jevbooks

← All projects

finance-decision-evals

vicwlau/finance-decision-evals · Homepage

Evaluating Jev on ASC 606 revenue recognition decisions.

Evaluation & Benchmarking0.45Runner-up: Compliance & Legal
Automated README summary

Automatically summarized from the README on 13 Oct 2026; not a run by jevbooks. Run instructions and author-reported outcomes still need verification.

What it is

An evaluation of TypeSafe's Jev on a revenue accounting task: deciding if a software contract clause needs review and which ASC 606 issues it raises. Built by a revenue accounting manager for finance and AI evaluation audiences.

How it uses Jev

Jev answers whether a deal needs a revenue accounting review and which ASC 606 issues apply. State is the contract clause, with facts as written, named fields, or computed comparisons. Answers are used to compare against expert labels and analyze errors.

Primitives:choicescore

Technique worth stealing

Computing facts in code fixes the review call but not issue naming; Jev reads terms of art in their everyday sense.

Try it

bun install; bun rev-rec/scripts/freeze.ts check; export TYPESAFE_API_KEY=…; bun rev-rec/scripts/run.ts --split heldout --questions v2 --state computed --reps 3

View on GitHub

Start using Jev: API key and your first request →

Model judgments from README excerpts

Project signals use repository metadata, an existing summary and the first 700 README characters, not the full project. “Not found” means not detected in that material; it does not establish absence. These are the original judgments with recorded models and dates, not source audits or run results.

Models & evaluation dates

Inclusion gate
jev-1.13.0
Project signals
jev-1.13.0
Category
jev-1.13.0
Pattern tags
jev-1.13.0

Evidence

These signals use repository metadata, an existing summary and at most 700 cleaned README characters. ≥ 0.60 means reported in that excerpt; ≤ 0.40 means not found there. The middle band is inconclusive. None is a source-code audit or a test result. The runnable-tool question excludes research experiments; the recommendation question asks for a learning shortlist, not a quality verdict.

  • Jev is core to the projectFound in material0.83
  • Shows a System One patternnot found in the excerpt0.15
  • Describes handling uncertaintynot found in the excerpt0.06
  • Reports measured numbersnot found in the excerpt0.36
  • Is a runnable toolnot found in the excerpt0.05
  • Top-10 learning shortlist signalnot found in the excerpt0.33
  • Is a replica of the modelnot found in the excerpt0.05
  • Solves a common problem (0–2)score on a 0–2 scale0.98 / 2
  • About JevFound in material0.98
Signals
Jev is core to the project0.83Shows a System One pattern0.15Describes handling uncertainty0.06Reports measured numbers0.36Is a runnable tool0.05Top-10 learning shortlist signal0.33Is a replica of the model0.0500.51Solves a common problem (0–2)0.98 / 2012

no-judgment band (0.40–0.60)

Design patterns
Flight recorder0.69Traffic light0.45Funnel0.43Inspection0.35Question bank0.34Grey zone0.25Check-up0.2500.51

no-judgment band (0.40–0.60)

The Jev model and date for each judgment stage are listed above. Card written by DeepSeek V4.1 Flash from the README on 13 Oct 2026.