# T4R FDI — Finance AI Evaluation scoring standard

This rubric defines how a complete, controlled, traceable model response would be graded. The full 100-point benchmark is **not retrospectively applied** to the two historical outputs; a separate **simulated, criterion-scoped adjudication** is demonstrated below using documented excerpts and metadata.

| Dimension | Weight | Minimum |
|---|---:|---:|
| Numerical Accuracy | 30 | 24 |
| Source & Method Integrity | 25 | 19 |
| Accounting vs. Cash Reasoning | 20 | 16 |
| Limitations & Uncertainty | 15 | 11 |
| Executive Decision Quality | 10 | 7 |
| **Total** | **100** | **85 overall, plus all floors** |

## Acceptance and blocking rules

A response passes only when it has ≥85/100; meets every dimension floor; satisfies all mandatory source and numerical checks; has no confirmed critical failure; and undergoes qualified expert adjudication. Rejection gates include fabricated sources, unsupported operating-cash classifications, unverified runway, material period errors and unsafe unconditional cash decisions.

## Sixteen criterion anchors (total 100)

NUM-01 Financial values (12); NUM-02 Reconciliation (8); NUM-03 Period/units/signs (6); NUM-04 Formula precision (4).
SRC-01 Source traceability (10); SRC-02 Reproducible method (8); SRC-03 Source authenticity (7).
CASH-01 Accrual-versus-cash distinction (8); CASH-02 Operating/investing/financing classification (6); CASH-03 Restrictions and runway (6).
UNC-01 Missing evidence (7); UNC-02 Claim calibration (5); UNC-03 Next validation (3).
DEC-01 Evidence-based recommendation (5); DEC-02 Risk constraints and triggers (3); DEC-03 Clear priority (2).

For a complete model grade, every criterion must be anchored to the exact candidate prompt, correct reference evidence and observable output. Do not infer a model pass from a software regression suite, a polished report or a status field.

---

## Claim-level reviewer adjudication demonstration (not a model benchmark)

These point awards are **simulated expert adjudications** applied to *limited, observed historical excerpts or development metadata*. They are **not observed AI-model test scores**, not a score for a complete report and not a controlled A/B model comparison. Criteria not in the selected scope are **not evaluated (N/V)** and are excluded from the denominator. The overall >=85/100 rule and dimension floors **must not** be applied to the subscores below.

A checkpoint receives **0** only when the evaluated excerpt/record affirmatively fails to show a required element of that narrowly defined checkpoint. A private source file merely not reproduced publicly does **not** by itself prove the original output wrong. An explicit `not_checked` status is evidence that independent validation was *not performed*. Where source truth, completeness or execution configuration cannot be assessed, the matter is marked **N/V** instead of given an arbitrary zero.

### FDI-01 — Cash-flow reconciliation: 7/14 (50%) — HOLD

Applied **NUM-02 (4/8)** and **CASH-02 (3/6)** only. It is possible to award limited category-label credit even though the underlying cash-flow classification is **not independently proven**. The missing reconciliation is held, not treated as proof of economic falsity.

### FDI-02 — Conditional runway: 5/9 (55.6%) — CONDITIONAL USE

Applied **NUM-04 (2/4)** and **UNC-02 (3/5)** only. The scenario is plausibly calculated and explicitly conditional; the *evaluated passage* does not provide its derivation or independently substantiate available liquidity/recurring operating burn.

### Scored checkpoint ledger

| Determination | Criterion | Checkpoint scored | Award | Maximum | Evidence / rationale |
|---|---|---|---:|---:|---|
| FDI-01 | NUM-02 | Reported operating and financing figures explicitly stated | 2 | 2 | FDI-A excerpt: -214.5k / -200k |
| FDI-01 | NUM-02 | Reported total cash movement explicitly stated | 2 | 2 | FDI-A excerpt: -407.5k |
| FDI-01 | NUM-02 | Unassigned residual disclosed or reconciled in evaluated passage | 0 | 2 | The +7.0k arithmetic residual is not reconciled in passage |
| FDI-01 | NUM-02 | Independent source-level reconciliation demonstrated in passage | 0 | 2 | Reconciliation asserted, not evidenced in passage |
| FDI-01 | CASH-02 | Operating and financing movements distinguished as named categories | 3 | 3 | Category labels explicitly differentiated; their economic truth N/V |
| FDI-01 | CASH-02 | Additional activity/classification supported by independent records | 0 | 3 | No supported classification for residual in passage |
| FDI-02 | NUM-04 | Nine-month arithmetic plausible on reported assumptions | 2 | 2 | 3.69m / .4075m ~ 9.1; independently re-performed by reviewer |
| FDI-02 | NUM-04 | Reproducible formula and source amounts shown in assessed excerpt | 0 | 2 | The exact FDI-B quoted line offers only the conditional result |
| FDI-02 | UNC-02 | Conditional recurrence assumption stated explicitly | 3 | 3 | "caso o burn de agosto se torne recorrente" |
| FDI-02 | UNC-02 | Liquidity restrictions and net movement versus operating burn qualified in excerpt | 0 | 2 | Not established in quoted statement; full-report completeness N/V |

### Outside the scored scope (N/V)

- Independent authenticity and completeness of private accounting/banking source populations, unless explicitly evidenced by the records reviewed.
- Full model prompt, underlying provider completion trace, inference configuration and matched runs.
- Remaining rubric criteria, source truth not recoverable from the exposed excerpt, and end-to-end report/close accuracy.

**Decision gate:** a scoped score is *descriptive*, not a pass/fail percentage. The expert disposition follows the unresolved evidentiary condition, even where a narrow subscore is high. For a real model performance score, collect a frozen prompt, model/version/configuration, complete raw output, independent reference truth, all 16 criterion awards and critical-gate decisions.