Struct RunReader 

pub struct RunReader { /* private fields */ }

Reader over one run directory.

Implementations

impl RunReader

pub fn open(run_dir: impl AsRef<Path>) -> Result<Self>

Open a run directory. Fails only if the manifest is missing or unreadable — everything else is tolerated per-file.

pub fn dir(&self) -> &Path

The run directory this reader was opened on.

pub fn manifest(&self) -> Result<RunManifest>

Parse manifest.json — the run’s immutable identity record.

pub fn status(&self) -> Result<RunStatus>

Parse status.json — the latest state + heartbeat snapshot.

pub fn info(&self) -> Result<RunInfo>

Listing entry for this run (state includes crash detection).

pub fn events(&self) -> Result<Vec<EventEnvelope>>

All parseable event envelopes, in log order. Torn or unknown lines are skipped; seq gaps let a consumer detect the skips.

pub fn node_timings(&self) -> Result<Vec<NodeSpan>>

Per-node execution spans in event order — the gantt/overlay substrate. Cache hits are standalone spans (a hit node never starts); an unclosed span means the run died mid-node.

pub fn cache_activity(&self) -> Result<CacheActivity>

Cache hit/miss counts, total and per node.

pub fn metric_series(&self, name: Option<&str>) -> Result<Vec<MetricPoint>>

Metric time series, optionally filtered by name. Reads the flat metrics.jsonl tee; falls back to deriving the same points from events.jsonl when the tee is absent.

pub fn health_flags(&self) -> Result<Vec<HealthFlagRecord>>

All HealthFlag events with wall time.

pub fn agentic_activity(&self) -> Result<AgenticActivity>

Agent-level activity, aggregated per step node.

Accounting rule: AgentStepCompleted totals are authoritative (they are cumulative — a resumed run re-counts its replayed effects). A Suspended cost stands in only until a later completion for the same node supersedes it, and turn counts seen on the wire (AgentTurnStarted) are used only for nodes that died without any accounting event at all.

pub fn agentic_timeline(&self) -> Result<Vec<EffectSpan>>

Per-effect execution spans in event order — the gantt substrate for agent runs. Concurrent same-label effects within a turn are matched first-in-first-out, which is the order the driver reports completions in.

pub fn graph(&self) -> Result<Option<Graph>>

The graph this run executed (graph.json), if snapshotted.

pub fn overlay(&self) -> Result<GraphOverlay>

Fold this run’s node spans and health flags into a rendering overlay: status + total duration + cache tier per node, ×N when a node ran more than once, deduplicated flags.

pub fn to_mermaid(&self) -> Result<String>

Mermaid rendering of the run’s graph, annotated with this run’s overlay. Errors if the run has no graph.json snapshot.

pub fn to_graphviz(&self) -> Result<String>

Graphviz rendering of the run’s graph, annotated with this run’s overlay. Errors if the run has no graph.json snapshot.

pub fn to_svg(&self) -> Result<String>

Self-contained SVG rendering of the run’s graph with this run’s overlay — no JavaScript, safe for notebook/report embedding. Errors if the run has no graph.json snapshot.

pub fn study(&self) -> Result<Option<Study>>

The study attached to this run, if any.

pub fn trial_timeline(&self) -> Result<Vec<TrialSpan>>

Trial lifetimes from study.json (empty for non-study runs) — the timeline/gantt substrate for HPO charts.