← Evals

Solve rate

Solved runs out of all runs for each agent version, with a 95% Wilson interval. Runs cut short by the machine (a killed process, the LLM server unreachable, the engine timing out) are left out.

• bdd945f 39% 7/18 20%–61%

Against another eval of the same questions

backend edge backend edge-thanveer multi-llm branch multi-llm main

Within noise: a difference this size could easily be chance with 18 and 18 runs. Fisher's exact test, two-sided, p = 0.092.

Tokens and time

Medians per run. Tokens come from RubyLLM's usage records; tokens per second is completion tokens over the run's wall time.

Every run

One dot per run, green solved, red stuck and grey interrupted, larger for more rounds; a hollow dot has no checks recorded. Hover or tap a dot for its top errors, and open it for the question's page.

Why checks fail

Each version's failed checks, as shares of the whole, by the hint in Question::HINTS their error matches first, or by the error's class when no hint matches. A check that repeats the error before it, in a streak of 3 or more, counts as a repeated failing move.

• repeated failing move 57
• no equation determines the … from what is known 47
• undefined method '…' for an instance of Physics::PhysicalObject 35
• is a word Ruby already uses 22
• ArgumentError 20
• gives nothing called … 17
• which none of its laws name/give to anything 13
• undefined local variable or method '…' for an instance of Physics::Solving 9
• 8 other categories 22

• bdd945f 242 in 18 runs

Most frequent errors

1. 51no equation determines the … from what is knownRuntimeError: no equation determines the mass of ice from what is known
2. 26undefined method '…' for an instance of Physics::PhysicalObjectNoMethodError: undefined method 'time_taken' for an instance of Physics::PhysicalObject; time_taken is a quantity of Velocity, through UniformCircularMotion, not a method: reach it from a motion, as <motion>.time_taken
3. 21no equation determines the … from what is knownRuntimeError: no equation determines the force on a wire on a from what is known
4. 11no equation determines the … from what is knownRuntimeError: no equation determines the acceleration of body during pushed from what is known
5. 11is a word Ruby already usesArgumentError: round is a word Ruby already uses: call it something else
6. 10which none of its laws name/give to anythingNameError: two_forces_at_right_angles takes force1 as force on body by pusher1, which none of its laws name
7. 9undefined local variable or method '…' for an instance of Physics::SolvingNameError: undefined local variable or method 'stone_motion' for an instance of Physics::Solving
8. 8no equation determines the … from what is knownRuntimeError: no equation determines the time taken of braking from what is known
9. 8gives nothing called …ArgumentError: two_parallel_wires gives nothing called field_at_a
10. 8undefined method '…' for an instance of Physics::PhysicalObjectNoMethodError: undefined method 'motion' for an instance of Physics::PhysicalObject; did you mean options?
11. 7which the solution's laws leave out…; or take oneNoMethodError: undefined method 'separation' for an instance of Physics::PhysicalObject; separation is a quantity of CoulombsLaw, ElectricField, TravellingWave, which the solution's laws leave out (add one to using:), not a method: reach it from an object, as <object>.separation, qualified as needed by from: <object>; or take one the solution's laws already name for that kind of quantity: radius, length_of_the_coil, distance_from_the_wire
12. 7ArgumentErrorArgumentError: copper_block_on_ice gives the specific heat capacity of copper twice: 0.39 J g-1 K-1 and 0.39e3 J kg-1 K-1

Where it ran

• agent 17b9257 · engine 453fb0d

  machine

  omarchy

  worktree

  /home/fazzledev/Work/phryzby-app--multi-llm-main

  commit

  bdd945f on HEAD

  clean

  yes

  ruby · rails

  3.4.7 · 8.1.4

  llm server

  edge-thanveer · http://127.0.0.1:8083 · llama.cpp b11371-99b95488c

  model

  qwen3.5:9b · Q4_K_M · 49152 context

  gpu

  Advanced Micro Devices, Inc. [AMD/ATI] Navi 33 [Radeon RX 7600/7600 XT/7600M XT/7600S/7700S / PRO W7600] (rev c1); Advanced Micro Devices, Inc. [AMD/ATI] Phoenix1 (rev c2)

  power

  on battery, 99% → on battery, 99%

  time

  14 Oct 09:01 → 14 Oct 09:22