeps every signal, source, and report tied to the same question. Your team can always see why the research was scoped that way.
Signals stay useful because they have structure.
A Signal is a candidate indicator of change with a title, summary, category, tags, scores, sources, and a grounding state. It can stay ungrounded, or your team can verify or reject it.
That structure makes the research reusable. Your team can merge duplicates, group signals into Frames, score them against Metrics, and collect them across Sessions.
Each source gets a verdict.
A Source is more than a link. It records whether it supports, contradicts, is unrelated to, or is unreachable for a Signal. Your team sees how the evidence relates to the claim.
A source that contradicts a signal can be as useful as one that supports it. Both stay in the record.
Verified and rejected are explicit states.
After weighing its sources, a Signal is Ungrounded, Pending, Verified, or Rejected. The outcome includes a confidence and rationale, so your team can see how it was reached.
A Verified outcome without a supporting source returns to Pending. Re-running an evaluation checks the current evidence again.
Self-correcting: a Verified outcome with no supporting source is automatically rewritten back to Pending. Terminal states are never sticky. The latest evaluation always wins.
Compare model output instead of trusting one answer.
Tasks and report types can use more than one model. When they agree, your team can see it. When they disagree, you know where to look closer.
No model is treated as ground truth. Every generation is recorded against the Workflow Run that produced it.
Keep a method your team can use again.
A Workflow is an ordered set of Tasks in a Workspace. Run it for a first pass, then run it again as the picture changes. Every Session inherits the same shared structure.
Because the structure lives in the Workspace, each new Session can deepen the same theme without starting cold.
Tune verification to your stakes.
You decide how strict grounding needs to be. Require peer-reviewed or official sources for regulated work, or allow broader sources for exploration. Every verdict stays reviewable, and your team can override, add, or unmerge by hand.
See the method on your own question.
Bring a research theme and we will run it.