Is it a fit?Book a call

Test report

The Proof Ladder

Six rungs, nine model families, 54 cells. One done. What runs next.

Read on 24 October 2026 at benchmarks.eigna.ai
  • 54CellsSix rungs times nine families
  • 1DoneTime series, rung a
  • 1PartialTime series, rung b
  • 52To runNo Eigna result yet
01

Six rungs, from the bare model to the chain that learns

Each rung compares two things on the same pinned test: the open base, and the Eigna version of it.

  1. Rung a, the model aloneBase model against the Eigna expert, on the same pinned test set.
  2. Rung b, with toolsBase plus tools against expert plus tools.
  3. Rung c, the decision modelBefore and after tuning: whether to act, how much, or when to ask a person.
  4. Rung d, expert and decision modelThe base pair against the tuned pair.
  5. Rung e, the whole chainFrom model to decision to action, with tools.
  6. Rung f, the whole chain, learningRetrained when new data arrives, to show what each retrain adds.
02

Nine families, 54 cells

Proof Ladder cells by family and rung
TLM, time seriesDonePartialTo runTo runTo runTo run
LLM, languageTo runTo runTo runTo runTo runTo run
VoiceTo runTo runTo runTo runTo runTo run
ImageTo runTo runTo runTo runTo runTo run
VideoTo runTo runTo runTo runTo runTo run
RoboticsTo runTo runTo runTo runTo runTo run
Typed routerTo runTo runTo runTo runTo runTo run
3DTo runTo runTo runTo runTo runTo run
GuardrailsTo runTo runTo runTo runTo runTo run
Done 1 Partial 1 To run 52
Done: the time-series model on rung a, Benchmark 001. Partial: the same model on rung b, where paper trading ran and the live session is blocked by a compute quota of 0.
  1. TLM, time seriesKronos-small, MIT licence. Has a result.
  2. LLM, languagegpt-oss-20b, Qwen3.5 9B, Mistral Small 3.2. No result yet.
  3. VoiceWhisper large-v3, Parakeet TDT, Kokoro-82M. No result yet.
  4. ImageFLUX.2 klein 4B, Qwen-Image, Z-Image-Turbo. No result yet.
  5. VideoWan 2.2, CogVideoX-2B. No result yet.
  6. RoboticsSmolVLA, pi0, GR00T N1.6. No result yet.
  7. Typed routerLaya 0.42B, Qwen2.5-3B with GRPO. No result yet.
  8. 3DTripoSG with a LoRA adapter. No result yet.
  9. GuardrailsAn open-weight classifier, to choose. No result yet.

Licences: Kronos-small (MIT) is verified. Every other base is checked before use.

03

What runs next

  1. 01Calibrate on untouched dataBand coverage 76 to 84% on a pre-registered validation split, CRPS no worse.Next
  2. 02Beat the simple baselinesA paired, significant CRPS win over last value and GARCH(1,1), at the 5% level.Planned
  3. 03More runs, more marketsFive or more runs; from 2 markets to 15, then all of the client’s markets.Planned
  4. 04Rung c, the decision layerA Qwen2.5-7B-Instruct decision adapter on 972 records a run beats the fixed rule on paper, after step 1.Planned
  5. 05The live paper sessionRung b live on one 4 vCPU machine, read only.Blocked: compute quota 0
  6. 06Public suitesfev-bench (100 tasks) and Chronos Benchmark II; TimesFM 2.5, Chronos-2, TiRex-2 and Toto 2.0 on the 243-day table, licences checked first.Planned
04

How a cell is filled

A cell is filled when the Eigna version and its open base meet the same pinned test, the result is scored in pairs, and it is published with its losses at benchmarks.eigna.ai. Licences are checked before any base is used.

Why we climb it in this orderbenchmarks.eigna.ai

Newsletter

Release notes.

Once a month: what we released, what we held, and why. Short, dated and true, losses included.

Double opt-in. One click to leave. We never share the list.

Each issue holds

  1. One release or testWhat passed the gate, what was held, and the reason in one line.
  2. One Journal pieceAn explainer from the people who build it.
  3. One numberWith its baseline and its date, never on its own.

The first issue comes in November 2026.

Test Lab

Add a cell with your own task.

A Test Lab runs one task on your own data in a sealed setup, scored against your baseline. You keep the report.

Request a Test LabBook a call