RESEARCH ENVIRONMENT
ST
SunderamMediator | India
Operator studio | seeded preview
Instrument control
Build evaluations, watch run health, inspect disagreements, and govern what the platform is allowed to claim.
Release authority
Gate 0 launch control
One global append-only decision controls new arena openings, invitations, participant offers, and attempt starts. The server—not this form—binds the named environment, Git release, Vercel deployment, deployment configuration, and India jurisdiction.
Gate 0 decisions are disabled in demo mode.
Use a separately configured staging or production deployment.
Run queue
Recent seeded activity
| Run | Participant | Evaluation | Time | Tokens | Status | |
|---|---|---|---|---|---|---|
| HL-0720-038 | meera.frames | Prompt Operator | 31:08 | 6,420 | In review | |
| HL-0720-037 | verify_first | Signal Check | 10:44 | 1,104 | Scored | |
| HL-0720-036 | analyst_zero | Analyst Arena | 86:12 | 14,880 | Evaluators resolving a disagreement | |
| HL-0719-035 | context_window | Signal Check | 11:52 | 1,318 | Scored |
Evidence pipeline
From idea to supported use
Brief, seeded states, reference run
Operational defects + participant signal
Agreement, stability, fairness
Ranking, hiring, or certification
Cohort runtime
Paid and invited arenas
Each arena is one immutable configuration version. Invitations name an account UUID; fixed offers use the separate compensation authority. This console never collects payout destinations.
No fictional arena operations.
Switch off demo mode to create a real private cohort.
Governed appeals
Blind review and immutable correction
The queue contains no participant name or direct identifier. Codex and Fable lock independently; their documents appear here only after both hashes exist and you reveal them.
No fictional appeal queue.
The seeded interface does not invent participant disputes or evaluator recommendations.
Participant rights
Privacy request queue
This queue may contain direct participant communication. Only the AAL2 privacy operator may advance it; model evaluators never receive it as scoring evidence.
No fictional privacy requests.
The demo never invents participant grievances, erasure requests, or operator responses.
Research governance
Pre-register, generate, audit
Every export is tied to one exact evaluation version, approved before its inclusion window, limited to separately consenting participants, and stripped of direct identity and free-text evidence.
No fictional research cohort.
Configured mode shows only governed purpose versions and real consent-limited snapshot facts.
Participant compensation
Fixed-offer payment queue
This ledger records payment facts only. The payout rail remains separately configurable, and no bank, wallet, or UPI destination belongs here.
No fictional payment queue.
The rehearsal does not invent participants, payout amounts, or payment receipts.