01Latency Evidence Infrastructure

Replay it exactly.
Prove it did not get slower.

BQL Engine 2.0 is latency evidence infrastructure for market systems: deterministic C++20 replay, CI latency regression gates, canonical symbol IDs, breaker-style protection logic, and audit-ready evidence bundles.

Deterministic C++20 replayCI-enforced p50 / p95 / p99 / p99.9 gatesCanonical symbol IDsAudit-ready evidence bundlesEnterprise ITCH / FIX roadmap
1.96M
events/sec
Tier A match-only throughput
1.20M
events/sec
Tier C full proof pipeline
100–200×
runtime
Tier C Phase 3 → Phase 4
128×
journal
overhead reduction
100%
digest
consistency across tiers
02Why BQL Exists

One deterministic contract for replay, latency, gates and evidence.

Low-latency teams do not fail because they lack benchmark scripts. They fail because benchmark results, replay state, runtime protection decisions, and evidence artifacts usually live in separate worlds.

BQL Engine 2.0 brings them into one deterministic contract: replay the same workload, preserve the same gate decisions, verify the same digest, and package the evidence every time.

Replay Invariant
Same Input + Same Gate Decisions Same Digest + Verifiable Evidence

BQL exists because latency claims without deterministic replay are not engineering evidence. They are anecdotes.

Against latency theater

// rejected
  • one-off benchmark scripts
  • manual spreadsheets
  • unstable replays
  • tail metrics without sample validity
  • incident logs that cannot reconstruct exact state
  • “it was faster on my machine” engineering
03Who this is for

Built for teams where one unexplained regression invalidates a release.

01

Head of Low-Latency Infrastructure

A release made latency worse and nobody can prove why.

02

Quant Platform Lead

Backtests and replays differ across environments.

03

Market Data / Exchange Systems Engineer

Incidents need deterministic reconstruction, not screenshots.

04

CTO / Engineering Director

Performance gates must fail CI before regressions reach production.

05

Compliance / Risk Technology Lead

Evidence needs to be reproducible, retained, and reviewable.

For teams where one unexplained latency regression can invalidate a release, BQL turns replay, latency, and evidence into a CI-enforced contract.

04Architecture

The deterministic proof path is explicit.

STAGE 01

Binary capture

ITCH feeds and gen_synth workloads enter as controlled event streams.

STAGE 02

Replay harness

The C++20 scheduler applies events through the same deterministic path.

STAGE 03

Order book

Canonical symbol IDs normalize ordering and shard-count behavior.

STAGE 04

DEG / SCM

Gate decisions are journaled so protection actions replay exactly.

STAGE 05

Evidence bundle

Hashes, benchmark artifacts, metrics, reports, and manifests package the proof trail.

Highlight
Replay invariant = Event Stream + Gate Decisions + Canonical Aggregate Digest
05Performance

Normalized latency tiers, not benchmark theater.

TierScopeCurrentThroughputStatus
Tier AMatch-only: parse + book update~510ms / 1M events1.96M events/secBaseline established
Tier BIn-process end-to-end~813ms / 1M events1.23M events/secSeparated
Tier CFull proof pipeline w/ 128MB journal~836ms / 1M events1.20M events/secEvidence pipeline
100–200× faster

Tier C total runtime improved from ~60–120s to ~0.5–0.9s.

128× faster journal path

Journal overhead reduced from 17.4s to 136ms.

0 synchronous flushes

Removed hot-path flushes and buffered journal writes.

73× throughput increase

Throughput improved from 2.1 MB/s to 153 MB/s.

13.7ms max journal latency

Down from 2,518ms in Phase 3.

100% determinism

Digest 0x722112ad4c431cb4 verified across measured tiers.

Benchmark context guard. Benchmarks validate deterministic replay and regression-detection behavior under controlled workloads. They are not a claim of production exchange throughput without environment-specific validation.

06Methodology

Methodology you can audit, line by line.

