AI EXPERIMENTATION AGENCY

Experiment command center

Research, run, compare, and price multi-treatment studies from one place.

LIVE STUDY · COMPLETE

Waitlist conversion — A vs B

swarm-2026-07-12T06-32-56-643Z-df7f0398
₹1,9991 treatment
RECOMMENDATION · ABLAZE-DECISION-V1
!INCONCLUSIVE

Sequence changes the answer.

B leads overall completion by 10 points and preference by four personas, but AB-order users lean A while BA-order users lean B.

LIVE PROOF

10 paired personas

20/20valid sessions100%paired success371msavg latency
PAIRED EVIDENCE
Variant A60%3 preferences · clarity 4VSVariant B70%7 preferences · trust 4
ORDER EFFECTAB group prefers A, 3–2BA group prefers B, 5–0
MANAGER PLAN

Research-grounded execution

ready
  1. 1Research audience contextLinkup · sourced jobs and objections
  2. 2Verify live surfacesDeployer · completed after revision
  3. 3Generate grounded panelPersona generator · 5 AB / 5 BA
  4. 4Run browser swarm20 isolated sessions
  5. 5Aggregate and decideEvidence-linked report
TOP FRICTION

What held personas back

8Not ready to share work emailA 4 · B 4
3Insufficient proof detailA 3 · B 0
3Value proposition feels abstractA 0 · B 3
L4 EVALUATION RELEASE GATE

landing-page-v1

CI enforced
3/3current score100%release floorBLOCKon regression
✓Every push runs the frozen setGitHub Actions · required gate
⊘Known regression rejected2/3 accuracy · release blocked
↗Evidence preserved on failurePassing and blocked artifacts
REPEATABILITY PROOF

Three completed live studies

gate passed
22/22valid pairs100%success rate$5 / $30OpenAI / 1M tokens
AIGPT-5.6 Sol pricinginput $5 · output $30 per 1M tokens
1Primary study10/10 valid · 0 retries
2repeatability-026/6 valid · 1.02s
3repeatability-036/6 valid · 1.22s
EMERGENT SPECIALIST

Segment deep dive

torn down
↗DYNAMIC ROLE

Triggered by order conflict.

The manager spawned a temporary segment interviewer from a persisted rule, stored its evidence-linked artifact, and removed the role after validation.