Public method
The scoring idea, evidence chain, failure semantics, and publication boundary remain inspectable.
Dronacharya · owned multi-sport evaluation
Evaluation with its context, available on request.
KhelVision owns the data, labels, code, training line, adapter, and evaluation artifacts behind the candidate. We do not publish context-free performance figures on the open web because a score separated from its ruler, panel history, operating point, and limitations is easy to misquote.
Disclosure boundary
The method is public. The figures travel with a conversation.
Evaluation results, including tolF1, are available on request. A qualified brief keeps aggregate and sport-specific results attached to the scoring protocol, data history, calibration status, limitations, and the release actually under discussion.
Available
Evaluation figures
Shared directly with suitable partners and reviewers together with the context needed to interpret them.
Serving identity
An accepted pilot identifies the release that actually served its run; candidate evaluation is not substituted for that fact.
Evaluation ruler
tolF1 scores source-aligned rally boundaries.
A human marks rally starts and ends against the authorised source timeline. The system predicts intervals on that same timeline. One-to-one matching prevents a prediction from claiming multiple human rallies, and a match must satisfy the agreed start and end tolerance windows.
F1 is calculated per recording and aggregated across the agreed evaluation panel so a long recording cannot silently dominate the result. Exact tolerances, operating points, panel construction, and reported figures are supplied in the request-only brief.
Scorer conformance is not model quality. A consistent tolF1 implementation does not prove representative data, media rights, venue generalisation, calibration, or the identity of a production serving release.
Context package
What accompanies a requested result.
- Metric contract
- Boundary matching, tolerance windows, aggregation, and failure treatment.
- Evaluation history
- Whether the panel was reused, adaptive, model-selecting, fresh, or independently held out.
- Operating point
- Locked comparison settings remain separate from calibration selected on development evidence.
- Scope
- Aggregate and sport-specific views are kept together with representation and imbalance notes.
- Release identity
- Candidate, custody, deployment readiness, and the release actually served are separate facts.
- Limitations
- Venue, camera, population, sport, rights, and workflow boundaries stay attached to any figure.
Owned evidence chain
Human labels remain the ruler.
The evaluation is grounded in owned, human-labelled rally boundaries with session-aware data roles. The current inventory is not treated as uniformly representative across sports, and fresh confirmation remains part of broader performance language.
Public Rally Annotator contracts make the label shape inspectable. Exact inventory, distribution, split roles, and quality-control findings belong in the contextual evaluation brief rather than in standalone marketing copy.
Release identity
Evidence, custody, and serving are separate statuses.
Evaluation retained
The owned evaluation artifacts and their provenance remain available for qualified diligence.
Candidate in custody
The evaluated candidate is preserved under internal provenance without exposing private correlation handles.
Exact release served publicly
Not claimed. Any assisted pilot must report its actual model, control-plane, and schema identity.
Responsible reading
What the public site can say.
- KhelVision has owned multi-sport development evidence for rally-boundary detection.
- The evidence supports qualified product discovery, not a universal benchmark claim.
- Sport representation, adaptive evaluation history, and calibration choices materially affect interpretation.
- Fresh-session confirmation and additional golden labels remain part of expanding the evidence base.
- Evaluation of a candidate does not imply that the same candidate serves a current customer run.
Request the complete brief: choose “Evaluation brief” in the pilot form and describe the sport, workflow, and decision the evidence needs to support. Do not submit footage or private links.