Concepts
Standalone explanations for every major isplay concept.
isplay is a replayable evidence system for agent runs. These pages explain the vocabulary behind the CLI, SDK, adapters, HTTP API, replay engine, experiment runner, and analyst reports.
Object model
Projects, runs, events, artifacts, model calls, tools, checkpoints, branches, replays, experiments, and analysis runs.
Capture model
What capture records, why it separates intent from execution, and how redaction fits in.
Event streams
Ordered events, sequence numbers, event families, custom framework events, and append semantics.
Context inventory
Searchable model-visible, tool-visible, state-only, and metadata-only decision inputs.
Checkpoints
State snapshots, branch points, hashes, schema versions, and practical placement strategy.
Branches and interventions
How hypotheses become checkpoint forks and targeted patches.
Replay policies
Model, tool, drift, budget, and temperature policy tradeoffs.
Fixtures and provenance
How missing divergent tool outputs are represented, satisfied, and qualified.
Experiments, effects, statistics
Hypothesis batches, arms, trials, low-N output, statistics, and ranked effects.
Evidence validity reporting
Validity labels, evidence nodes, evidence edges, and report wording.
Adapter capabilities
Native replacement, proxy replacement, post-result replacement, blocking, and observability-only modes.
Storage model
Postgres tables, JSONB projections, artifact files, hashes, and catalog reconstruction.
Schema vs executor support
How to read features that are modeled in schemas but not fully executable in the local engine.
How To Read These Pages
Use concept pages for durable mental models. Use guides when you need a task recipe. Use reference when you need exact command names, route paths, enum values, fields, and limitations.
Evidence discipline
Most isplay concepts exist to prevent overclaiming. A replay is not automatically proof, a fixture is not automatically reality, and an experiment with one trial is not automatically a statistically supported conclusion.