Speed Climbing

What the model says, and what actually limits it

The route is fixed, so a record time implies an average vertical speed and nothing else is free. This app solves the other direction: given the published geometry and one calibrated elite body, how fast can the climb be, and which limit is the one that stops it being faster?

6.844 s — 6.714 s of climbing plus a 0.130 s reaction, for a 70 kg climber with an 0.80 m arm and a 1.02 m leg. The centre of mass rises 12.288 m at a mean 1.830 m/s.

For comparison the current world records are 4.54 s (Zhao Yicheng, 10 May 2026) and 5.99 s (Emma Hunt, 4 July 2026). The model's climber would have to have every capability scaled by ×1.52 to run the men's record and ×1.11 to run the women's.

The answer is not force, and it is not one thing

Raising each capability by 6% one at a time and re-solving gives these elasticities MEASURED — d(ln T)/d(ln p), so more negative means that knob buys more speed:

Neither leg strength nor hand grip appears in the answer at all. Leg isometric force has an elasticity of −0.008 and leg contraction velocity of +0.0005; hand-hold friction and hand-hold adhesion are 0.0000 to four decimal places. The structure underneath is a cross: what binds at the feet is the hold (friction and adhesion), and what binds at the hands is the limb (force and contraction velocity). A leg is never the thing that gives way — long before the muscle runs out, the foothold does — and a hand's grip on a big incut jug is never the thing that gives way either, because the arm behind it saturates first. Buying leg strength or stickier shoes for the hands would change this climb by less than a millisecond.

The second mechanism is the limb reconfiguration rate at −0.255. Of the 45 limb moves in the modelled ascent, MEASURED 11 are limited by nothing except the time it takes to throw the limb to the next hold, accounting for 1.441 s of the 6.714 s climb. Twenty-eight moves are limited by the contact cone (4.690 s), five are ballistic — three limbs cannot hold the body statically at all and its own momentum carries it (0.386 s) — and exactly one is limited by how hard a limb can push sideways (0.197 s).

Where this corrects what it was told, and what it corrects in print

The obvious question is: Is the limiter contact force, limb reconfiguration rate, reaction time, something else — and does your answer survive the sensitivity checks? The measurements say that framing is incomplete in two ways.

• “Contact force” conflates two different quantities. The force a limb can generate and the force a hold will accept are separate limits with opposite answers here, and they bind at different limbs: arm force matters (−0.209), leg force does not (−0.008); hand-hold grip does not matter (0.0000), foot-hold grip is the largest capability lever there is (−0.343). What actually binds on 28 of 45 moves is the hold's friction-plus-adhesion cone, and it binds at the feet. Naming “contact force” as one candidate hides the distinction that carries the result.
• There is no single limiter. Foot-hold adhesion (−0.343), foot friction (−0.301) and limb reconfiguration (−0.255) are three different mechanisms inside a 0.09 band, with arm force (−0.209) just behind them. A model that reported one of them alone would be reporting a rounding difference.
• Reaction time cannot be the limiter, by rule. It is 0.130 s here, 1.90% of the total, and IFSC 9.12(A)(1) makes anything under 0.100 s a false start. At most 0.030 s of it is available to be won, or 0.44% of the time.
• Body mass beats every capability. At +0.752 it is more than twice the largest capability elasticity. The fastest way to a faster time in this model is not to get stronger.

