Skip to content

fabric storage - #13650

Closed
briansrls wants to merge 88 commits into
mainfrom
integration/warm-badger-442
Closed

briansrls wants to merge 88 commits into
mainfrom
integration/warm-badger-442

Conversation

@briansrls

Copy link
Copy Markdown
Contributor

Auto-opened by session-dashboard for session warm-badger-442.
Pushing to integration/warm-badger-442 advances this PR.

Worker attestation

Before flipping this PR to ready for review, confirm each item:

  • Title describes the change (not the session id or branch).
  • PR body summarises what and why (replace the TODO below).
  • Tests run: name the command (e.g. npm test, cargo test) and the result.
  • If this closes a work item, the body contains a Closes #N directive.
  • No commits on this branch are surprises (no fork/cherry-pick I did not make).
  • No secrets / credentials / large binaries staged.

Summary

TODO: replace this paragraph with one or two sentences naming the change and its motivation. Reviewers read this first.

Test plan

  • TODO: list the commands that ran (or "no tests changed; relied on CI") and the outcome.

gunbc-ci-auto-heal and others added 30 commits October 4, 2026 23:52
…ier in both realizations

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…he slice

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…share the carrier; regen fixed point holds

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…threshold

Dropping a slice of a shared RRB list consumed it into the iterative-drop worklist,
cloning every member out of nodes it shared with its source: O(len) per drop, so an
offset walk was quadratic in drop alone (measured: 12000 m:count calls = 13.2s at
49k members; the skip arms themselves 207ms). Detach only past 64 nested drops.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…n follows)

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…d; fixed_point_equal=true)

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
… enrol the slice cost controls on the generated job

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…e controls

The portable round-trip deep test overflowed the fleet runner's test stack on the
first CI run that executed it; it passes on main and this branch alike off the fleet
(3/3 each), so it is a separate finding about the portable form.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…ness eval-step budget

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…gen follows)

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…_size

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…red-regen converged)

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…gen follows)

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…red-regen converged)

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…roots

The substring rules *token* and *credentials* dropped 48 and 3 tracked source files of the clone,
so stage 0 would have measured a smaller workspace than the operator's. The row now names credential
files exactly (Claude .credentials.json, Codex auth.json, netrc/git/npm/pypi files, SSH key names,
key/cert suffixes, cloud CLI and docker configs, dotenv). Roots are the srv1 host paths: the clone,
/home/briansrls/.claude@gmail3 and /home/briansrls/.codex/sessions. Witness 28/28.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…id_ed25519_sk)

Review 76565: the exact-name rows missed the _sk key names outside .ssh.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…-class excluded files and bytes in the receipt

Operator decision 2026-10-05 (default B): the pack carries agent configuration and state, never
conversation content. A second exclusion class, TranscriptExclusion: projects/**/*.jsonl (Claude
session and subagent transcripts), history.jsonl (Claude and Codex), .codex/sessions (Codex rollouts).
The Codex root is ~/.codex (auth.json excluded as a credential). shell.Find FilesAndSymlinksWithMode
gains a size column (%s, appended so every existing positional parser is unchanged); the tally reports
excluded files and bytes per class, and names every credential-excluded path (review 76579).
Witness 29/29.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…hed partition, no hand-rolled predicates

Review 76603: WorkspaceSourceExcluded carries bytes: ByteSize from the parse; a size that does not
parse (or is negative) makes the line WorkspaceSourceUnparsed instead of a silent 0; the double parse,
workspace_is_int, workspace_some, workspace_exclusion_class_eq and workspace_line_excluded_bytes are
deleted -- the tally partitions excluded sizes by matching the exclusion class. Witness 30/30 (new
red an_unreadable_size_refuses_the_line_rather_than_counting_zero).

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…-open SuffixUnder arm

Review 76617: WorkspaceSegmentRule (Equals | Prefix | Suffix | Pair) is what the per-segment matcher
takes; WorkspaceExclusionRule = AnySegment { rule } | SuffixUnder { .. }. The SuffixUnder => false
arm and the four pass-through arms are gone, so a path-level rule cannot reach the segment matcher.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…s; withhold the 10 s / 60 s verdicts

- extdeps.crypto.seeded_filler: coreutils' cited seeded random source (truncate -s N, then
  openssl enc -aes-256-ctr -pass pass:<seed> -nosalt), exactly N bytes, size read back.
- workspace_source: TranscriptExclusion files (never credentials, never symlinks) staged as
  placeholders seeded by member path, packed into one tar with the real members (two -C/-T
  lists); readback refuses credentials, unplanned transcripts, and placeholders absent or of
  another size (verbose listing, map-keyed). Receipt carries files/bytes and the caveat.
- workspace_checkpoint_instrument: caveat in the receipt; four bound lines print
  'withheld: churn not measured'; quota prints as quota_static.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A real transcript of a placeholder's size passed the readback as clean. Each staged placeholder now
carries the SHA-256 of its filler bytes; after packing, the placeholder members are extracted back
out of the archive and hashed, and the readback refuses per cause: absent, not a regular file
(the archive's own type letter), wrong size, unread, digest differs. The archive build and
readback are one function (workspace_pack_archive), exercised wet on the real two-root tar.
Refusal text says 'workspace exclusion' (it covers transcripts too).

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
… over the pack's transcript rows (fleet-converge workspace_transcript_rate_observe)

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…with the interval fold

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…ks; located refusals before extract

- agent_state_slot (dispatch input, env WORKSPACE_AGENT_STATE_SLOT): admitted only when it names an
  observed /home/briansrls/.claude@<slot>; absent or unknown refuses naming the observed slots; no
  default; receipt carries slot and path. Clone and ~/.codex stay fixed rows.
- workspace_pack_archive makes the placeholder root first, so a workspace with no transcripts packs.
- Names/types/sizes are judged from the listings before extraction, so a missing member is refused by
  name rather than as a failed extract.
- Wet witnesses branch on mktemp before any effect.
- fleet-converge.yml regenerated (main_wet_one).

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…the Claude root from that admission and the same dispatch input

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…threshold; the slice step now runs only the runtime crate's controls (rust-unit-tests runs v1-compiler whole)

Generated artifacts conflicted and were taken from main; regen follows.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
gunbc-ci-auto-heal and others added 27 commits October 6, 2026 23:24
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…and main

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…s the runtime_rust.dag list_skip/list_take block; regen follows)

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…red-regen converged; v1_rt.rs equals its emission)

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…unk_42, wet schedule rows re-appended)

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…oth rows sets

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…obe and WorkspaceTranscriptRateObserve; fleet-converge.yml regenerated next)

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…sum rendered at the receipt edge (review 77655)

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…riptRateObserve; yml regenerated next)

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
#13429 a651914

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…regen follows)

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…deps_version_semver.rs re-emitted)

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
… regenerated from the merged authorities)

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…ated below

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…1b6477

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…_tests_claim_interpolation_hole_census.rs)

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…riptRateObserve; floor_route_gap: main's chunk_31 comment kept beside #13429's chunk_42; yml regenerated next)

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
@gunbai-bot

gunbai-bot Bot commented Oct 9, 2026

Copy link
Copy Markdown
Contributor

Closing as a duplicate. The dashboard re-opened this integration branch's PR when its manager was archived. Its content is in the v1 closeout mega branch #13641 (for qwen-argv, #13626 is in and #13632 is deliberately left out). Branch kept. — sent from neat-wolf-604

@gunbai-bot gunbai-bot Bot closed this Oct 9, 2026
@gunbai-bot gunbai-bot Bot mentioned this pull request Oct 10, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant