EVIDENCE CENTER · ENGLISH RECORD

Verifiable Computing.

Evidence is not a marketing appendix. It is part of the Runtime product. Current product evidence, historical research records, negative evidence and claim boundaries remain explicitly separated.

Evidence pages remain English to preserve record continuity.

LATEST EVIDENCE

Current evidence first.

Current · Product-facing

SCALEPASS

65.536B

Structured nodes validated

L213 Scale Validation

Bounded RCPSP run controlling a 2.821T potential computational space without exhaustive evaluation.

BOUNDARY · Claim boundary: structured scale validation, not exhaustive evaluation of 2.821T states.

PORTFOLIOPASS

162 / 162

Frozen case equality

Portfolio Optimization

Exact equality across frozen reference cases for portfolio optimization runs.

BOUNDARY · Claim boundary: deterministic equality on frozen input set.

RECOVERYPASS

100%

Checkpoint replay integrity

Runtime Recovery + Replay

Deterministic replay of verified execution from checkpointed runtime state.

BOUNDARY · Claim boundary: replay integrity under controlled runtime profile.

QOBLIB · EXTERNAL BENCHMARK VALIDATION

5 / 5 submitted instances passed official CI validation.

PR #84 passed 3/3 submitted instances and PR #85 passed 2/2. Both pull requests remain OPEN and NOT MERGED; reviewer feedback is still in progress.

5/5

CI PASS

#84

3/3

#85

2/2

QOBLIB PR #84

Maximum Independent Set

OPEN · NOT MERGED · REVIEW

PROBLEM DEFINITION

Given a graph, find the largest set of vertices such that no two selected vertices share an edge.

WHY IT IS HARD

The number of possible selections grows combinatorially. QOBLIB uses unweighted MIS as one of its external optimization benchmark classes.

Submitted instances

3 / 3 PASS

Official checker

Overall OK

C125-9 · result 34 · QOBLIB-listed optimal

p_hat1500-3 · result 94 · QOBLIB-listed optimal

hamming10-4 · result 40 · published best-known

HEAD 1dca0fb1903de765bbe93c575d89c96105b416df

Open GitHub PR

QOBLIB PR #85

LABS

OPEN · NOT MERGED · REVIEW

PROBLEM DEFINITION

Find a ±1 binary sequence that minimizes the sum of squared off-peak autocorrelations.

WHY IT IS HARD

Changing one bit can affect many correlation terms at once, creating a difficult discrete search landscape.

Submitted instances

2 / 2 PASS

Official checker

Overall OK

labs067 · energy 241 · published best-known

labs099 · energy 577 · published best-known

HEAD 4076ab2ff2493877a5527a20f8d3578057f7bda7

Open GitHub PR

Classical execution only. CI validation is not PR acceptance or merge, does not independently prove optimality, and does not claim quantum advantage, new public best, or universal solver superiority.

PRESERVED PUBLIC RECORDS

Existing evidence remains reachable.

EXACT_CORRECTNESS · EXACTPASS

gr21 / 21 cities

TSP · TSPLIB · Held-Karp DP · Runtime 2707 = Reference 2707

View Evidence

STRUCTURED_GRAPH_SCALEPASS

65.536B-node structured graph

Structured scale validation · reference equality · not exhaustive traversal

View Evidence

RUNTIME SEARCH CONTROL · ARCHIVEPASS

2.821T potential RCPSP space

Bounded historical run · 818 states evaluated · not exhaustive 2.821T evaluation

View Evidence

MULTI-PROBLEM · RESEARCH ARCHIVEVERIFIED

Historical multi-problem Runtime evidence

Historical research scope retained separately from Golden Core V1 portfolio scope

View Evidence

EVIDENCE STREAMS

New evidence is filed by capability and industry.

Historical exactness records

Research archive

Historical problem-class records are separate from current product claims.

• TSP

• Knapsack

• Max-Cut

• QUBO

• RCPSP

Scale research

Research archive

Structured scale and logical state-space records retain separate claim boundaries.

• Structured graph

• Logical state-space

• Resource-boundary experiments

Runtime control research

Research archive

Preserved Runtime control and architecture research streams.

• Preserve / reuse / re-evaluate

• Selective recompute

• PASS / HOLD / STOP

• Distributed execution studies

CLAIM BOUNDARY

Scale Evidence, Exactness Evidence, Golden Core V1 evidence and historical multi-problem research evidence are not merged into one superiority claim. Golden Core V1's corrected Stage 11 scope is the three portfolio sizes 40 / 42 / 50. Historical TSP / Knapsack / Max-Cut / QUBO / RCPSP records remain Research Archive evidence unless separately promoted under a current claim gate.

Open append-only Research Archive