Corrections to published sources

• The IFSC route plan is not lettered A–K. DERIVED Wikipedia's Speed climbing wall article says each panel's 11 columns are “lettered A-K from left to right”. The IFSC Speed Licence Rules route table references columns L and M as well as I, so on an A–K reading 7 of the 31 hold references do not parse at all. The lettering is the 21-letter Italian alphabet, which has no J and no K: A B C D E F G H I L M. Read that way, all 31 holds land within 0.0 mm of an independently drawn Wikimedia Commons plan of the same route. Read as A–K, the 24 references that still parse leave at least 7 of the 31 holds on that independent plan with nothing on them at all.
• The IFSC document contradicts itself by 0.5 mm. DOCUMENTED Speed Licence Rules 1.1.7(ii) puts the starting hold at “mm 1 887”; clause 1.3.4 of the same 15-page document puts it at “1 887,5 mm”. The grid settles it: the lowest big hold is DX2 F1, panel 2 row 1, at 1500 + 187.5 = 1687.5 mm up the wall plus the 200 mm socle. 1.3.4 is exact; 1.1.7(ii) rounds.
• The 1 mm “error in the standard” is an error in the encyclopaedia. Wikipedia computes the panel height as (188 + 9 × 125 + 188) = 1501 mm and presents the 1 mm as a defect of the specification. The specification says 187.5 mm, and 187.5 + 9 × 125 + 187.5 = 1500.0 exactly. The standard is self-consistent here; the article's own rounding is not.
• A figure attributed to the specification is not in it. Wikipedia states the stop device's centre “has to be 13140 mm above the starting hold” and cites the Speed Licence Rules. The string 13140 appears nowhere in that document. Its only constraint on the finishing pad is the height of the pad's bottom edge, 14 800 ±20 mm above the ground.
• The record tables disagree with their own prose. The same Wikipedia article says the men's record is “4.58 seconds ... in April 2026” and the women's is “6.06 seconds ... in August 2024”, while its own tables end at 4.54 (10 May 2026) and 5.99 (4 July 2026). The prose is two records stale in both categories, and its counts (“broken 20 times”) disagree with its rows (23 and 23).
• One record row cannot be right. The men's table lists 4.798 s before 4.859 s, both on 12 April 2024. A record progression cannot increase. Both rows ship here exactly as published, with the pair flagged rather than silently re-sorted.

Where the published specification does not decide the answer

• The stop device has no position. The route plan gives a grid reference and an orientation for all 31 holds and leaves both columns empty for the 32nd row, DX 10 | STOP DEVICE. Its horizontal position, its size and its orientation are undefined. This app centres it between columns A and B of panel dx10 at the documented height and marks that RECONSTRUCTED.
• There is no climbing order. The specification fixes 31 positions and publishes no sequence, and the table's own order is not even monotone in height — SN9 10M is listed before two DX9 holds that are lower. The ascent order used here is DERIVED, and it uses 31 of the 31 holds across 45 limb moves, 19 of them hand moves.
• The 15 m is never said to be vertical or along the surface. Panels are “1500 mm ±1 mm” and the wall “must overhang regularly 5°”, but the two readings differ by 15000 × (1 − cos 5°) = 57.08 mm over the full wall. Only the in-plane reading lets a physical 1500 mm board satisfy the ±1 mm tolerance, so that is the reading used — and under it the start hold sits 6.42 mm below the 1887.5 mm the same document states, which is inside its own ±10 mm tolerance but twelve times the 0.5 mm the two clauses differ by.
• The false-start threshold is tested at a resolution the rules never fix. The Definitions page says Reaction Time is “measured to at least 1/100 s”; 9.15(B) requires the timing system to resolve 1/1000 s; 9.12(A)(1) tests against 0.100 s. A reaction resolved only to 1/100 s cannot decide that threshold — a displayed 0.10 is consistent with 0.095 and with 0.104. This app tests at 1/1000 s.
• Lane separation is stated twice, differently, in two places. The specification caps the gap between the two lanes at 500 mm; the encyclopaedia says “0 to 1000 mm”, which appears to have merged that clause with the separate 1000 mm of free space required outside each lane.
• Nothing specifies a competitor. No mass, no reach, no anthropometry appears anywhere in either document, so every body number behind the headline is CALIBRATED and none of it is documented. That is the honest status of the 6.844 s.

The convergence table behind that answer

Every number above depends on a discretisation, so none of it is reported without showing it survives refinement. MEASURED

Sampling converges to 0.027% between 24 and 96 samples per move. The pose solver is exactly iteration-independent, which it was not in the first version: annealing over a fixed sweep count made the answer oscillate by six per cent between 60, 120, 240 and 480 iterations because each count landed in a different local minimum. A discretisation that does not converge cannot be believed, and that one was caught only by tabulating it.

Two honest negatives. The torso segment count does not move the answer at all across a 64-fold refinement — every row is 6.84420 — because rotational inertia never becomes the binding constraint anywhere on this route. Saying so is more useful than quoting a converged number as if it had been in doubt. Limb stiffness converges upward to 6.790 s, so an infinitely stiff limb would be 0.8% faster than the modelled one; at 8 000 N/m the limb is so compliant that one move has no solution at all and the climb does not complete.

A third check does not pass cleanly, and it should not. Scaling every force capability by f² and every velocity by f is an exact time-rescaling of the whole problem except for gravity, which does not scale. The product T×f is therefore expected to drift, and it does: 7.238, 6.797, 6.844, 6.336, 6.086 for f = 0.8 to 1.25. The drift is the gravity term's signature, and a model that passed this test exactly would be wrong.

