Foundation
State windows, the temporal model, and direct history queries establish the evidence vocabulary before any cross-source comparison happens.
Concepts
How Spanfold turns state windows into analysis.
Learn the model in order: record state as windows, understand the time axes, inspect one history, compare two histories, then move into segments, live finality, cohorts, matrices, roll-ups, hierarchy, and advanced analytics.
Problem
Temporal systems fail in the gaps between latest values.
A latest-value table can say a device is healthy now, but not when it was unhealthy, which provider saw the state first, whether another provider missed it, or whether that evidence was available at a decision point. Spanfold's concepts build a vocabulary for those questions.
Learning path
Move from the primitive to the advanced cases.
State windows
Why Spanfold records predicate truth as temporal evidence instead of only latest state.
Temporal model
Processing position, event time, known-at filtering, and live horizons.
Querying history
Inspect one recorded history before comparing sources.
Comparing histories
Target, against, scope, normalization, alignment, and comparator rows.
Segments and tags
Split windows on analytical boundaries while keeping descriptive metadata attached.
Live finality
Clip open windows at a horizon and separate provisional rows from final evidence.
Cohorts
Compare one source against a group treated as a single derived lane.
Source matrix
Read directional pairwise comparisons across a provider set.
Roll-ups
Aggregate child windows into parent windows while preserving the segment context that still matters.
Hierarchy
Explain parent roll-up activity from child contribution windows.
Advanced analytics
Use coverage, gap, containment, lead/lag, as-of, finality, and row aggregations for deeper questions.
Solution
Record first, analyze second.
Spanfold does not ask every comparator to re-derive state. The runtime records the time ranges where predicates were active. Analysis then works over those ranges: normalization, alignment, comparison, export, and explanation.
That split keeps the concepts composable. A roll-up can produce parent windows that are compared later. A cohort can become the comparison side. A source matrix can run the same plan across every provider pair.
How we get there
The ideas layer without changing the primitive.
Every page in this section keeps returning to the same primitive: a named window over a key, source, partition, range, segments, and tags. The advanced features are built by selecting, aligning, deriving, and summarizing those windows rather than replacing the model.
Comparison
Comparing histories introduces target and against lanes, scope, normalization, alignment, and row families such as overlap, residual, and missing.
Context
Segments, tags, and live finality explain when evidence should split, when it should only be labeled, and when it is provisional.
Derived analysis
Cohorts, matrices, roll-ups, hierarchy explanation, and advanced analytics reuse the same rows to answer larger operational questions.