Methodology

Where an organiser has published a result or a standing, RoboPodium reproduces it faithfully, including when we would have ranked things differently. Those outputs carry an Official label.

RoboPodium also computes its own views. Those carry a RoboPodium-derived label, a named methodology version, a cutoff time and a fingerprint of the exact inputs, so any figure can be recomputed and checked. A derived view is never presented as an official competition result.

Robot sport mixes autonomous, teleoperated and hybrid control, and organisers do not treat that consistently. WHRG football is autonomous-only; WHRG free combat ranks autonomous, teleoperated and semi-autonomous entrants in one cohort for one medal; the WHRG relay has a robot revert to manual control mid-race after handing over the baton.

RoboPodium records four layers: what the rules permit, what a competitor declared, what mode was actually used, and, only where control genuinely changes within a contest, the individual segments. Where the mode is not established, it is recorded as unknown. Unknown is preferred to a confident guess.

Where an organiser ranks control modes together, so do we. We may additionally publish a segmented view, clearly marked as our analysis rather than the competition’s verdict.

The organiser-defined competition entry is the primary sporting identity. A team, an organisation, a robot model, a physical machine, an operator and a software stack are distinct things and are never collapsed into one another.

A persistent physical robot identity is created only where evidence supports it: a serial, a persistent organiser identifier, an explicit team attestation or a uniquely corroborated marking. A recurring name is not evidence. In WHRG football many machines are organiser-loaned and pooled between stages, so for most competitors no physical robot identity exists at all, and the database refuses to invent one.

• Composite results are stored in parts. A WHRG athletics time is the finish time plus accumulated penalties; we store the raw time and every penalty with its rule reference, so the official figure can be rebuilt and checked rather than merely repeated.
• Field-event attempts are individual records, including failed ones. Best mark and countback order are derived from them.
• An administrative scoreline, such as a walkover or an awarded result, stays official and counts in the table, but is marked by origin so it never contaminates played-goal analysis.
• Results carry a publication state. A result can be provisional, official, amended, disputed, withdrawn or superseded. The closing of an appeal window does not make a later organiser amendment impossible, and the model does not pretend otherwise.
• Organiser terminology is preserved. The organiser’s own words are stored as published, alongside the normalised value and a reviewed English translation: 比赛中退 (retired during competition), 取消比赛资格 (disqualified), 成绩无效 (mark invalid) and 弃权 (forfeit / walkover).

Every material fact points at a locator inside an immutable, checksummed snapshot of a specific version of a specific document. Snapshots cannot be edited or deleted; a correction is a new snapshot and a new revision.

AI may extract, classify, translate and propose. It may not publish, merge ambiguous identities, assert physical robot continuity, recognise records or grant itself verification authority. That boundary is enforced in the database, not just in policy.