Provenance

Every quantity in this app carries one of six markers. The counts below are recomputed from the shipped files by the page harness, so the tally cannot drift from the code.

• DOCUMENTED stated outright in an IFSC document: the wall, the grid, all 31 hold positions and orientations, the false-start threshold, the timing resolution, the display rounding, the final quotas, the Annex 4 ladder, the ranking rules.
• qualified documented, but of something adjacent: the world-record tables come from an encyclopaedia rather than the governing body, because the IFSC's own records page is on a host that now 301s to a domain where the path is a 404. The Wikimedia route drawing used to cross-check the hold positions is a contributor's own work, not a CAD file. Counting these as DOCUMENTED would flatter the tally.
• DERIVED computed from documented values with no new assumption: the Italian-alphabet column mapping, the millimetre coordinates of every hold, the ascent order, the panel arithmetic.
• MEASURED produced by running the model: the finish time, the per-move binding limits, the elasticities, every convergence row.
• CALIBRATED chosen to match published data about people rather than about this wall: body mass, segment lengths, isometric forces, contraction velocities, the Hill curve's shape parameter, limb stiffness.
• RECONSTRUCTED invented because nothing decides it: the stop device's horizontal position and size, the starting pad's position on the floor, the shape of the master hold beyond “a ball and a long tail”, the rival's pace.

Tag counts are recounted from the shipped files by the page harness, so this page cannot claim a tally the code does not carry.

A blind spot, stated rather than hidden

The oracle that rebuilds the climb from emitted positions alone closes an energy budget, and that budget is homogeneous of degree one in body mass — work, kinetic energy and potential energy all scale together — so it is structurally blind to a wrong mass. Only the second half of that oracle, which asks whether the reconstructed forces were ones the reconstructed contacts could actually have carried, can see it, because a force ceiling is not mass-proportional. The harness asserts the blind spot as a fact, so it cannot quietly stop being true.

What is faithful, and what is a decision of mine

Faithful to the specification

• 15 000 mm wall on a 200 mm socle, continuing 500 mm above the last panel; two 3 000 mm lanes; a continuous 5° overhang.
• 1500 × 1500 mm panels, 10 high by 2 wide per lane, named sn and dx and numbered from the bottom; an 11 × 10 grid at 125 mm pitch with a 187.5 mm vertical edge inset and a 125 mm horizontal one.
• All 31 holds — 20 big, 11 small — at their published grid references and orientations, transcribed row by row from the route plan.
• The finishing pad's bottom edge at 14 800 mm above the ground.
• Reaction below 0.100 s is a false start; times ranked to 1/1000 s and displayed rounded down to 1/100 s; the Annex 4 ladder for fields of 4, 8 and 16; the quota table; the qualification stagger; the ranking order including the rule that a false starter is ranked equal last and behind a competitor who merely fell.

My decisions, not the specification's

• The ascent order, the stop device's position and size, the starting pad's placement, the hold shape, every body parameter, and the rival's pace.
• Eight playable sections instead of 45 individual limb moves: a limb move every 0.15 s is not a human input rate.
• A slip penalty of 0.12 s and a three-slip fall limit — a game rule with no counterpart in the IFSC rulebook, which has no notion of a partial slip at all.
• The rival climbs a scripted pace rather than being simulated; only the player's ascent is solved.

Sources, and what could not be opened

• IFSC Speed Licence Rules — Speed Walls, version 1, 1 February 2022, 15 pages. The archived copy is stored truncated at exactly 1 048 576 bytes while the crawler recorded the original as 1 601 911; a byte-range request returns the missing 553 335 and the two halves concatenate into a valid PDF. Text extraction on the truncated file silently loses the tail, which includes the homologation clauses.
• IFSC Rules 2024 v1.1, 88 pages — article 9 (Speed) and Annex 4.
• Wikipedia, Speed climbing wall and Speed climbing.
• Wikimedia Commons, Voie officielle d'escalade de vitesse by user Ghusse, CC BY-SA 4.0 — used only as an independent cross-check of the hold coordinates, never as their source.
• Could not be opened: the current IFSC rules index. ifsc-climbing.org now 301s to worldclimbing.com, where /about/rules returns 404. Every rule quoted here is therefore from the archived 2022 and 2024 editions, and a later edition may have changed any of it. The IFSC's own world-records page is on the same dead host, which is why the record tables are marked qualified.

World-record progression

23 men's and 23 women's records on the standardised 15 m wall, shipped exactly as published including the two internally inconsistent rows. qualified