[vjovanov]

Closes #619

Before, the failed-own-record test could report an ambiguous edit spawns both with only mock:mock:d in its append log. Both invocations had started, but d deliberately failed before delayed c executed its body, so normal abort handling interrupted c before its append. Now the private fixture lets c complete successfully before d deliberately exits 1, preserving the scenario under either startup order.

What this changes

The clarification at §FS-rhei-agent-visit-pairing.5 explains that spawning an invocation does not guarantee its body's side effects survive sibling failure; tests requiring completion before deliberate failure establish that order inside their fixture.

The shared failure body waits up to five seconds for c's successful completed record matching the scenario's task, state and current visit. A prerequisite timeout prints the record path and last observation and exits 97, distinct from d's intended exit 1.

The original test and the new regression share the real scenario. An independent scheduler exercises c-first and delayed-c/d-first startup and snapshots c's record at each deliberate d exit without waiting. Assertions retain both target appends and the ambiguity warning, check durable records, and require only d to retry with a fresh attempt log while successful c and orphan x stay untouched and the ticket remains in work. This continues enforcing §FS-rhei-agent-visit-pairing.2: d's failed own record decides, and successful orphan x cannot answer for it.

Both scheduler record reads explicitly decode UTF-8 so Unicode fixture paths survive non-UTF-8 default text decoding. The change is confined to test fixtures and the specification clarification; production scheduling, pairing, abort policy, public interfaces and persisted formats are unchanged.

Verification

• The regression first failed before the repair at the exact missing-c assertion, with independent records showing both starts, interrupted c, d exit 1 and successful x. It was committed ignored, then enabled by the repair.
• On 34bee098e3, both exact tests passed: agent_target_edit_tests::failed_own_record_coverage_survives_delayed_sibling_startup and agent_target_edit_tests::an_orphan_never_answers_for_an_invocation_whose_own_spawn_failed. The unchanged triage reproducer passed both startup orders.
• Both real tests also passed with Unicode paths under CP1252 text decoding on Linux with UTF-8 filesystem semantics. Both scheduler encoding controls passed. This independently verified the fix for review's only major finding, R1-01; no findings remain unresolved, rejected or deferred. Native Windows/macOS execution remains for hosted CI.
• The final local gate on 34bee098e3 ran the unchanged reproducer and pre-commit run --all-files, both exit 0, in 3m40s total. All configured hooks passed: CI-pinned grund 0.18.0 config/check, staged fissile, formatting, workspace Clippy/build/tests, gate script tests, links and attribution checks. Hosted CI and merge remain for the ship step.

Checklist

• The most-specific spec point is updated and cited with a § marker.
• pre-commit run --all-files passes locally.

AI workflow: `rhei`, 15 agent invocations across 1 model; 8 tasks completed, 8 in progress.

1. github-issues-agent-grounds-rhei-619-implement-7c69c7b6.ticket supervising (visit 1) — cdx, openai/gpt-6.1-sol — 1m13s — 338.2k in / 2.8k out
2. github-issues-agent-grounds-rhei-619-implement-7c69c7b6.ticket.triage assessing (visit 1) — cdx, openai/gpt-6.1-sol — 6m11s — 1.9M in / 14.3k out
3. github-issues-agent-grounds-rhei-619-implement-7c69c7b6.ticket.triage reproducing (visit 1) — cdx, openai/gpt-6.1-sol — 3m31s — 673.0k in / 8.4k out
4. github-issues-agent-grounds-rhei-619-implement-7c69c7b6.ticket supervising (visit 2) — cdx, openai/gpt-6.1-sol — 2m32s — 634.2k in / 6.2k out
5. github-issues-agent-grounds-rhei-619-implement-7c69c7b6.ticket.specify specify — cdx, openai/gpt-6.1-sol — 9m54s — 3.1M in / 19.6k out
6. github-issues-agent-grounds-rhei-619-implement-7c69c7b6.ticket supervising (visit 3) — cdx, openai/gpt-6.1-sol — 1m17s — 440.2k in / 3.0k out
7. github-issues-agent-grounds-rhei-619-implement-7c69c7b6.ticket.implement implement — cdx, openai/gpt-6.1-sol — 4m32s — 1.4M in / 6.9k out
8. github-issues-agent-grounds-rhei-619-implement-7c69c7b6.ticket supervising (visit 4) — cdx, openai/gpt-6.1-sol — 1m31s — 350.8k in / 3.7k out
9. github-issues-agent-grounds-rhei-619-implement-7c69c7b6.ticket.review-1 review — cdx, openai/gpt-6.1-sol — 5m08s — 1.7M in / 12.1k out
10. github-issues-agent-grounds-rhei-619-implement-7c69c7b6.ticket supervising (visit 5) — cdx, openai/gpt-6.1-sol — 1m26s — 460.3k in / 3.7k out
11. github-issues-agent-grounds-rhei-619-implement-7c69c7b6.ticket.fix-1 fix — cdx, openai/gpt-6.1-sol — 4m24s — 1.5M in / 5.4k out
12. github-issues-agent-grounds-rhei-619-implement-7c69c7b6.ticket supervising (visit 6) — cdx, openai/gpt-6.1-sol — 1m27s — 291.2k in / 3.7k out
13. github-issues-agent-grounds-rhei-619-implement-7c69c7b6.ticket.review-2 review — cdx, openai/gpt-6.1-sol — 3m19s — 964.7k in / 7.8k out
14. github-issues-agent-grounds-rhei-619-implement-7c69c7b6.ticket supervising (visit 7) — cdx, openai/gpt-6.1-sol — 1m20s — 424.0k in / 3.1k out
15. github-issues-agent-grounds-rhei-619-implement-7c69c7b6.ticket supervising (visit 8) — cdx, openai/gpt-6.1-sol — 1m29s — 409.6k in / 3.6k out

Accounting	Value
cost	$4.29
total tokens	14.6M
input tokens (incl. cache)	14.5M
input cache read	13.6M
input cache write	-
output tokens (incl. cache)	104.2k
output cache read	-
output cache write	-
coverage	Complete