…ll — closes D-kr2-st4

Mechanical follow-on swap unlocked by K-9 (substrate `f8487059`) +
KR-7a (`b6415692`). events.py:emit_kora_event body replaces
`raise ChainEventEmitNotAvailableError()` with
`await mcp_client.invoke('kora__append_event', {...})` returning the
substrate-assigned event_id. ~30 LOC across events.py + provider.py +
tools/iso_node.py + tests + docs.

Production-test posture per IsoKron PM #27:
* K-9's kora__append_event handler is a notImplementedHandler stub on
  substrate main; substrate-team's dispatch tier (queued) un-stubs +
  resolves Layer-A wsk_* auth → Layer-B actor_kind='kora'.
* KR-7's code shape is sound; mock tests verify correctness.
* Production deploys wait on substrate dispatch tier landing —
  identical posture to KR-7a (token-not-yet-issued; ships green).
* Verify-at-first-live-emit step per spec § 3: confirm
  event_log.actor_id resolves to the 0076-seeded canonical Kora actor
  (actor_kind='kora' AND workspace_id=<Flynn clerk_org_id>); if it
  resolves to a token-UUIDv5 instead (cowork-claude-pm precedent),
  small substrate patch needed.

events.py:
* emit_kora_event body: live call to mcp_client.invoke. K-9 contract
  output: {'event_id': '<uuid>'} — projected to a plain str return.
* Defensive: None mcp_client raises ValueError ("caller must resolve
  via IsoKronConnection.get_mcp_client() before calling"); unexpected
  response shape raises RuntimeError.
* mcp_client param is now required (was Optional with deferred-error
  default). Forward signature stays caller-stable from KR-2 ST4.
* ChainEventEmitNotAvailableError class kept exported tagged
  [kora.isokron.deprecated] for one release so any downstream pinned
  tests resolve. Will be removed when KR-N audits show no remaining
  references.

provider.py:_attempt_chain_event_emit:
* Stop catching ChainEventEmitNotAvailableError (no longer raised).
* Fetch IsoKronMCPClient via self._connection.get_mcp_client() (wired
  in KR-7a); failure to obtain the client is caught + logged at
  ERROR ([kora.chain.emit.failed] ... MCP client unavailable: ...)
  so lifecycle hooks (on_session_end / on_delegation / on_memory_write
  / sync_turn) stay alive on substrate downtime.
* Catch IsoKronMCPInvocationError + any defensive Exception at the
  lifecycle boundary — log at ERROR (not WARNING) since chain-event-
  emit failure is operator-visibility-worthy per PM-lean ("substrate-
  side issue worth surfacing"). Sessions don't crash; emits drop with
  a loud log line.
* Successful emits log INFO [kora.chain.emit] with the event_id and
  invalidate the per-workspace events cache so the next
  system_prompt_block §6 re-reads.

tools/iso_node.py:_handle_iso_node_supersede:
* The kora.node.superseded emit call site previously imported
  emit_kora_event directly and passed mcp_client=None. Refactored to
  route through provider._attempt_chain_event_emit so all chain-emit
  paths share the same error-handling + cache-invalidation logic.

Tests:
* test_events.py — replaced test_emit_kora_event_raises_deferred_
  write_error with four MCP-call-path tests:
  - invokes 'kora__append_event' with expected arg shape; returns
    the substrate event_id.
  - propagates IsoKronMCPInvocationError on substrate-side error.
  - rejects None mcp_client with ValueError (defensive).
  - rejects unexpected response shape with RuntimeError (defensive).
  Plus one deprecation-runway test asserting the
  ChainEventEmitNotAvailableError class is still importable post-KR-7.
* test_provider_end_to_end.py — _FakeProviderConnection now exposes
  get_mcp_client() returning a _FakeMcpClient that records invokes.
  E2E asserts both emits (on_delegation + on_session_end) hit
  kora__append_event with the spec-pinned tool name + arg shape; logs
  [kora.chain.emit] INFO each time. Scratchpad-write deferrals stay
  at 3 (D-kr2-st3 still open).
* test_isokron_provider_skeleton.py — unchanged (didn't reference
  D-kr2-st4 specifics).

BUILD_DEVIATIONS:
* D-kr2-st4-no-chain-emit-mcp-tool moved from Open to Closed with
  Rule-5 spec-quote + production-test posture + deprecation-runway
  note. 3 deferrals remain open (D-kr2-st2 / D-kr2-st3 / D-kr3-st2).

README "Operator pitfalls":
* Deferred-surface table trimmed 4 → 3 rows. Recently-closed list
  gains the KR-7 closure under D-kr2-st4.
* Individual "Chain event emission deferred" pitfall replaced with
  the KR-7-shape description: routes via kora__append_event;
  successful emits log [kora.chain.emit] INFO; substrate failures log
  [kora.chain.emit.failed] ERROR; production posture flagged.
* D-kr2-st2 row updated — K-7 already merged 2026-05-20; PM drafts
  the swap bucket.

Local gates:
* ty check — 7,337 diagnostics, zero-delta vs KR-7a baseline.
* pytest tests/plugins/memory/ — 351/351 passing.
* Full suite via xdist (-n auto): 24,561 / 203 failed / 52 errors /
  129 skipped. Higher xdist isolation variance this run (kanban_tools
  test_board_param errors aren't reproducible in isolation; none
  touch isokron).

3 deferrals still open pending CC#1 substrate buckets:
* D-kr2-st2-capability-matrix-mirror ← KR-N swap (K-7 merged; PM drafts)
* D-kr2-st3-no-scratchpad-write-mcp-tool ← K-8
* D-kr3-st2-no-relationlink-write-mcp-tool ← K-10

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>