#3411 Telemetry records swamp behavior and swamp model invocations under different command names

Summary

swamp behavior run and swamp model method run run the same code, but telemetry records them under different command / subcommand values. A comment in the code says swamp-club scores XP and quests on those fields, so a user who follows the preferred spelling may be scored differently from one who uses the legacy spelling. The swamp-club side was not checked; this needs confirming there.

Found in an audit of the two command groups (see also swamp-club#3409). This is the only place the audit found where the two spellings behave differently rather than just read differently.

What is recorded

1. Typed spelling is kept in command / subcommand. swamp behavior run … records command: "behavior", subcommand: "run" (or "behaviour" for the alias). swamp model method run … records command: "model", subcommand: "method". Only commandPath is canonicalised. See the doc comment on the resolver in src/cli/telemetry_invocation.ts: "command, subcommand and the command words in args keep the spelling as typed, because swamp-club scores XP and quests on command/subcommand and an alias can score differently from its canonical name." This is pinned in src/cli/telemetry_invocation_test.ts.
2. Synthetic entries are always model-spelled. Runs recorded for workflow steps and for serve-executed runs use command: "model", subcommand: "method", commandPath: ["model", "method", "run"], whichever spelling the user typed (src/libswamp/workflows/telemetry_bridge.ts, src/serve/telemetry.ts).

So one logical action, "run a method", can arrive as model/method, behavior/run or behaviour/run.

To confirm

• Does swamp-club score behavior/run and behaviour/run the same as model/method, for XP and for quests that count model method runs?
• The same question for the other twins: behavior create|get|edit|delete|validate|evaluate|search|cancel|describe|history|output|type against their model forms.

Possible fixes

• Server side: treat the behavior spellings as equivalent when scoring.
• Client side: send a canonical command name for scoring, keeping the typed spelling in a separate field if it is still wanted.

Related, not a defect

The OTEL span attribute swamp.args also differs: swamp model output get <id> records get, while swamp behavior output get <id> records <REDACTED>. A comment in src/cli/telemetry_integration.ts (behaviorArgSchemas) says this is deliberate, because behavior output <name> and behavior history <name> are shorthands that take a name first. It affects span attributes only, not recorded telemetry.

Acceptance

• A method run is scored the same whether typed as swamp model method run, swamp behavior run or swamp behaviour run, and the same holds for the other twin commands.
• The decision (server-side equivalence or client-side canonical name) is recorded where the telemetry contract is documented.