Golden-state validation

deterministic fixtures and digest checks.

Tail latency purity

p99.9 and p99.99 measured with validity flags.

CI performance contract

p50 / p95 / p99 / p999 budgets fail builds on regression.

Evidence retention

benchmark JSON, Prometheus metrics, reports, manifests, and provenance bundles.

CPU pinning

core 3 default on Linux to avoid core 0 interrupt noise.

Reproducible reporting

HTML analytics report, Prometheus textfile metrics, JSONL artifacts.

07Selective Coordination Mode

Smallest breaker trips first.

Selective Coordination Mode brings the “smallest breaker trips first” principle from power systems into trading-engine protection logic. Instead of halting everything when one subsystem slows down, BQL disables or sheds only the affected zone first, preserving cleaner incident boundaries and replayable decisions.

Shed Ladder
Telemetry
Journaling
Risk Checks
Core Match
HALT
NORMAL

all features enabled

WARN

early breach signals detected

DEGRADED

optional work reduced

SHED

non-critical work aggressively shed

HALT

fail-fast and preserve incident artifacts

08Evidence

Evidence you can inspect, diff, and retain.

Artifacts
  • bench.jsonl
  • metrics.prom
  • report/index.html
  • sha256 manifests
  • benchmark CSV / log / summary
  • system information
  • compiler information
  • canonical ID evidence docs
  • signed or packaged release bundles
Reproducibility status
  • Public harness available
  • Golden digest verified
  • Benchmark bundle generated
  • External replication: seeking review by low-latency C++ engineers, quant infrastructure reviewers, and reproducibility reviewers.
09Enterprise

Enterprise BQL 2.0 Pilot

The public version is the proof harness. The enterprise version is the production-shaped integration layer for real market-data workflows, deployment controls, dashboards, and support.

DURATION
4–6 weeks
INPUT
one representative market-data or replay workload
NEXT STEP
30-minute technical review, workload fit check, pilot scope, and evidence plan
Deliverables
  • deterministic replay harness
  • baseline latency profile
  • CI latency gate
  • reproducible benchmark report
  • Prometheus/Grafana dashboard plan
  • legal/compliance evidence bundle
  • deployment recommendation
  • executive readout
Features
  • real ITCH/FIX ingest
  • Prometheus/Grafana dashboards
  • CI/CD integration templates
  • enterprise auth/SSO
  • reproducible benchmark reports
  • evidence bundles for legal/compliance
  • customer pilots/reference users
  • clear commercial license and support terms
10Licensing

Open proof harness. Commercial production path.

The public repository is available for research, review, education, and proof-harness evaluation. Production or commercial deployment should follow the commercial license and enterprise support path.

11Founder

Founder.

Built by Jean Blanc.

Blanc Quant Systems is led by Jean Blanc, an award-winning engineer, quantitative systems builder and infrastructure strategist with 15+ years at the intersection of reliability engineering, asset performance, risk analytics, R&D and data-driven innovation.

Jean’s career has been built around high-consequence infrastructure: systems where reliability, traceability, performance and evidence matter. That background shapes the design philosophy behind Blanc Quant Systems.

The company is not being built as a generic market-technology project. It is being built from an infrastructure mindset: deterministic workflows, reproducible results, auditable performance, transparent benchmarking and proof-oriented system design.

In serious systems, performance claims are not enough. The system must be able to prove itself.

That means every major capability is designed around evidence: deterministic input handling, replayable execution, measurable latency, benchmark artifacts and a clear chain from input to output. Blanc Quant Systems is built for environments where speed matters, but where credibility, repeatability and auditability matter just as much.

15+ years

reliability engineering, asset performance, risk analytics, R&D, and data-driven innovation

High-consequence systems

environments where reliability, traceability, performance, and evidence matter

Proof-oriented thesis

performance claims are not enough; the system must prove itself

12Contact

Ready to turn latency proof into an enterprise evidence contract?