What StørmPrep guarantees
StørmPrep guarantees semantic normalization of raw events into canonical envelopes and deterministic feature vectors for StørmAI. Outputs are reproducible for the same schema version and context.
Contract: raw event → canonical feature vector
Deterministic preprocessing with schema governance.
Inputs
Raw events, schema definitions, and trust context for admission.
Processing
Canonical envelopes, semantic mapping, and deterministic feature construction.
Outputs
Canonical events and versioned feature vectors for StørmAI.
How it works
Three steps from raw event to deterministic feature output.
Validate schema
Schema validation and trust checks at admission.
Map to canonical classes
Normalize events into canonical envelopes with preserved semantics.
Construct deterministic features
Produce versioned feature vectors with reproducible transforms.
Interfaces
- Inputs: raw telemetry and event streams.
- Outputs: canonical event envelopes and deterministic feature vectors.
- Contracts: schema versions with signed feature schema governance.
- Latency/ordering: bounded admission ordering with tiered latency targets per stream.
Semantic admission filter + deterministic feature builder.
StørmPrep turns raw events into canonical classes with preserved causality.
It then constructs deterministic features that remain reproducible.
Downstream inference relies on this stability for audit and replay.
Deterministic preprocessing contracts.
- Schema-validated inputs mapped to canonical event classes.
- Deterministic feature vectors for the same event and context.
- Versioned, signed feature schemas verified by StørmTrust.
- Ordering windows and causality preserved from admission.
- Context snapshots recorded for audit and replay.
Determinism guarantees
- Schema validation and canonical envelopes at admission.
- Versioned feature schemas verified by StørmTrust.
- Deterministic transforms with reproducible outputs.
- Drift prevention through schema pinning and change control.
Failure handling
- Reject or quarantine events on schema mismatch.
- Preserve ordering and anti-replay constraints per session.
- Record rejection reasons for audit and replay.
Capabilities
Semantic enforcement with deterministic feature construction.
Typed event classes preserve meaning
Identity, process, network, OT, and admin events retain causality and lineage with explicit class contracts. So what: downstream reasoning preserves meaning.
Inline and asynchronous context
Inline enrichment delivers enforcement-critical context; async joins add forensic depth without altering decision-time inputs. So what: decisions use known context without retroactive changes.
Stable schemas for AI inference
The same event and context produce the same vector, with schema versions signed and verified by StørmTrust. So what: inference is reproducible and auditable.
Provenance for every feature
StørmPrep records trust state, ordering window, and context snapshot references for audit and replay safety. So what: audits can replay with the exact inputs.