Methodology

Why bipedal?

Parity is a biological question, not a performance question. We are not tracking the fastest mechanism — wheels will beat legs at flat-ground speed indefinitely. We are tracking whether robots can match humans under human biomechanical constraints. The question we exist to answer is: "Can a machine built like us outperform us at what we do?" A wheeled robot winning a 100m sprint tells us nothing about that question. A bipedal robot running 9.57 seconds does.

Scope: Why World Athletics–Ratified Records

Bioparity anchors on the current World Athletics–ratified world records because that list is the highest sanctioned human performance ceiling we have. Those records also come with codified eligibility rules — wind thresholds, track surface standards, equipment compliance, anti-doping verification — that translate cleanly onto a robot attempt. A robot that beats Cheptegei's 12:35.36 over 5000 metres, on a standardized 400 m track, under legal conditions, has cleared the same bar a human did. The rules give us something rigorous to test against.

The ratified list also covers the full range of individual bipedal locomotion events — sprints, middle distance, long distance, hurdles, steeplechase, high jump, long jump — rather than the IOC's quadrennial, television-selected subset. The Games are a window into a subset of these records every four years; the records themselves are the continuous ceiling. We anchor on the ceiling.

Winter events and equipment-mediated events (pole vault, throws, wheeled disciplines) stay out of scope for the same reason as before: non-bipedal or equipment-dominant locomotion is outside the biological parity question. A robot that beats a speed skater is beating a human on steel blades, not on human legs. A robot that throws a javelin further than Zelezny is beating a human with a spear, not a human body. Neither tells us what this ledger is here to measure. Archery is the one exception already seeded — it is shoulder-and-eye biomechanics, not locomotion, but the test stays bipedal-only.

Scope: What's Tracked

Twenty-five events, split by discipline and gender:

Throws (shot put, javelin, hammer, discus), pole vault, and swimming are excluded because either no current humanoid bipedal robot platform has demonstrated capability in the modality, or the event is equipment-dominant rather than locomotion-dominant. When a real bipedal humanoid throws a real regulation implement, clears a regulation pole vault, or swims a sanctioned 50 m freestyle, the event will be added.

What Counts as an Attempt

Bioparity tracks capability parity, not head-to-head competition. A robot does not have to race a human to set a verified parity attempt. It has to perform the event, under sanctioned conditions, to the measurement standards of the event's governing body. Solo runs count. Timed trials count. Head-to-head races count. What does not count: simulation, wind-aided results outside legal thresholds, shortened distances, assisted propulsion, or any result where the robot's locomotion was substantially provided by a human or a wheeled platform.

The parity meter measures against World Athletics world records. An attempt that would beat the record but fails sanctioning rules (wind, surface, equipment, handler intervention) is recorded as ineligible and does not move the meter.

Every performance in Bioparity is also tagged with an autonomy level — autonomous, assisted, teleoperated, or unknown. Autonomy does not determine eligibility, but it does determine what a result means. A teleoperated robot finishing a half marathon is a very different result from an autonomous one. Both are tracked. Neither is hidden.

Three recording rules govern how large-field events, teleoperated entries, and approximate-format events are handled. See Recording Rules below.

Recording Rules

Teleoperation is automatically ineligible.

Any performance where autonomy is classified as teleoperated is automatically ineligible for parity meter purposes. Teleoperation does not represent humanoid capability parity because the locomotion decisions are being made by a human pilot, not the robot. Teleoperated entries are recorded for completeness — they document what machines and pilots can do together — but they do not move the parity meter.

Large-field events are capped at five entries per autonomy tier.

For sanctioned events with more than ten completing performances, Bioparity records the top 5 finishers per autonomy tier (autonomous, assisted, teleoperated, unknown) where verifiable times exist. The full field count is noted on the event detail page. This curation keeps the ledger usable without suppressing the shape of the field.

Approximate or category-mismatched events are not treated as parity attempts.

