Results and reproduction

The Results center is a directory of nine research projects and training programs. Open a project to read its goals, method schematic, evaluation design, planned milestones and participation roles. The initial stage is Training preparation, with applications open; browsing or applying does not start computation.

Join a research project#

Connect an existing wallet and sign a login message to associate an application with your account. Choose at least one participation role and one contact channel, then confirm use of that information for project coordination. Device details are optional and self-reported. No transfer, deposit, model download or reward authorization is part of the application.

The server returns an application ID, receipt time and review status after persistence. Use My applications to revisit, edit or withdraw a record. Resubmission retains its identity and change history. If a response is interrupted, retry the pending request or inspect your applications before changing it. Switching wallets clears private drafts and receipts.

Interest, applications, selection and first preference are separate. Counts represent authenticated accounts, not independently verified people. Active application counts include submitted, reviewing, waitlisted and selected records; rejected and withdrawn records are excluded. Missing statistics remain unavailable, rather than becoming zero. The community coordinates batches and device matching after review; selection itself does not launch training.

Published scientific records#

Published research records retain explicitly authorized observations, evidence and reproduction packages. These are distinct from project plans and from the Explorer's competition and financial history. A reproducible calibration example does not establish a completed behavioral capability or a current round win. Historical result links and authorized downloads retain their original scientific identities.

Read a result in order#

1. Read the purpose and declared scientific scope.
2. Check the dataset, graph, model and parameter identities.
3. Read the stimulus, control/backgrounds, duration and evaluator.
4. Inspect paired A/B metrics and off-target scope.
5. Read the verification evidence, limitations and source/license disclosures.
6. Inspect permitted downloads and the explicit reproduction action.

Do not treat the figure alone as the complete result. Raw events, compact summaries and display samples are separately identified.

Read, verify and reproduce a publication

1. 01Read the scopeInputs, evaluator and limitations
2. 02Check the packageAuthorization, manifest and file hashes
3. 03Run explicitlyMatching model and complete extent
4. 04Compare evidenceNumerical agreement within the declared scope

Scientific identities#

SimulationID binds dataset, graph, model, parameter roots, numerical rules, initial state, canonical stimulus, actual seed, time step and run extent. Wallet, filename, display title, downstream readout and outer round are excluded.

ReadoutID binds a deduplicated sorted simulation set, response/off-target roots, windows, evaluator and eligibility. It allows a different interpretation of existing simulation output without pretending that the underlying execution is new.

ContributionID describes who contributed what under a task, budget and authorization. AcceptanceID records the separate frozen acceptance judgement and evidence. These identities avoid circular definitions and duplicate rights.

Download a reproduction package#

Use the result's declared download links. Delivery enforces the allowlisted file, permission/expiration, MIME, size and SHA-256. Range recovery allows supported downloads to continue without requiring another complete transfer.

Keep the manifest and verify downloaded hashes before execution. Read the package's prerequisites, model extent and expected outputs. Reproduction may require the complete model and significant resources; browsing the evidence does not automatically acquire it.

Run reproduction explicitly#

A reproduction action is an intentional compute operation. It is distinct from selecting a historical observation. Use the matching model and frozen parameters, preserve existing caches and compare the required outputs under the package's declared contract.

Successful reproduction confirms numerical agreement for that package. It does not automatically establish a new discovery entitlement, current competition win or external biological accuracy.

Publication and privacy#

Private candidates, salts, archive paths and customer tickets are not public result material. Publication requires its actual authorization. A public projection or source repository cannot be used to bypass private package access.

Public counters deduplicate the relevant scientific identities. Downloads are download events, not user adoption; result count is not the count of independent experiments in living organisms.

Connect to payment evidence#

For a rewarded unit, follow its candidate and unit into Explorer. The on-chain 1,024-byte Result is a canonical settlement summary. It is not the complete scientific package. Explorer guide and Economic model explain how scientific and financial evidence stay linked.