Documentation / Examples

Transcript Lake examples

This is the canonical catalog of supported user outcomes for the development channel. Each example uses the installed transcript-lake interface, names its risks and side effects, includes observable verification and a representative failure path, and owns its

This is the canonical catalog of supported user outcomes for the development channel. Each example uses the installed transcript-lake interface, names its risks and side effects, includes observable verification and a representative failure path, and owns its cleanup decision.

Shared prerequisites#

  • macOS. Building the CLI needs a Rust toolchain at version 1.85 or newer; running the installed binary needs nothing else.
  • Installed Transcript Lake development build or exact release archive.
  • An operator-owned LAKE_DATA root.
  • DuckDB 1.5.x only for query, signal, and compact examples.
  • Oko only for reindex examples.
  • No production credentials are required or accepted.

Every example assumes the user has read onboarding. Replace placeholders only with operator-owned local paths or exact immutable versions. Never use a shared production Lake for a destructive or recovery example.

Risk labels#

  • Read-only: does not intentionally mutate product or provider state.
  • Local mutation: writes only operator-selected local product state.
  • Derived mutation: writes rebuildable Parquet or export artifacts.
  • Destructive/recovery: changes which state root is active or removes retained local data; requires explicit operator decision.
  • Provider-facing: reads local files owned by coding-agent runtimes; never modifies them.
  • External tool: invokes DuckDB or Oko and inherits that tool's local behavior.

No catalog example is billable, network-facing at runtime, or credentialed. Release download and publication are separate network operations.

Coverage matrix#

ActorOutcomeInterfacePreconditionsRiskCanonical example
EvaluatorProve a native live append without private dataCLI/synthetic filesIsolated invented fixtureLocal mutationSynthetic live append
EvaluatorAudit all masking classes and a derived mirrorCLI/files/DuckDBIsolated invented fixture; DuckDB for compactRead-only plus optional derived mutationSynthetic masking audit
New operatorInstall and create the first live archiveCLIClean supported host; local sessions optionalLocal mutation, provider-facingFirst local archive
OperatorRead safe zero-state guidance and product identityCLIInstalled productRead-onlyInspect zero state
OperatorTour the complete installed command surfaceCLIInstalled product; optional tools per commandMixed, explicitly stagedCLI tour
OperatorKeep an existing Lake currentCLIExisting valid LakeLocal mutation, provider-facingLive stream
AnalystRetrieve cross-runtime session evidence and locate events by literal textCLI/DuckDBLake partitions; DuckDBRead-only, external toolQuery sessions
OperatorRead one past conversation back in full, in orderCLI/DuckDBLake partitions; DuckDB; a session idRead-only, external toolRestore a conversation
AnalystJoin Lake with Oko signal stateCLI/DuckDBLake, DuckDB SQLite extension, Oko indexRead-only, external toolCross-source signals
OperatorCreate and clean Parquet mirrorsCLI/DuckDBLake partitions; DuckDBDerived mutation, external toolCompact to Parquet
Oko operatorReconstruct every supported runtime projectionCLI/filesLake partitionsDerived mutationRebuild Oko projection
Oko operatorReindex Oko after projectionCLI/processCompatible oko-cliDerived mutation, external toolReindex Oko
Hook maintainerStream validated Tama segments without legacy duplicationCLI/filesTama ready directoryLocal mutation, provider-facingImport hook decisions
OperatorRebuild after source rewrite or cursor damageCLIPreserved old Lake; empty replacement rootDestructive/recoveryRebuild into an empty root
OperatorUpgrade or roll back exact artifactsCLI/GitHub releaseImmutable archive, checksum, backupDestructive/recovery, network installUpgrade and rollback
OperatorClean derived state and uninstallCLIStopped stream and writersDestructive/recoveryReset and uninstall
Release ownerBuild attributable immutable release assetsRelease scriptClean exact tag; qualification approvalLocal mutation, publication preparationBuild release assets
OperatorDiagnose invalid input, dependency outage, source failure, and writer conflictCLIScenario-specificRead-only or isolated local mutationRepresentative failures
UserStream Gemini or QwenNot supported; see product boundaries
UserDelete individual events or mutate vendor transcriptsDestructiveNot supported

Selecting and running an example#

  1. Choose one row matching the desired outcome.
  2. Read its status, risk, environment, preconditions, inputs, side effects, and failure path before copying commands.
  3. Pass an isolated root with global --data-dir for local mutation, recovery, or failure work.
  4. Run the stated installed transcript-lake commands; source-tree scripts appear only in the release-maintainer example.
  5. Compare the observable result with the expected shape; do not interpret exit zero alone as success.
  6. Follow the explicit cleanup or retention decision.

Global safety and cleanup#

  • Never point recovery examples at the only copy of a Lake.
  • Never delete vendor transcript stores; examples do not require it.
  • Stop the stream before moving, replacing, or deleting a Lake root.
  • Treat Lake metadata and unmasked short text as sensitive even after secret masking.
  • Examples create no cloud resources and request no credentials.
  • Delete only paths created or explicitly selected by the example.
  • Preserve failure evidence before cleanup when diagnosing corruption or partial state.