Narrativ's data engine turns messy, high-volume activity into market-grade outputs. We do that by separating two things:Signals = normalized events (what happened)
Indices = computed metrics (what it means over time)
At a glance#
Signals and indices are always computed and delivered under the active permission state and with traceable provenance.
Signals#
A signal is a canonical, normalized event derived from raw inputs. Signals are designed to be:Consistent: different sources map into the same event shape
Attributable: each signal carries provenance (source + timestamp + permission state)
Composable: signals can be aggregated into features and indices
Auditable: signals can be traced back to the input class and processing path
engagement events (views, likes, follows)
content interactions (shares, comments, saves)
narrative markers (topic mentions, sentiment changes)
behavioral indicators (frequency shifts, session changes)
Indices#
An index is a computed metric built from signals over a defined time window. Indices are designed to be:Windowed: computed over explicit time horizons (e.g., 1h, 24h, 7d)
Stable: less sensitive to noise and duplicated inputs
Explainable: methodology is defined and readable
Versioned: changes are explicit, not silent
Every index has an explicit definition that typically includes:
Name + purpose (what it measures)
Inputs (which signals, which sources, which permission classes)
Filters (cohort, geography, category, policy constraints)
Window + sampling (time range and update frequency)
Normalization (how raw variance is standardized)
De-duplication and linking (where applicable)
Weighting (how signals contribute to the final value)
Methodology version (so customers can reproduce and compare)
Quality flags (confidence and anomaly indicators where applicable)
Why we separate signals and indices#
Most systems either ship raw events (too noisy) or ship opaque scores (not auditable). Narrativ separates signals from indices so outputs can be:High-signal: indices reduce noise while preserving meaning
Permissioned: permitted use is enforceable at every step
Auditable: outputs can be traced back to sources and consent state
Composable: the same signal layer can power multiple indices and products
B2B delivery: signals, indices, alerts, exports
Market resolution: indices as objective settlement sources
Rewards accounting: indices and signal quality contribute to participation measurement
Provenance and permissions#
Signals and indices are never "free-floating." They are always computed and stored with:source provenance (managed vs user-permissioned, plus origin metadata)
timestamp + window context
permission version (what was allowed at the time of processing)
auditability hooks (so eligibility, payouts, and deliveries can be explained)
For details on how inputs are ingested and processed, see Data Engine.
For how permitted use is defined and enforced, see Consent and Permissions.
