Analyze occurrences without losing the interval evidence.

Suppose target has [10:00, 10:30) and against has [10:02, 10:10) plus [10:12, 10:28) in the same lane. A two-minute stitch forms one against Episode with two fragments. Its envelope is [10:02, 10:28), its active magnitude is the union of the two fragments, and its internal-gap magnitude is two minutes. Lower the tolerance below two minutes and it remains two Episodes instead.

Release availability differs: the repository's 0.2.0-preview.1 release set includes the C# Episode API and Spanfold.Artifacts package, while its CLI remains checkout-only. The Rust spanfold library and spanfold-cli 0.1.1 are published.

using Spanfold;
using Spanfold.Episodes;

var result = history
    .CompareEpisodes("Provider outage QA")
    .Target("provider", selector => selector.Source("provider-a"))
    .Against("detector", selector => selector.Source("detector-b"))
    .Within(scope => scope.Window("DeviceOffline"))
    .Normalize(normalization => normalization.OnPosition())
    .StitchGapsUpTo(1L)
    .RelateWithin(0L)
    .Run();

var splits = result.RelationsOfKind(EpisodeRelationKind.Split);
var neutral = result.Summary;

// Only do this when the target really is authoritative.
var scorecard = result.AsReference();

use spanfold::{
    ComparisonNormalizationPolicy, ComparisonScope, ComparisonSelector,
    EpisodeRelationKind, TemporalTolerance,
};

let result = history
    .compare_episodes("Provider outage QA")
    .target(
        "provider",
        ComparisonSelector::for_source("provider-a"),
    )
    .against(
        "detector",
        ComparisonSelector::for_source("detector-b"),
    )
    .scope(ComparisonScope::window("DeviceOffline"))
    .normalization(ComparisonNormalizationPolicy::default_policy())
    .stitch_gaps_up_to(TemporalTolerance::processing_positions(1)?)
    .relate_within(TemporalTolerance::processing_positions(0)?)
    .run()?;

let splits = result.relations_of_kind(EpisodeRelationKind::Split);
let neutral = result.summary();

// Only do this when the target really is authoritative.
let scorecard = result.as_reference();

For event-time analysis, select event-time scope and normalization in both runtimes and use a duration/timestamp-tick tolerance with one compatible clock. Portable schema version 1 does not carry that clock contract, so event-time Episodes currently remain a direct-API journey.

{
  "schema": "spanfold.episode.analysis",
  "schemaVersion": 1,
  "name": "Provider and detector offline episodes",
  "target": { "name": "provider", "source": "provider-a" },
  "against": { "name": "detector", "source": "detector-b" },
  "windowName": "DeviceOffline",
  "normalizationAxis": "processingPosition",
  "stitchTolerance": 1,
  "relationTolerance": 0,
  "liveHorizon": 14
}

The accompanying flat window JSON Lines use string keys, string-or-null partitions, non-negative processing positions, and a numeric or null endPosition. Version 1 rejects equal target and against sources, negative tolerances or horizons, and timestamp normalization. Portable execution is intentionally narrower than the direct APIs.

{"windowName":"DeviceOffline","key":"device-1","source":"provider-a","partition":"eu","startPosition":1,"endPosition":4}
{"windowName":"DeviceOffline","key":"device-1","source":"provider-a","partition":"eu","startPosition":5,"endPosition":8}
{"windowName":"DeviceOffline","key":"device-1","source":"detector-b","partition":"eu","startPosition":2,"endPosition":7}
{"windowName":"DeviceOffline","key":"device-2","source":"provider-a","partition":null,"startPosition":10,"endPosition":12}
{"windowName":"DeviceOffline","key":"device-2","source":"detector-b","partition":null,"startPosition":11,"endPosition":null}

dotnet run \
  --project packages/dotnet/src/Spanfold.Cli/Spanfold.Cli.csproj \
  -- episodes episode-plan.json windows.jsonl --format json

dotnet run \
  --project packages/dotnet/src/Spanfold.Cli/Spanfold.Cli.csproj \
  -- episodes episode-plan.json windows.jsonl --format markdown

spanfold episodes \
  episode-plan.json windows.jsonl --format json

cd packages/rust
cargo run -p spanfold-cli -- episodes \
  ../../episode-plan.json ../../windows.jsonl --format markdown

Result v1 carries the analysis and window names, axis, tolerances, evaluation horizon, target and against source metadata, each side's summary and ordered Episodes, the neutral comparison summary, and exhaustive relations referencing those Episode indexes. The repository includes the complete shared plan, window JSONL, and expected JSON result. Use the portable result for cross-runtime exchange; keep a native direct result alongside it when you need full fragment objects and source-record lineage.