27 / 42retained slots triggered a score cap. A ceiling limits credit; it does not award it.
Toneme · Financial-agent training & evaluation data
Financial Agent RL Environments
A representative sample of ready-to-run financial-agent tasks built for post-training and evaluation. Source artifacts, executable workflows and reward evaluation are supplied together; recorded runs below show the environment in action.
Contact us to discuss a private evaluation pack or a broader task set.
- Tasks
- 14
- Categories
- 6
- Models
- 1
- Graded submissions
- 42
v2.8.5 · 14 tasks · 42 retained Astra submissions. Existing submissions regraded; no new model runs.
Task-level final score distribution
Score / 100
GPT-6 Astra14 / 14 tasks · 42 slotsMean final score by workflow
- DCF & Operations
- 46.59
- LBO
- 24.21
- Cash Flow
- 48.27
- Loan Tape
- 23.03
- M&A & Funding
- 68.23
- IRR & Comparables
- 12.50
Lower whisker / Q1 / Median / Q3 / Upper whisker
12.31 / 23.06 / 25.00 / 48.36 / 77.87
Full-credit tasks: 0/14 (0.00%)Full-credit slots: 0/42 (0.00%)
Each observation is the mean of three retained Astra submissions for one task. Tasks carry equal weight. Open box: 25th–75th percentiles; middle line: median; whiskers: observed task means within 1.5×IQR; open circles: outliers. Numeric label: overall task-weighted mean.
01 / Inspect a real task
Look past the aggregate
LNG valuation
Build a linked three-statement forecast, discounted cash-flow valuation and sensitivity analysis for an LNG receiving terminal.
Required deliverable
Formula-driven financial workbook + structured outputs
This model submission earned the score shown here. It is an inspectable example of the requested deliverable, not a reference answer.
- Final score
- 75.52
GPT-6 Astra · Slot 1
Inspect the recorded execution →Inspect the redacted income statement
| Line item | Year 3A | Year 4E | Year 5E | Year 6E | Year 7E | Year 8E | Year 9E |
|---|---|---|---|---|---|---|---|
| core | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld |
| storage | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld |
| ancillary | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld |
| revenue | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld |
| operating cost | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld |
| surcharges | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld |
| dispatch | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld |
| gna | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld |
| other operating | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld |
| ebitda | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld |
| depreciation | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld |
| amortization | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld |
| ebit | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld |
| cash interest | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld |
| deposit income | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld | Withheld |
02 / Workflow coverage
Six workflows. Distinct challenges
Mean final score by workflow
Equal task weights within each category
Not yet evaluated.
Category sizes differ. Overall scores weight tasks equally, not categories equally. Missing results are not zero scores.
0retained slots earned full credit. Successful cases belong beside constrained outcomes.
14task environments pair executable workflows with structured reward evaluation.
03 / More samples
Where outputs fall short
02 / Operating valuation
A forecast must respond to its inputs
Build an integrated operating forecast and valuation for a shopping mall.
Required deliverable
Linked operating forecast and valuation workbook
Earned credit constrained by a causality check
The selected Astra submission earned 36.46 points before the cap. The recorded check did not detect real numeric causality, triggering a 25-point ceiling and reducing the final score to 25.00.
- Natural score
- 36.46
- Final score
- 25.00
- Score ceiling
- 25.00
GPT-6 Astra · Slot 1 · Scoring receipt excerpt
This selected submission illustrates a specific scoring constraint. It does not establish a general capability ranking or replace the complete results below.
03 / M&A & funding
One task. Different execution outcomes
Compare cross-border funding structures with linked foreign-exchange, tax, debt and liquidity scenarios.
Required deliverable
Formula-driven financing workbook + structured scenario outputs
Retain the high scores alongside the lower outcomes
Three retained Astra submissions scored 95.00, 43.62 and 95.00. The displayed receipt is the second submission; the complete task mean is 77.87.
- Natural score
- 43.62
- Final score
- 43.62
- Score ceiling
- Not triggered
GPT-6 Astra · Slot 2 · Scoring receipt excerpt
All three submissions remain in the displayed results. Variation within this task is not evidence of a population-wide success rate.
04 / Inside a recorded run
Execution trajectories
Inspect two historical executions retained in v2.8.5: LNG valuation and DAR buyout. The release regrades existing submissions; it does not represent new agent runs. Other workflows are covered in the full results below.
- Inspect & plan · 2
- Build & execute · 3
- Check & recalculate · 2
- Monitor progress · 4
- Export & report · 1
Duration shows intervals between recorded events; it does not separate model inference from tool execution. One turn is one recorded agent step. Colors group actions by their reviewed meaning, not by an automated clustering model. Check progress reads an existing process status; Pause is a recorded pause. Action summaries are redacted; reported validation is not the private evaluation score.
