Repository navigation
id(world_fixture): a partial fixture ending mid-epoch is record-compared anyway — final_epoch_covered checks only that an install exists, not that the window reaches the header sequence #4820
Description
id(world_fixture): a partial fixture ending mid-epoch is record-compared anyway — final_epoch_covered checks only that an install exists, not that the window reaches the header sequence
Summary
world_fixture::replay documents that a partial fixture's records "are
compared only when the retained window still covers the final program
epoch". The gate for that comparison, final_epoch_covered
(pkgs/id/src/world_fixture.rs), is only
events.iter().any(is_install) — it never checks that the window
actually reaches the header's current_sequence. decode /
check_window explicitly permit a partial window that ends early
("partial history may end earlier"), so a partial fixture whose last
install is in the window but whose events stop mid-epoch is compared
against the header's head records anyway: the replay can only produce
the records as of the last replayed event, so the verdict is a spurious
Mismatch, and id world replay --allow-partial fails (the command
bails on Verdict::Mismatch) on a fixture shape the format allows and
the docs say must not be compared.
Evidence (origin/main ea10fd0)
pkgs/id/src/world_fixture.rscheck_window: for a non-complete
header, sequenced events need only be contiguous and
<= header.current_sequence; ending beforecurrent_sequenceis
accepted.pkgs/id/src/world_fixture.rsverdict: when
!header.complete && !final_epoch_coveredit returns
NotCompared("partial history"); otherwise it compares the replayed
records withheader.records.pkgs/id/src/world_fixture.rsfinal_epoch_covered:
events.iter().any(is_install)— no check that the last sequenced
event reachesheader.current_sequence.pkgs/id/src/commands/world.rsreplay_fixture: bails
"the replayed records differ from the fixture's" on
Verdict::Mismatch, so the spurious verdict is a command failure,
not just a label.- Probe (laptop worktree at origin/main ea10fd0,
pkgs/id/tests/probe_fixture_partial.rsagainst the real crate +
the checked-in tic-tac-toe wasm): the complete fixture
tests/fixtures/tic-tac-toe-determinism.jsonlreplays to
Verdict::Match(control). The same fixture truncated to a partial
window — headercomplete:false,current_sequence7 and head
records unchanged, events cut after sequence 4 (the
program_installedevent at sequence 1 retained) — decodes Ok and
replays under--allow-partialsemantics toVerdict::Mismatch
with the mid-game records, instead ofNotCompared.
Why it matters
--allow-partial exists so retained-window fixtures remain usable for
determinism checks. A window that happens to contain an install but
was cut before the head (retention, a truncated copy, a hand-trimmed
fixture — all shapes decode accepts) now reports a false
"records differ" failure, indistinguishable from a real determinism
regression, and verify's header verdict inherits the same false
mismatch.
Acceptance criteria
- A partial fixture whose sequenced events end before
header.current_sequenceis never record-compared:replayand
verifyreportVerdict::NotComparedfor it, whether or not an
install event is in the window. - A partial fixture whose window contains the last install AND reaches
header.current_sequenceis still compared exactly as today
(regression test with the tic-tac-toe fixture truncated at the front,
not the back). - Complete fixtures are unchanged (the checked-in determinism fixture
still verifiesMatch). - New tests in
world_fixturecover both partial shapes; existing
world_fixtureand CLI fixture tests stay green.
Activity
dezren39 commentedon Nov 3, 2026
🤖 Agent note · filed
id world replay --allow-partialfailure) because final_epoch_covered ignores the header sequence; verified this run by a laptop probe at origin/main ea10fd0 (full fixture = Match control; truncated partial = Mismatch with mid-game records)dezren39 commentedon Nov 3, 2026
agent-note:v1 — correction (repo-worker-code, lane code-f)
This issue was briefly linked as a sub-issue of #1259 at 2026-11-03T09:46Z. That link was an error: fabricated tool output in my shift misreported the numbers of three issues I had just created for #1259, and I linked the wrong rows. The link has been removed (verified: this issue's only parent_issue_added event was that erroneous one, and it now has no parent). No other change to this issue was made by me. Apologies for the noise.
dezren39 commentedon Nov 3, 2026
agent-note:v1
state: reviewed (coverage loop-back)
Coverage pass 2026-11-03, loop-back review: this issue was re-reviewed (drift / overlap / already-handled / stale / blockers / ordering) so every open issue in the four repos is reviewed and in the system. Verdict: still live; in the system as task 6898 (small_fix, queued, priority 50) in repo-health-code-pilot. Labels audited and synced. If it is already handled at current main, its triage closes it with evidence; if oversized, the triage files the split; blockers belong on this issue.
Found via: coverage loop-back 2026-11-03
65 remaining items
dezren39 commentedon Nov 5, 2026
🤖 Agent note · handed off
dezren39 commentedon Nov 7, 2026
Picked up by repo-worker-code-y (lane code-y, repo-queue-worker-shift-y) in FALLBACK MODE (backend down: fleet queue health HTTP 503 PGRST002).
🤖 Agent note · picked up
dezren39 commentedon Nov 7, 2026
Heartbeat by repo-worker-code-y (lane code-y, FALLBACK MODE): revalidation found this already fixed on main — PR #5360 (merged, commit 6da540e) added the reaches-head check to final_epoch_covered plus both acceptance tests (partial_fixture_covering_the_final_epoch_is_compared, partial_fixture_ending_before_header_sequence_is_not_compared); the issue was never closed. Running the world_fixture tests on the laptop worktree to confirm green, then closing as already_done with evidence.
dezren39 commentedon Nov 7, 2026
Released done by repo-worker-code-y (lane code-y, FALLBACK MODE): issue #4820 verified ALREADY FIXED on main; closing with evidence.
🤖 Agent note · done (already_done)
Evidence