Humanoid competitions sometimes feature events that approximate but do not match sanctioned human events — for example, a “100m hurdles” event where the barrier specifications, gender category, or other sanctioning details do not align with World Athletics' ratified version. Bioparity may record such performances for completeness, but they are marked ineligible with the specific mismatch documented in the eligibility reason. The parity meter is computed only from performances that meet full sanctioned-conformance on the World Athletics version of the event.

Verified vs. experimental performances

Every human world record in the ledger is real and cited to its sanctioning body. Every robot performance is a real, documented attempt — no illustrative placeholders. The engine distinguishes status cases regardless:

• Parity — a robot within the metric-type epsilon, even if numerically better
• Robot Lead — outside epsilon and beating the human
• Fallback — best experimental row when no eligible+validated row exists
• Eligibility fail — a wind, surface, or equipment violation that would otherwise win
• Hard fail — non-bipedal, tethered, or terrain-mismatched submissions that never enter the ledger at all

A row graduates from experimental or unverified to verified when a contributor supplies a source_url from one of the recognized sanctioning bodies listed below. Real performances are welcome via PR; see /submit.

Hard fail constraints

A robot performance is rejected outright — never enters the ledger — if any of:

• locomotion_type is not bipedal
• energy_source is not integrated (tethered power and external pneumatics fail)
• terrain_match is not true (must compete on the regulation surface for the event)

Eligibility (independent of validation)

A performance that passes hard-fail can still be ineligible for record purposes. Each condition below independently sets eligible = false; the rejection reason lists every failing condition, not just the first:

• wind_speed_mps > 2.0 (null is allowed for indoor events — null does not fail)
• wind_legal == false
• surface_standardized == false
• equipment_compliant == false

Validation status

A separate axis from eligibility. Three values:

• verified — citation from a recognized sanctioning body. Required for primary ledger status.
• experimental — real but unsanctioned performances (e.g., exhibition events, lab demos). Surfaces with a "Fallback: experimental" flag when no verified+eligible row exists.
• unverified — pending or unsourced. Never selected for status.

Status precedence

Computed at read time. Stored values are inputs only — derived fields are never persisted to disk.

1. If no selectable robot performance exists, status is Human Lead (no robot attempts) or Human Lead (no eligible robot performance).
2. If |robot − human| / human ≤ ε for the event's metric type, status is Parity — even when the robot is numerically better.
3. If outside ε and the robot beats the human in the right direction, status is Robot Lead.
4. Otherwise, status is Human Lead.

Two parity meters

The homepage shows two denominators side by side, never one without the other:

• Primary: share of events robots have actually attempted that are at Parity or Robot Lead.
• Secondary: share of all tracked events at Parity or Robot Lead.

The primary alone overstates progress; the secondary alone understates it. Both are true; together they are honest.

Recognized sanctioning bodies

A robot performance is treated as verified only when cited from one of these bodies:

• World Athletics — World Athletics
• FINA — World Aquatics (FINA)
• ISU — International Skating Union
• IWF — International Weightlifting Federation
• UCI — Union Cycliste Internationale
• World Archery — World Archery Federation
• ISSF — International Shooting Sport Federation
• FIS — International Ski and Snowboard Federation
• IBU — International Biathlon Union
• IBSF — International Bobsleigh and Skeleton Federation
• FIG — International Gymnastics Federation
• World Rowing — World Rowing Federation
• Guinness World Records — Guinness World Records
• ProRL — Professional Robotics League
• Beijing E-Town Half Marathon — Beijing E-Town Humanoid Half Marathon
• World Humanoid Robot Games — World Humanoid Robot Games

Projections

Linear least-squares regression on compliance-valid + eligible performances. Requires at least 3 points. Suppressed when r² < 0.3, the trend regresses, projection lands in the past, projection lands beyond 2100, or status is already Parity / Robot Lead.

Confidence is reported as high (n ≥ 6, r² ≥ 0.7), medium (n ≥ 4, r² ≥ 0.5), or low. A projection is a curve through past data — not a prediction.