Benchmarks
Proof before performance claims
The benchmark record separates measured results from targets. Frozen cohort hashes prevent after-the-fact sample selection; measurement boundaries prevent throughput figures from being presented as something they are not.
Live engine proof
17,656 NERVI · 39,211 READY
This block is read from the live reconstruction database at request time. It is not a static marketing claim. The 20/20 score is earned only when every declared target is actually measured; the page does not round a near miss up to 20.
- Search coverage PASS
- 224,275
- Documentary PASS
- 44,874
- NERVI PASS
- 17,656
- READY PASS
- 39,211
- Documentary → NERVI gap
- 27,218
- NERVI → READY gap
- 0
- READY conversion
- 222.08%
- Schema
- 0808
| Measured axis | Live score | 20/20 condition |
|---|---|---|
| Rigor / predictive proof | 18.00 / 20 | 20 only when PNE is live with falsifiable calibration: either the DB model is live, or >=10 app-level predictions were durably pre-registered before work and scored against canonical NERVI outcomes. |
| Integrity / fail-closed | 20.00 / 20 | 20 only with zero worker errors, zero public verification failures and zero rollbacks in the last hour. |
| NERVI throughput | 0.00 / 20 | 20 only when the exact first-PASS end-to-end 60-minute lower 95% bound across Coverage→Documentary→NERVI→READY→NERVI100→verified publication is >=100,000/hour. |
| Verified publication | 0.02 / 20 | 20 only when the same exact first-PASS proof clears 100,000 verified publications/hour and there are zero verification failures/rollbacks. |
| Closure / backlog drain | 0.85 / 20 | 20 only when Documentary→NERVI backlog is zero or the measured net rate implies <=1 hour to drain it. |
Overall: 7.77 / 20 · At least one live gate has not yet earned 20/20. · Measured 2026-10-07 19:59:12 UTC. · World-best certificate: UNPROVEN.
Net production proof
1,158.3 NERVI/h
Net unique current-PASS growth over 0.97 observed hours;90 PNE-tagged editorial runs in the last hour.
Publication proof
100 verified / hour
Queue due: 2,899 · verification failures: 0 · rollbacks: 0 · corpus authorization: LIVE.
Queue states: PENDING=890 · VERIFIED=110
Top publication blockers
- 2 × candidate changed; set-based READY reconciliation
Live operational counts
Larger live counts — not certification
These sets are measured live from the database and are not interchangeable. They describe operational state only; literal 100% corpus claims are governed solely by the strict corpus certificate.
Live operational counts are unavailable at this moment.
Historical frozen proof
14 distinct READY records in the original public milestone
The block below is retained as an immutable historical benchmark. It must not be confused with the live counts above.
Current proven aggregate · historical record
14 distinct READY records
The aggregate is append-only across three measured milestones: 10 + 3 + 1. It is not presented as one statistical 14/14 cohort. The latest milestone is Johanna Ambrosius, closed automatically through evidence-change, documentary coverage, NERVI and READY without adding a canonical claim or publishing.
Statistical baseline
READY wave — run 21273
| Records claimed | 50 |
|---|---|
| Complete candidates | 10 |
| Writer audits PASS | 30 / 30 |
| NERVI PASS | 10 |
| READY PASS | 10 |
| Closure errors | 0 |
| Unintended publications | 0 |
| Complete → READY | 100% (10 / 10) |
Ten successes without an observed failure prove that the production path works. They do not establish industrial reliability. The exact two-sided 95% lower confidence bound for 10/10 is approximately 69.15%. The next public targets are 100/100, 1,000/1,000 and 10,000/10,000 on frozen cohorts.
Frozen scale benchmark
NERVI-SCALE-BENCH-1.0
- Manifest SHA-256
- 7b699a47a79baa8a093e562ef0681220451e053c2615b5775e3e7d5ffe1a9ff7
- Nested cohorts
- 10 / 100 / 1000
- Stored-evidence compile + validation
- 55,630 records/hour
- Boundary
- Stored-evidence compilation and validation only; external evidence acquisition and source latency are outside this throughput figure.
Industrial standard
What would count as strong evidence
VALÉORINE will treat 1,000/1,000 zero-failure Complete→READY conversion on the frozen protocol as the first industrial reliability threshold. With no observed failures, that result corresponds to an exact two-sided 95% lower bound of approximately 99.6318%, before any allowance for distribution shift.