Summary

KR-P2-CLEANUP ST1 of 5. KR-P2-I-integration ST4 (deferred from #34) — now unblocked by KR-P2-E shipping the SeaTicketPoller. The operational-state machine moves automatically as the consumer loop turns.

Transitions

Both the normal-release path and the lease-lost path land back at READY since the claim is gone from the runtime's perspective. The TRANSITION_TABLE only has one ACTIVE → READY arrow (trigger column = "claim released"); the trigger string is free-form for the chain-event payload, so both paths share it.

Fail-soft per spec §2 ST1

A new _signal_operational_state(target, trigger) helper wraps each transition:

• get_holder() returns None (no agent session has yet run wire_operational_state, so the holder is uninitialized) → DEBUG-log + skip. Consumer loop still claims + releases cleanly.
• holder.transition_to raises → WARN-log with target + trigger; cycle continues.

Operational-state observability is best-effort during the consumer loop. Substrate-side records (sea_ticket.claimed_for_edit / sea_ticket.claim_released + the existing kora.sea_ticket.resolved emit from KR-P2-E ST4) remain the durable trail.

Tests — tests/test_sea_ticket_poller_operational_state.py (340 LOC, 4 cases)

• Full cycle: [("ready","active","claim acquired"), ("active","ready","claim released")] fired in order; holder lands back at READY.
• Lease-lost branch: both transitions still fire (claim → ACTIVE; lease lost → READY) even though kora__release_claim never fires substrate-side. Only one MCP call (the claim).
• Fail-soft, holder uninitialized: full claim + release cycle runs; DEBUG-log "holder not initialized" lines acknowledge the skip.
• Fail-soft, transition_to raises: substrate cycle still completes both calls; two WARN-log lines (one per attempted transition).

Tests are isolated in their own file (test_sea_ticket_poller_operational_state.py) — they use init_holder + per-test _reset_holder_for_tests rather than mocking, so the singleton's first-wins semantics don't bleed across tests. The existing tests/test_sea_ticket_poller.py stays unchanged.

Honest scope

• Holder initialization is out of scope. wire_operational_state(provider) is called from agent/agent_init.py per-session, not from sea_ticket_poller_lifecycle.py. Gateway-only deployments (poller running, no agent sessions yet) stay observationally quiet — that's the spec's accepted trade. A future "init holder at gateway startup" change is its own bucket (and would need to dedup the per-session listener to avoid N-times emit).
• No changes to agent/operational_state* modules.
• No changes to the kora.sea_ticket.resolved / kora.operational_state.transitioned event payload shapes.

Sub-task chain (this bucket — all independent)

Each ST branches from main independently — no cascade-merge dance.

Test plan

• CI green on pytest tests/test_sea_ticket_poller.py tests/test_sea_ticket_poller_operational_state.py
• Once merged, after an agent session has wired the holder: substrate event_log shows kora.operational_state.transitioned events with payload.trigger ∈ {"claim acquired", "claim released"} interleaved with sea_ticket.claimed_for_edit / sea_ticket.claim_released

🤖 Generated with Claude Code