01
One specification
Describe the capability, content, context, quality, rights, budget, and delivery once. Sally translates it into contributor qualification rules.
Managed data procurement
Human data with the context machines cannot infer.
Sally is being designed to turn a precise buyer need into a rights-reviewed, quality-measured, reproducible data product—without asking a buyer to browse people's private libraries or negotiate with contributors one by one.
The buyer receives a finished data product
Not a directory of personal files, inconsistent listings, or unverified consent claims.
02
One evaluated data product
Inspect a representative sample, schema, limitations, quality report, and runnable example before commercial review.
03
One rights package
Review named purposes, allowed and prohibited uses, provenance, consent evidence, retention, revocation treatment, and known gaps.
04
One reproducible revision
License a finalized manifest with pinned assets, context, rights, quality reports, schema version, and checksums.
One workflow, four perspectives
Individual contributor
At matching, sees a platform-reviewed buyer category and purpose, why selected items may match, and what context is missing. Before any binding rights grant or inclusion, sees the verified recipient identity, exact purpose, terms, and allocation.
Data and ML team
Gets consistent episodes, distributions, a representative sample, data dictionary, machine-readable metadata, loader examples, benchmark guidance, and an immutable revision.
Legal and procurement
Gets one accountable counterparty, a permission matrix, provenance and consent evidence, prohibited-use controls, standard documents, delivery receipts, and a complete audit trail.
Business AI agent
Gets stable identifiers, structured data-product metadata, explicit availability, machine-readable use constraints, quality fields, version history, and deterministic request and delivery states.
The proposed procurement path
- 1
Define the model behavior, acceptance test, data requirements, use scope, and budget.
- 2
Privately identify candidate supply without disclosing a contributor library.
- 3
Invite contributors to answer only the questions and rights checks that affect qualification.
- 4
Assemble a candidate revision and disclose its real distributions, gaps, and limitations.
- 5
Evaluate a documented representative sample under evaluation-only terms.
- 6
Finalize one immutable revision, license it, deliver it, and record contributor allocations.
Checking buyer capability gates…