SignalOps
Deterministic signals from governed evidence
SignalOps ingests, processes, correlates, and replays time-based and stream-based observations. Versioned algorithms turn normalized events into signals that carry their own evidence, so analysts review what merits review.
Operating principles
Turn data into signals. Turn signals into evidence. Conviction belongs to the analyst.
Deterministic
Versioned logic produces reproducible outcomes. The same evidence, at the same version, yields the same signal, every time.
Evidence-backed
Signals retain the facts and lineage behind them: which observations fired, from which sources, under which algorithm version.
Analyst-controlled
Review is explicit. A signal is not a recommendation and never becomes an interruption without policy saying so.
Architecture
Three layers, one boundary
SignalOps is a standalone subsystem: it does not touch document ingestion code paths, and its failures never degrade Ask, retrieval, or graph construction on the main platform.
Core platform
Source registration, schema contracts, pipeline orchestration with idempotency, retries, replay, and backpressure. PostgreSQL and TimescaleDB persistence; broker-abstracted topic routing.
Processing runtime
Statistical detection, change and drift detection, semantic enrichment, and windowed computation (tumbling, sliding, session, calendar) over durable stream state.
Use-case packs
Domain behavior without changing core infrastructure. MarketOps and CyberOps run on the same primitives: signals, evidence ledgers, insight candidates, lifecycle policies.
Why time series need relationships
Streams answer "what." Relationships answer "so what."
A price series says what happened. An options surface says what the market expects. A filing says why the business changed. A sector rank says where the pressure sits.
Time-series tooling treats each stream as an island. SignalOps persists every stream as governed evidence first, then correlates: state transitions across feature domains, convergence rules that require independent sources to agree, and graph proposals that connect events to entities.
The result is relational deduction across heterogeneous environments: not "this metric moved," but "this valuation shift, this options behavior, and this sector context agree, and here is the evidence."
The hierarchy
Data → Signals → Evidence → Opportunity → Analyst decision → Outcome → Signal assertion → Validation → Calibration
Every stage is persisted and replayable. The loop closes: confirmed signals become testable assertions, assertions are validated against outcomes, and validation feeds algorithm calibration.
Platform capabilities
Signals with receipts
Signal definitions and families
Named, versioned signals (for example, market_data.options_daily_prices, security.identity_behavior) with confidence aggregation across detectors.
Lifecycle with noise control
Signals are durable, queryable detector facts. Insights are deduplicated, explainable work items. Alerts are exceptional, policy-driven interruptions. A valid signal is never silently discarded, and never automatically becomes an alert.
Replay and backtesting
Historical detector evaluation with point-in-time correctness: signals can be re-derived from persisted evidence and audited after the fact.
Engine federation
SignalOps proposes graph mutations, event artifacts, and insight candidates through approved Engine extension contracts. The Syncratic Engine remains the authority for accepting them.
Use case · MarketOps
Deterministic market surveillance for research teams
MarketOps is the flagship SignalOps domain: equity, options, technical, and financial evidence turned into a consistent daily review process. It is research decision support, not a trading system: it places no orders, alters no portfolios, and issues no recommendations.
Market State Intelligence
One canonical, versioned snapshot per asset per session, composed of normalized feature domains. State transitions persist level change, z-score, percentile, acceleration, regime change, and cross-bucket migration: not raw ticks, but explainable movement.
Strategic algorithms
Valuation Composite (explainable 0-10 relative valuation) and Distressed Opportunity Scoring (strategic research rank combining value, operating quality, cash generation, and debt profile), refreshed weekly from financial statements.
Tactical algorithms
Exhaustive Reversal (extended movement with fading or climactic participation), implied-volatility regime context across the 30/60/90-day options surface, and Sector Rotation Intelligence ranking relative strength and momentum.
Convergence, not noise
The Convergence Opportunity Builder only elevates review items when at least two independent sources agree. Earnings Opportunity Intelligence scores pre-earnings setup quality from persisted evidence, weighting strategic and tactical signals by proximity to the event.
Signal Assurance Framework
Signals that are later graded
A confirmed signal becomes a testable assertion with an expected outcome. The Signal Assurance Framework validates those assertions continuously: time-to-materialization as the primary metric, path-quality excursion tracking, and evaluation across assets, regimes, score bands, and algorithm versions.
Validation preserves point-in-time correctness: no future information leaks into the grade. The result closes the loop that most analytics stacks leave open: the system learns, from evidence, which of its signals deserve weight.
Boundary
Scores are research artifacts, not price targets or recommendations.
Every MarketOps surface states this. Determinism and evidence are the solution; conviction stays with the analyst.
Also on SignalOps
CyberOps applies the same primitives to security evidence: firewall observations retained durably, deterministic detections, and lifecycle policies that separate record-only evidence from investigation work.
Governed knowledge assurance
See signals with their evidence.
We will walk the SignalOps pipeline on live MarketOps data: ingestion, deterministic detection, the review queue, and signal assurance, end to end.