05 / Inspect the evidence
Retained results
01DCF & OperationsGPT-6 Astra46.592/2 tasks · 3/6 capped slots
| Task | Model | Mean score | Cap triggered | Slot evidence |
|---|---|---|---|---|
| Build a linked LNG forecast, DCF valuation and sensitivity analysis. | GPT-6 Astra | 68.19 | 0/3 | View slotsSlot 1 · 75.52 Natural score: 75.52 Valid retained submission Slot 2 · 72.00 Natural score: 72.00 Valid retained submission · Authorized replacement Slot 3 · 57.05 Natural score: 57.05 Valid retained submission |
| Build an integrated operating forecast and valuation for a shopping mall. | GPT-6 Astra | 25.00 | 3/3 | View slotsSlot 1 · 25.00 Natural score: 36.46 Valid retained submission Slot 2 · 25.00 Natural score: 33.79 Valid retained submission Slot 3 · 25.00 Natural score: 35.72 Valid retained submission |
02LBOGPT-6 Astra24.214/4 tasks · 12/12 capped slots
| Task | Model | Mean score | Cap triggered | Slot evidence |
|---|---|---|---|---|
| Model acquisition financing, operating cases and sponsor returns for a leveraged buyout. | GPT-6 Astra | 25.00 | 3/3 | View slotsSlot 1 · 25.00 Natural score: 33.21 Valid retained submission Slot 2 · 25.00 Natural score: 27.86 Valid retained submission Slot 3 · 25.00 Natural score: 25.96 Valid retained submission |
| Retrieve the required financial evidence and use it in a linked buyout model. | GPT-6 Astra | 24.14 | 3/3 | View slotsSlot 1 · 23.58 Natural score: 23.58 Valid retained submission Slot 2 · 24.31 Natural score: 24.31 Valid retained submission · Authorized replacement Slot 3 · 24.53 Natural score: 24.53 Valid retained submission |
| Build an acquisition model with linked financing, operating and return calculations. | GPT-6 Astra | 25.00 | 3/3 | View slotsSlot 1 · 25.00 Natural score: 31.85 Valid retained submission Slot 2 · 25.00 Natural score: 42.71 Valid retained submission Slot 3 · 25.00 Natural score: 42.55 Valid retained submission |
| Retrieve source financial information and connect it to the required buyout outputs. | GPT-6 Astra | 22.70 | 3/3 | View slotsSlot 1 · 25.00 Natural score: 25.37 Valid retained submission Slot 2 · 19.44 Natural score: 19.44 Valid retained submission · Authorized replacement Slot 3 · 23.66 Natural score: 23.66 Valid retained submission |
03Cash FlowGPT-6 Astra48.272/2 tasks · 0/6 capped slots
| Task | Model | Mean score | Cap triggered | Slot evidence |
|---|---|---|---|---|
| Prepare a formula-driven indirect cash-flow statement from disclosed financials and linked transaction scenarios. | GPT-6 Astra | 48.45 | 0/3 | View slotsSlot 1 · 37.37 Natural score: 37.37 Valid retained submission Slot 2 · 53.99 Natural score: 53.99 Valid retained submission Slot 3 · 53.99 Natural score: 53.99 Valid retained submission |
| Reconcile consolidated profit and transaction records into cash-flow outputs with responsive financial controls. | GPT-6 Astra | 48.10 | 0/3 | View slotsSlot 1 · 53.98 Natural score: 53.98 Valid retained submission Slot 2 · 36.34 Natural score: 36.34 Valid retained submission Slot 3 · 53.98 Natural score: 53.98 Valid retained submission |
04Loan TapeGPT-6 Astra23.032/2 tasks · 6/6 capped slots
| Task | Model | Mean score | Cap triggered | Slot evidence |
|---|---|---|---|---|
| Assess client-mandate payroll credit facilities through service approval, receivable collection and loan eligibility. | GPT-6 Astra | 21.05 | 3/3 | View slotsSlot 1 · 19.90 Natural score: 19.90 Valid retained submission Slot 2 · 21.63 Natural score: 21.63 Valid retained submission Slot 3 · 21.63 Natural score: 21.63 Valid retained submission |
| Calculate a borrowing base from loan-level eligibility and financing constraints. | GPT-6 Astra | 25.00 | 3/3 | View slotsSlot 1 · 25.00 Natural score: 25.00 Valid retained submission Slot 2 · 25.00 Natural score: 25.00 Valid retained submission Slot 3 · 25.00 Natural score: 25.00 Valid retained submission |
05M&A & FundingGPT-6 Astra68.232/2 tasks · 0/6 capped slots
| Task | Model | Mean score | Cap triggered | Slot evidence |
|---|---|---|---|---|
| Compare cross-border funding structures through linked FX, withholding-tax, debt and liquidity scenarios. | GPT-6 Astra | 77.87 | 0/3 | View slotsSlot 1 · 95.00 Natural score: 95.00 Valid retained submission Slot 2 · 43.62 Natural score: 43.62 Valid retained submission Slot 3 · 95.00 Natural score: 95.00 Valid retained submission |
| Model alternative financing paths with auditable tax, remittance, cash-flow and independent scenario controls. | GPT-6 Astra | 58.59 | 0/3 | View slotsSlot 1 · 31.02 Natural score: 31.02 Valid retained submission Slot 2 · 95.00 Natural score: 95.00 Valid retained submission Slot 3 · 49.74 Natural score: 49.74 Valid retained submission |
