Customize mappings

Trove’s Tier A metrics are the cross-harness comparison layer — a fixed, normalized schema so that “cost per turn” or “tokens this week” means the same thing whether the signal came from Claude Code, Cursor, or Aider. The five Tier A metrics are:

Metric	Type	Key attributes
trove.harness.events	counter	event.kind (chat.turn, tool.call, shell.exec, file.edit, session.start/end)
trove.harness.tokens	counter	direction (input/output), model
trove.harness.cost.usd	counter	model, cost.method (exact / estimated)
trove.harness.turn.duration	histogram (s)	event.kind
trove.harness.errors	counter	error.kind (rate_limit, auth, tool_failure, network, policy, unknown)

Everything a harness emits natively (its Tier B metrics) always passes through untouched. Mappings only control how raw signals are additionally projected onto Tier A. You edit those rules on the Mappings tab.

Two kinds of rule

How a harness contributes to Tier A depends on its coverage type:

• Synthesis rules (synthesize-from-native) — for native-OTel harnesses. Each rule names a native metric (for example claude_code.token.usage), the Tier A metric it feeds (tokens), and an attribute map from the raw keys to Tier A keys. Trove implements these as a collector-side transform / metricstransform step, so synthesis is additive and never disturbs the native metric.
• Hook rules (hook-rule) — for watcher/wrapper harnesses. Each rule matches a raw event name (for example afterAgentResponse) and either emits into a Tier A metric with a set of attributes, or emits nothing. Setting a rule’s emit to “nothing” is how you suppress double-counting — for example mapping beforeSubmitPrompt → (none) and counting the turn only on afterAgentResponse.

Adjust a rule

1. Open the Mappings tab and pick the harness you want to tune. Each harness shows its current Tier A rows.

   Note

   Screenshot coming soon.

2. Edit a row. Typical edits:

   • Retarget an event. Change which event.kind a raw event maps to (for example, classify a shell-execution event as shell.exec rather than tool.call).
   • Stop a double-count. Set a redundant raw event’s emit to “nothing” so only one of a before/after pair contributes a chat.turn.
   • Toggle a synthesis row. Turn synthesize-from-native on for the tokens or events metric so a native counter is also copied into Tier A; turn it off for a clean native-only namespace.
   • Disable a metric for one harness. Turn off a harness’s cost.usd row entirely if you don’t trust its estimate.

3. For cost, adjust the rate table if your model isn’t priced the way you pay. Add a model entry with your own per-1k input/output USD rates; Trove uses those for the matching model attribute when computing trove.harness.cost.usd. Rows priced from a rate table report cost.method = estimated; rows that carry a real cost from the harness report cost.method = exact.

Caution

The five built-in Tier A metrics are locked — you can’t rename or delete them, because cross-user and cross-team dashboards depend on the schema meaning the same thing everywhere. You can extend the catalog with your own custom metrics, but a custom name only carries across teams if the consuming side has the same catalog. (Custom metric rules are honored by native-OTel harnesses today; watcher/hook harnesses don’t yet honor them.)

Preview the diff before applying

Trove doesn’t apply mapping edits blind. Before anything reaches the running collector, you can preview exactly how your changes alter the generated pipeline.

1. Make your edits. Trove computes a live diff of the generated collector config — the new/changed transform and metricstransform stanzas your rules produce — so you can read precisely what will change before committing.

2. Review the diff. Confirm the rules synthesize the metrics you expect and nothing you didn’t intend.

3. Apply changes. Trove validates the mapping, persists it to the single versioned state.json, regenerates collector.yaml, and signals the collector to reload. Expect a brief 1–3 second collector blip while the new config takes effect.

Reset to defaults

Every harness ships with a sensible default mapping baked in. If you’ve edited yourself into a corner, you can always go back:

1. On the harness’s mapping, choose Reset to defaults.

2. Trove restores that harness’s shipped default rows, regenerates the collector config, and reloads. Your edits to other harnesses are untouched.

Tip

Defaults are deliberately conservative about cost. Token and event synthesis are commonly on by default so cross-harness dashboards populate immediately; cost synthesis is often off by default to avoid a confusing double-count until you opt in.

Normalize costs & find dead seats Put Tier A cost and events to work across every harness.

Enable a harness Turn telemetry on before you start mapping it.

Mappings reference The full Tier A schema and rule grammar.

Add identity tags Attribute Tier A metrics to people for per-user dashboards.