WHO MAY USE IT
Consent & permitted use
An explicit consent receipt, a defined recipient class, and a machine-readable license. Repository rights and organization permission are eligibility requirements.
For data buyers · Early conversations
Good data has
a backstory.
Explore a planned source of coding trajectories with explicit consent, traceable model calls, and evidence of what the work achieved.
TRAJECTORY MANIFESTEXAMPLE
OTR-TRAJ / v1 proposed
A record you can inspect.
Sessionexample_session_014
ConsentExplicit · receipt attached
ProvenanceRouter + device evidence
OutcomeTest run evidence
LicenseRecipient & use constrained
FormatVersioned manifest
Example structure. No real session or dataset is represented.
The evidence stack
More than a conversation transcript.
A useful coding trajectory connects a request to the work around it and to a measurable result. Each layer carries its own evidence and its own limitations.
WHERE IT CAME FROM
Request provenance
Router evidence and device-signed context are designed to connect the submitted record with the model calls that actually passed through the gateway.
HOW THE WORK HAPPENED
Work context
Changes, tool events, and task structure help explain how a result was reached. Sensitive or unauthorized content must pass separate exclusion checks.
WHAT THE WORK ACHIEVED
Outcome evidence
Tests, CI, or other independently sourced checks provide a stronger signal than an agent saying it is done. Evidence levels are labeled instead of flattened into one score.
Designed for real diligence
Verify the record.
Understand the lot.
The planned buyer experience puts manifests, schema contracts, license restrictions, and acceptance criteria alongside the data itself.
Versioned deliveries
Canonical trajectories and manifests, with proposed JSONL and Parquet exports. Schema changes should be explicit, documented, and testable.
Measured quality
Published sampling methods, defect categories, and confidence intervals. No quality percentages are presented here as measured results.
Rights that follow the data
Permitted uses, recipient classes, retention, and takedown obligations attached to each delivery. Evaluation-only access and clean rooms remain later-stage options.
Supply begins with permission.
The marketplace is not open. Dataset availability depends on developer consent, supplier terms, verified rights, and release readiness. We’re speaking with prospective buyers to shape useful, responsible data products.
Can I buy a dataset today?
No live catalog or available inventory is claimed in this preview. Early conversations focus on task categories, outcome evidence, format requirements, permitted uses, and acceptance criteria.
Can a buyer also be a model supplier?
Yes, the design supports that possibility, with separate permissions and data boundaries. Being a supplier would not grant access to competitors’ traffic or broader contribution data. Every delivery still requires an eligible license.
Do provenance records prove code quality?
Provenance establishes evidence about origin and integrity. It does not establish correctness by itself. Test results, CI, and other outcome evidence are separate layers and must be assessed on their own terms.
Which commercial formats are being explored?
Versioned datasets and filtered lots are the starting point. Streams, funded campaigns, evaluation-only holdouts, and clean-room access are roadmap concepts with separate technical and contractual requirements.
Build with us
What would useful evidence look like?
Tell us about your training or evaluation problem, the task categories you need, and how you would measure quality.
