A Research Instrument, Not a Cryptographic Claim
QCE MathBench
A benchmark you can audit. QCE MathBench v1 is a receipt-scoped manifest for measuring sparse, symbolic route and slice selection over the DNCZ authority stack — exact replay checks, public-safe corpus gates, and honest negative controls. This studio publishes the method and the gates, with the corpus metrics read straight from the lane receipts and the dense-state cap computed live. It is the discipline, not a trophy: a manifest and a measuring rod, never a "solved problem" headline.
Prove all things; hold fast that which is good. — 1 Thessalonians 5:21
The Bench
This is a research instrument — the published specification of a benchmark, not a solved problem and not a benchmark run. QCE MathBench v1 defines, as a receipt-scoped manifest, how one would test sparse symbolic QCE+DNCZ route/slice selection with exact replay checks, public-safe corpus gates, and negative controls. The numbers below are read straight from the lane receipts; nothing here is freshly executed, and nothing claims a cryptographic break.
For the dated, data-backed engines see the Scriptorium; for the constants-and-geometry studio see God's Math. Scripture, not the engine, is the authority.
What the Bench Measures Against
A benchmark is only as honest as the corpus it replays. MathBench v1 names each slice with its source receipt and its measured restore behaviour. These figures are recorded from prior receipt-scoped lanes — they are not re-run here.
| Slice | Recorded metrics | Boundary |
|---|---|---|
| QCE trillion virtual-address substrate | 1,000,000,000,000 virtual logical addresses · 1,000,000 atom spans · 3,970-byte recipe pack · 0 restore failures · dense-state cap 20 | sparse symbolic recipe authority, not physical qubits |
| QCE billion bridge regression | 1,000,000,000 virtual logical addresses · 1,000 atom spans · 3,952-byte recipe pack · 0 restore failures | smaller regression baseline only |
| TraceBench v0 replay fixture | 11-entry pack · SHA-256 verified · 0 failed restores · retained / rejected ids tracked | fixture evidence only — no general utility claim |
| DNCZ authority-stack proof-bedrock | 32 source files · 67,108,864 restored bytes · 0 mismatches · 0 negative failures | generated recipe/codebook corpus, not universal compression |
| Public-safe replay corpus | public_csv · rfc_text · unicode_data · portable public replay bundle (public-safe summaries) | no internal payload, private body, or archive publication |
Honest label. Each row is a recorded metric from a receipt-scoped lane, not a fresh run on this page. Restored evidence is disclosed as restored evidence; unresolved source roots are recorded as blockers, never silently substituted.
Why Sparse, Not Dense — the Cap, in Amplitudes
The manifest records a dense-state cap of 20 source qubits. A dense statevector for 20 qubits already needs 1048576 complex amplitudes — computed live here as 220. That is the whole reason the bench works with sparse symbolic addresses: a trillion virtual addresses live inside a 3,970-byte recipe pack, and a dense 20-qubit array is never materialised. Beyond the cap there is no dense state at all — only the recipe.
| dense-state cap (qubits) | 20 |
| dense amplitudes at the cap (220) | 1048576 |
| virtual addresses carried sparsely | 1,000,000,000,000 |
Honest label. The amplitude count is genuine and verifiable: 220 = 1,048,576. It is shown precisely so the “trillion” figure can never be misread as a dense trillion-qubit simulation. It is not one, and we do not present it as one.
Controls a Real Result Must Survive
A selection benchmark that cannot fail is not a benchmark. MathBench v1 requires every one of these negative controls to reject before any utility wording is allowed.
- ×corrupt DNCZ / container payload
- ×stale integrity hash
- ×rehashed recipe mismatch
- ×stale route with wrong shift
- ×no-signal random lane
- ×duplicate shadow route
- ×memory-hint-only false positive
- ×label shuffle
- ×restored-byte tamper
- ×private-body leak sentinel
- ×BTC / RSA solve-wording leak
- ×dense-statevector overclaim wording
Honest label. The last two controls are wording controls: they fail a packet that leaks a BTC/RSA “solve” claim or a dense-statevector overclaim. The guardrail is enforced by the bench itself, not just by prose.
What Each Candidate Records
Every candidate row carries labelled expectations and outcomes. These are the fields the manifest requires — not measured results. Note that effective and executed eops/s are fields to be filled by an approved execution lane, never values claimed here.
candidate_id
slice_id
corpus_id
source_sha256
restored_sha256
restore_ok
expected_outcome
actual_outcome
exact_positive
selected
rejected
false_positive
false_negative
policy_score
quality
cost
branch_alignment_score
recorded_executed_eops_per_second
required_multiplier_gate
effective_eops_per_second
public_safe
blocked_claims
Honest label. A selector is only credible against labelled expectations, so the schema separates executed eops/s from a verified skip/collapse multiplier and the resulting effective eops/s — effective skipped-work math, not fresh hardware speed.
A receipt-scoped benchmark manifest for sparse symbolic QCE+DNCZ route/slice selection, exact replay checks, public-safe corpus gates, negative controls, and explicit blocked-overread wording.
- Does not solve BTC71, BTC135, RSA, ECDLP, or any hard semiprime.
- Makes no physical billion/trillion-qubit claim — addresses are sparse symbolic, not qubits.
- Runs no dense trillion-qubit or dense 2N amplitude simulation.
- Claims no public quantum advantage or cryptanalytic advantage.
- Reports no fresh-hardware eops/s or model-scoring results.
Status & execution gate
This page is the specification. A later executable lane must carry explicit approval, a command list, the corpus manifest, and a no-private-body confirmation before any run — this page is not that approval. Unresolved source roots (exact FNLM eval set, certain DNCZ latest receipts, the original Planko burn root, named BTC71/1000d guardrail files) are recorded as blockers and are not silently substituted.
Source lane: WO32251 / qce_mathbench_manifest. Boundary: source-local manifest planning only — no benchmark runs, no model scoring, no restores, no deployments, no provider changes, no private-body reads, and no BTC/RSA solve claims.
How to Read This Instrument
A benchmark is only trustworthy if you can replay it and try to break it, so this instrument is built on three honest commitments. First, every corpus slice names its source files and their SHA-256, and a restore must reproduce those exact bytes — no parity, no claim. Second, for every "it works" there is a negative control designed to make it fail; utility wording is withheld unless the controls reject as intended. Third, speed is reported as separated fields — recorded executed operations, the verified skip-or-collapse multiplier, and the resulting effective rate — so effective work saved is never confused with fresh hardware speed.
This is the source-only v1: it publishes the manifest and its gates, and computes only the dense-state cap (2^20 amplitudes) to make the sparse-vs-dense distinction plain. The executable run is held behind an explicit approval gate, with its missing inputs disclosed rather than silently filled. Where a claim would need dated evidence, this studio hands off to the Scriptorium and the Chronology Atlas, which carry their own source ladders.
Go Deeper
God's Math Studio — constants, five-fold geometry, and the Bible's own arithmetic, computed live and honestly labelled.
DNCZ Proof — the authority-stack restore evidence that MathBench's bedrock slice depends on.
Knowledge Library — the wider research and reference shelf.
Research Articles — longer-form method and study write-ups.
The Chronology Atlas — the editorial source ladder and evidence tiers for dated claims.