[Japabu]

The tracker carried records whose defect was fixed, whose code no longer exists, or whose test was cut. This deletes sixteen of them. Records only; no code changes.

Each record was checked against the tree at 422915e2c: whether the function, file or test it cites exists, and, where the code was replaced, whether the defect is still reachable in what replaced it. The commit message carries the same evidence in prose.

record	action	evidence
a-crash-leaks-the-blocks-of-its-last-commit	delete, gone with the shadow-paging crate (#1152)	No bitmap written in place: a bucket's state is an alloc-btree key journaled with its change. lib/bcachefs/src/engine/tests/crash.rs judge mounts every sampled crash state and requires fsck to find nothing; fsck's derived-btree walk reports alloc keys the pointers do not derive
a-btree-split-leaves-nodes-holding-a-few-entries	delete, gone with the crate	split_node/pack are gone; publication and reclaim split through mkfs::spread (even fill), reclaim at two thirds (engine/mount/publish.rs, reclaim.rs)
a-write-over-a-files-committed-page-is-not-shadowed	delete, fixed by design	Data is copy-on-write (engine/mount/data.rs); data_and_logged_truncate_survive_torn_reordered_writes overwrites committed bytes and judges each crash state as one commit or the other
a-file-on-data-is-refused-at-250-runs-while-blocks-are-free	delete, gone with the crate	Extents are keys of the extents btree; MAX_ENTRY_SIZE, file_entry_fits and the cited test do not exist
deviceerror-classify-trusts-a-lying-transfererror	delete, gone with the crate	DeviceError, TransferError, classify: no hit outside issues/
a-console-holders-line-state-guards-writers-that-no-longer-exist	delete, code gone	ConsoleLine, MID_LINE, MAX_CONSOLE_LINE removed by a788d314b; kernel/CLAUDE.md holds no line-buffer caveat
the-i8042-tallys-release-acquire-pair-has-no-negative-control	delete, code gone	tally.rs and i8042_tally.rs removed by febfd21e8
copy-meets-a-remap-holds-a-cpu-the-thread-it-waits-on-may-be-queued-behind	delete, code gone	The spin, remap_race::hold, was deleted by 2f366a930. A user copy now runs through the process's own mapping (kernel/src/user_ptr.rs header), so a racing unmap meets the fault path user_copy_faults and unmap_touch cover
fs-turns-judges-the-scheduler-with-the-lock	delete, test gone	fs_turns cut by ff1a11dbd
hda-client-stall-prints-a-stream-error-and-passes	delete, test gone	hda_client_stall cut by ff1a11dbd
a-loaded-suite-reds-a-volume-checker-on-both-arms	delete, held by a host test	The leak was real and reproduced (toybox_cp_volume red run alone on both arms, through a deliberately failed write). 0d73b9957 gave toyos-fat32 its repair contract, and lib/fat32/tests/refused_writes.rs refuses every write of every structural call and requires of the independent toyos-fat32-check that no cluster is orphaned
a-double-panic-at-boots-edge-says-nothing-but-its-name	delete, sighting of a deleted test	log_poll_outlives_a_close deleted by ad6dc0781; finding 2 resolved in kernel/src/panic.rs; finding 1 not seen in 612 boots and the log paths it named were replaced (88a17feee, a788d314b). It also cited a record that no longer exists
logkeeper-still-says-a-budget-refusal-leaves-the-device-untouched	delete, fixed	system/logkeeper/src/policy.rs holds neither sentence
a-log-rings-owner-is-named-only-when-logkeeper-reads-its-registration	delete, fixed	Rings are laid out with a placeholder owner and the supervisor names the owner before registering; a_placeholder_owned_ring_keeps_its_slots_from_every_pid_until_named (sdk/toyos/src/log/proof.rs, d565ff418)
an-fsync-of-a-served-file-syncs-its-whole-volume	delete, built and measured; row 28 of toyos-builds-itself no longer cites it	persist.write_4k_fsync_beside_1m 3.03 ms against persist.write_4k_fsync 2.36 ms on the AMD bench, Linux 4.18 ms: evidence
kernel-issues-cite-the-cut-https-tests-as-cover-that-runs	delete, fixed	a-claims-own-refusals-are-read-by-nothing names neither test; the remaining mentions are the tracks' plans for the metal row

Records: 752 on main at 34cb4c918, 736 on this branch.

Decisions a reviewer may want to question

• The fsync record's exit asked for base against head on metal. The base is the storage crate Consolidate storage around native bcachefs and userland filesystems #1152 removed, so that pair cannot be run. The code establishes the behaviour: Files::fsync (system/fileserver/src/pages.rs) writes back only the named file and those the volume reports carried, DataVolume keeps Volume::carried's default of false (volume.rs), and DataVolume::commit (data.rs) persists only the named node before the journal commit. The AMD reading confirms it: the megabyte written beside the fsynced file would add about 3.8 ms at persist.seq_write_fsync's 266 MB/s if the fsync carried it, and the row costs 0.67 ms more than the fsync alone.
• The double-panic sighting is deleted without its first finding's cause being found. It was one red under host load of a test that no longer exists, and the log and console paths it suspected were replaced. The leaked-cluster record is not of this kind: its leak was reproduced, and the deletion rests on the host test above.