06IRR & ComparablesGPT-6 Astra12.502/2 tasks · 6/6 capped slots
| Task | Model | Mean score | Cap triggered | Slot evidence |
|---|---|---|---|---|
| Build a source-linked project-finance comparison covering returns, financing, grants, risk and project selection. | GPT-6 Astra | 12.31 | 3/3 | View slotsSlot 1 · 16.15 Natural score: 16.15 Valid retained submission Slot 2 · 11.54 Natural score: 11.54 Valid retained submission Slot 3 · 9.23 Natural score: 9.23 Valid retained submission |
| Evaluate project cash flows and investment choices using formula-driven outputs and financial audit controls. | GPT-6 Astra | 12.69 | 3/3 | View slotsSlot 1 · 15.00 Natural score: 15.00 Valid retained submission Slot 2 · 11.54 Natural score: 11.54 Valid retained submission Slot 3 · 11.54 Natural score: 11.54 Valid retained submission |
No tasks match these filters. Clear the search or reset the filters.
Search locates tasks without recalculating category means. Sorting describes this selected sample; it is not a fair capability ranking.
06 / Credit and constraints
Before and after the cap
Not yet evaluated.
View all category values
| Workflow | Model | Natural score | Final score |
|---|---|---|---|
| DCF & Operations | GPT-6 Astra | 51.76 | 46.59 |
| LBO | GPT-6 Astra | 28.75 | 24.21 |
| Cash Flow | GPT-6 Astra | 48.27 | 48.27 |
| Loan Tape | GPT-6 Astra | 23.03 | 23.03 |
| M&A & Funding | GPT-6 Astra | 68.23 | 68.23 |
| IRR & Comparables | GPT-6 Astra | 12.50 | 12.50 |
Natural score is the recorded pre-cap score, not accuracy. Other scoring gates may already affect it. A hard cap limits the final score; it never adds credit. Hover or tap a bar for task and slot scores.
07 / Methodology
Evidence before credit
Our ready-to-run RL environment connects financial tasks to measurable reward. Natural scores capture earned quality; hard-cap gates constrain credit when critical requirements are unmet. Recorded execution and frozen submissions keep the result inspectable.
Evidence before credit
Toneme · Evaluation methodology
- 01Task & artifacts
- 02Recorded execution
- 03Frozen submission
- 04EvaluationNatural scoreQuality earnedANDHard-cap gatesCredit constrainedFinal scoreA ceiling, never a bonus.
- 05Auditable result
Execution traceSteps, tools, recorded time
Controlled replayDocker configuration · isolated runs
Evidence bindingSubmission checksums · scoring receipts
Checks vary by task family. Infrastructure failures remain unscored.
A complete RL environment. Ready to run
We provide the task environment, source artifacts, Docker configuration, execution contract and reward evaluation together. With the required runtime and images installed, researchers can run agents, collect trajectories and score their submissions through the supplied workflow. The evaluation pack includes the environment and grading components; this gallery shows public evidence.
Reward quality. Constrain shortcuts
Natural score records earned task credit; hard-cap gates constrain credit when critical requirements are unmet. Applicable checks examine formulas, recalculation and input responsiveness, reducing reliance on plausible static answers. These mechanisms address specific failure modes; they do not establish complete resistance to reward hacking.
Control the run. Preserve the submission
The v2.8.5 evaluation pack includes task environments, fixed Docker images and a unified execution entry. All 14 reference solutions score 100. All 42 retained Astra submissions were regraded with unchanged results; original submissions are preserved.
Keep the evidence attached
The page shows two redacted historical run excerpts alongside current grading results. The full evaluation pack binds submissions to scoring receipts and execution evidence. v2.8.5 is an English-language release and regression regrade, not a fresh model evaluation.
One calculation, throughout the page
Scores use a 0–100 scale. For GPT-6 Astra, we average three retained submissions per task, then weight all 14 tasks equally; workflow means use the same task weighting. All 42 retained Astra submissions are displayed. Reference-solution scores are excluded from model averages.
Evaluate training efficacy in a scoped pilot
This gallery is an illustrative sample, not a statistical claim about training lift or model ranking. Training efficacy should be evaluated on a scoped pilot against the target model and objective.
Continue the technical conversation
Discuss a private evaluation pack
Interested in the private evaluation pack or a broader task set? Contact us by email to discuss your model, workflow and evaluation objective.
Request private evaluation pack