1 reporting.generation_throughput

reporting.generation_throughput

Evaluate the frozen, three-scene computational throughput assignment.

Oracle owns admission, tensor comparisons and the paired log-cost reducer. This owner records their verdict in SENPAI schema-v2; it adds no statistical gate or scientific computation. W&B publication remains a separate operation of aria_nbv.reporting.performance_evidence.

1.1 Functions

Name	Description
evaluate_assignment	Write a new, immutable evaluator snapshot after canonical validation.
evaluator_fingerprint	Hash the supported evaluator and shared measurement reader bytes.
main	Run the real evaluator; return zero only for its retained-gain verdict.

1.1.1 evaluate_assignment

reporting.generation_throughput.evaluate_assignment(
    assignment_path,
    *,
    brief_path,
    output_dir,
    goal_slug,
)

Write a new, immutable evaluator snapshot after canonical validation.

Counts describe the fixed 18-attempt engineering roster. The three scenes and two order-balanced production blocks are not a population estimate. Missing cost reductions remain absent metrics, never zero speedups. PASS requires the Oracle reader’s complete scientific, resource and frozen observed-variation verdict. No learning or corpus-admission claim follows.

1.1.2 evaluator_fingerprint

reporting.generation_throughput.evaluator_fingerprint()

Hash the supported evaluator and shared measurement reader bytes.

Paths are package-relative to make this identity independent of checkout location. This fingerprint is frozen before candidate acquisition; the candidate’s single computation patch must leave these owners unchanged.

1.1.3 main

reporting.generation_throughput.main()

Run the real evaluator; return zero only for its retained-gain verdict.

# reporting.generation_throughput { #aria_nbv.reporting.generation_throughput }

`reporting.generation_throughput`

Evaluate the frozen, three-scene computational throughput assignment.

Oracle owns admission, tensor comparisons and the paired log-cost reducer.
This owner records their verdict in SENPAI schema-v2; it adds no statistical
gate or scientific computation. W&B publication remains a separate operation
of :mod:`aria_nbv.reporting.performance_evidence`.

## Functions

| Name | Description |
| --- | --- |
| [evaluate_assignment](#aria_nbv.reporting.generation_throughput.evaluate_assignment) | Write a new, immutable evaluator snapshot after canonical validation. |
| [evaluator_fingerprint](#aria_nbv.reporting.generation_throughput.evaluator_fingerprint) | Hash the supported evaluator and shared measurement reader bytes. |
| [main](#aria_nbv.reporting.generation_throughput.main) | Run the real evaluator; return zero only for its retained-gain verdict. |

### evaluate_assignment { #aria_nbv.reporting.generation_throughput.evaluate_assignment }

```python
reporting.generation_throughput.evaluate_assignment(
    assignment_path,
    *,
    brief_path,
    output_dir,
    goal_slug,
)
```

Write a new, immutable evaluator snapshot after canonical validation.

Counts describe the fixed 18-attempt engineering roster. The three scenes
and two order-balanced production blocks are not a population estimate.
Missing cost reductions remain absent metrics, never zero speedups. PASS
requires the Oracle reader's complete scientific, resource and frozen
observed-variation verdict. No learning or corpus-admission claim follows.

### evaluator_fingerprint { #aria_nbv.reporting.generation_throughput.evaluator_fingerprint }

```python
reporting.generation_throughput.evaluator_fingerprint()
```

Hash the supported evaluator and shared measurement reader bytes.

Paths are package-relative to make this identity independent of checkout
location. This fingerprint is frozen before candidate acquisition; the
candidate's single computation patch must leave these owners unchanged.

### main { #aria_nbv.reporting.generation_throughput.main }

```python
reporting.generation_throughput.main()
```

Run the real evaluator; return zero only for its retained-gain verdict.