Checks

No code or instruction file changes, so no host check applies. For each deleted slug, git grep <slug> on the branch finds nothing.

🤖 Generated with Claude Code

[Japabu]

Review of #1220 at 7fdc4309b (base 34cb4c918)

Assessment. Records only: 16 records deleted (−738 lines) and row 28 of toyos-builds-itself.md updated. I checked every deletion at origin/main. All 16 records have been overtaken by the code, but one record's closing evidence misstates what the record says. No deleted slug is cited anywhere on the branch, in the four hygiene branches merged together, or in the changed files of any open or in-progress branch. The branch merges cleanly with #1221, #1222, #1224 and current main. The four together leave 732 records.

Findings

BLOCKER: a-loaded-suite-reds-a-volume-checker-on-both-arms. The PR row and the "Decisions" bullet say "reds were under host load and green alone". Commit 7fdc4309b says "green alone each time". The record says the opposite. Its "fifth name" section (origin/main:issues/a-loaded-suite-reds-a-volume-checker-on-both-arms.md:126-133) records toybox_cp_volume answering ALONE … red again on both the branch arm and origin/main. It calls the leaked cluster "real and reproducible", reached through a deliberately failed write. The deletion is still right, but on other evidence. 0d73b9957 (2026-09-26, after every sighting) gave toyos-fat32 its repair contract (lib/fat32/src/repair.rs). lib/fat32/tests/refused_writes.rs refuses every write of every structural call, including append_two_clusters and first_allocation. It then judges the result with the independent toyos-fat32-check, which requires "no cluster orphaned". The commit message is where a close's evidence lives (issues/README.md). As written it hides a reproduced leak behind a host-noise label. Fix: correct the commit message and the PR row to say the leak was reproduced, and cite 0d73b9957 and refused_writes.rs as what now holds it.

Fsync record (the risk the author flagged): meets its exit. Code inspection already shows the behaviour; the metal reading confirms it:

• Files::fsync (system/fileserver/src/pages.rs:619) writes back only the named file and files the volume reports as carried.
• DataVolume does not override Volume::carried, whose default is false (volume.rs:225).
• DataVolume::commit (data.rs:693-698) persists only the named node before the journal commit.

On the AMD bench the evidence is the median of four boots' p50 for the M arm (c9d6eff6f, after #1152): 3.03 ms for persist.write_4k_fsync_beside_1m against 2.36 ms for write_4k_fsync. If the fsync carried the other file's megabyte, it would add at least about 3.8 ms at 266 MB/s, more than five times the 0.67 ms seen. The base-against-head pair cannot be run because the base is a crate that no longer exists. Nothing more is owed.

• SUGGESTION: add the three code sites above to the commit message, because they establish the claim more strongly than the timing does.

Checked and accepted:

• The five storage records. split_node, pack, MAX_ENTRY_SIZE, file_entry_fits, BitmapAllocator, DeviceError, TransferError and the classify method are gone. In the engine:
  • Bucket state is a journaled alloc key. fsck's derived() (engine/fsck.rs:701-729) reports any alloc key whose usage the pointers do not derive, and crash.rs's judge runs it on every sampled crash state.
  • Data is copy-on-write (mount/data.rs header), and the torn-write tests exist.
  • Splits go through mkfs::spread.
  • toyos_blockdev::Error has no trait door left that could mint a verdict with no transfer behind it.
• Console holder, i8042 tally, remap_race, fs_turns, hda_client_stall. The code or test is gone. fs_turns' claim is held by sdk/toyos/src/sync.rs's the_value_goes_to_the_tickets_in_the_order_they_were_taken, which is the alternative that record's exit offered.
• copy-meets-a-remap….
  • SUGGESTION: "remap_race deleted" was already true when the record was written. The real reason nothing is owed is that a user copy now runs through the process's own mapping (kernel/src/user_ptr.rs header), so a racing unmap meets the same fault path that user_copy_faults and unmap_touch cover. Say so in the commit.
• Double panic. Finding 2 is resolved in kernel/src/panic.rs. Finding 1 was a host-noise sighting, which root CLAUDE.md lets go unrecorded, and its suspected log and console paths were replaced. A recurrence now reports the first crash itself.
• logkeeper sentences, ring owner, HTTPS citations.
  • policy.rs no longer holds the sentences.
  • The supervisor lays every ring out with a placeholder owner (system/supervisor/src/main.rs:327,336), and proof.rs:401 tests that.
  • The remaining https_tls13 mentions are rows the internet-clients track plans to build, not claimed cover.

SEND BACK

[Japabu]

Orchestrator verification at 74cea1d, after review r1 SEND BACK (text-only blockers).

This PR changes only files under issues/ (checked with git diff --name-only against main). I read the fixes for each blocker the review named on this PR and they are resolved as the review asked; the review found no deletion that hides a live defect. Records only, so no build or test is owed beyond the tracker gates in the merge queue. Queueing.