Human preference data with a protocol.
Scatha runs text-only contributor panels for AI personalization and evaluation: consent records, recurring sessions, QA review, and pseudonymous exports.
{
"participant_id": "SC_1042",
"consent_scope": "text_eval_only",
"session_type": "persona_fidelity",
"export_format": [
"jsonl",
"csv"
]
}A clean system for messy human signal.
Baseline profile
Communication style, preference tradeoffs, and boundaries collected before model testing.
Scenario tasks
Repeatable prompts that capture decisions, rewrites, explanation preferences, and drift.
Fidelity labels
Contributor judgments on whether AI-predicted responses sound like their preferences.
Export record
Dataset card, QA status, consent scope, and buyer-use restrictions attached to each cohort.
Built like a research instrument.
Scatha is the protocol layer for permissioned human preference data: scoped collection, documented consent, reviewable sessions, and export records that buyers can audit.
Confirms adult contributors and text-only task boundaries before access.
Stores scope, timestamp, policy version, and permitted use categories.
Collects structured text responses, ratings, rewrites, and longitudinal follow-ups.
Marks submissions valid, rejected, duplicate, or needs-review before export.
Packages approved records without direct identity fields for buyer evaluation workflows.
Your data is not your identity.
You choose what to contribute, you see the scope, and v1 is text-only. Scatha is not asking for photos, voice, face, fingerprint, medical details, sexual data, files, or passive identity collection.
No fuzzy promises.
Contributors license text contributions under documented terms. Scatha does not claim to rent, own, or reproduce a person. v1 excludes biometric and sensitive media collection.
Pilot-stage protocol.
Scatha is currently accepting limited pilot interest. Contributor payments and production data collection begin only after server-side consent records, review workflows, and payout controls are live.