Repository navigation
Description
Phase 1 completion amendment (2026-09-04)
The user requires functional failures to be repaired in Phase 1. The committed amendment supersedes earlier baseline-only closure and deferral wording below. Required functional, correctness, integrity, resource, safety-deadline and cleanup gates must pass before PHASE1_TERMINAL_PASS. Optional latency/storage optimization of correct operations remains Phase 2.
Preserve all original baseline failures and producing identities; record corrected candidate evidence separately. Fix shared causes and recollect only genuinely invalidated coverage. Do not shrink fixtures, omit cases, weaken oracles, raise budgets/timeouts or add benchmark-only bypasses. Keep affected family issues and #35 open until corrected execution and independent verification pass; #21 stays open for later phases/release.
Parent: #21
Depends on: #22 (shared infrastructure and committed specification).
Problem statement
LayerFS needs an authentic initial baseline for complete agent work episodes before deciding product improvements.
Goal and family scope
Run 1/10/100/500 complete dependent episodes with alias observations, append/truncate, directory movement, atomic replacement and retained outputs.
Inventory: 4 new timed cases, 12 prescribed initial performance sample slots, and 0 standalone proof recipes. The canonical planning file is docs/roadmap/0.1/0.1.3/mixed-load-bearing-workload.md. The snapshot below makes the current proposal reviewable until its committed revision is published.
Stage 1 work
- Freeze this family and its exact fixture/schedule/oracle identities under the shared contract; implement its complete membership in the existing harness with one canonical definition and thin adapter.
- Qualify route, call counts, size/resource bounds, independent oracles and selected-case self-checks. Resolve harness defects before collecting baseline evidence.
- Run one selected case/seed to establish a valid execution path, then collect every prescribed initial sample and separate proof variant on the pinned unoptimized product. Required extended cases are explicit but remain required.
- Retain raw failures/timeouts as classified product findings. Keep setup, performance, verification, cleanup and external command wall distinct.
- Produce a source-bound family report with all outcomes and measured questions for later optimization; do not tune product algorithms within this issue.
Acceptance criteria
- Complete family implementation, unique IDs, exact counts, deterministic fixtures/schedules and independent oracle checks.
- Committed specification and product/harness/build/runtime/input identities linked in the evidence manifest.
- Every required case/subcase executed with a retained initial outcome; no missing mandatory counter, unimplemented member or silent skipped slot.
- All prescribed valid sample repetitions retained, with separate independent verification of distinct input/schedule variants. Failed/invalid outcomes remain explicit and ineligible for performance claims.
- Preparation reuse, per-sample isolation, byte limits and cleanup checked; actual first-use/cache-hit command wall reported.
- Family baseline summary delivered to the consolidated report issue; known product failures remain failing release gates and are recorded for Stage 2.
Files to read
docs/roadmap/0.1/0.1.3/mixed-load-bearing-workload.md, docs/roadmap/0.1/0.1.3/testing-rules.md, docs/general/benchmark_rules.md, and the existing source/test references in the specification below.
Stage 1 boundary
Build the benchmark and retain its initial outcomes. Product algorithm, storage-format and performance optimization are deferred to Stage 2. Improve benchmark preparation/reuse and correct harness defects now; do not silently change the measured product to make results pass. Pin the unoptimized product identity and distinguish harness changes from product changes.
A slow result, product failure or timeout is an initial finding, not permission to lower a gate, drop a case or start tuning. Keep outcome, harness validity, verification, resources, cleanup and coverage completeness separate. Failed/invalid cases cannot support performance claims. Unimplemented cases, missing mandatory observations and unexecuted slots remain incomplete.
Shared execution rules
- Reuse
fs-bench-pro, existing generators, runner helpers, custody, qualified pristine inputs, independent writable samples and reports. Add only needed family adapters. - Use 1/10/100/500 of the declared unit; no Cartesian size/count matrix. Each workload file is at most 500 MiB and aggregate logical content is strictly below 1 GiB at every intermediate state, including temporary files and Git internals. Physical Store/spool/cache budgets are separately reported.
- Resolve the selected case/seed before preparation. Reuse compatible artifacts; prepare only required inputs. Measured creation/import/history must still execute all work.
- Iterate one selected case and seed; ordinary warm-prepared 1–5 second command wall is a target to qualify, not a blanket guarantee. Required extended cases are explicitly selected and cannot be omitted from complete baseline coverage.
- Separate performance and verification. Added benchmark hashes, full manifests, reopen and fault injection stay outside performance timing; intrinsic CAS/Git hashing remains measured.
- Collect several metrics from one execution and regenerate reports from raw evidence without rerunning successful workloads. Retain all valid results and failures, with exact source, fixture, oracle, runtime and cache identities.
Planning specification snapshot — mixed-load-bearing-workload.md
Source file: docs/roadmap/0.1/0.1.3/mixed-load-bearing-workload.md. Local draft SHA-256: c0f6caba356ada86dfee3df44c8ba0169d1ce0ed3fdf1b99cbd2b1645d727661. This snapshot is reviewable now; it is not a committed admission contract. The infrastructure issue must publish/freeze the source revision before implementation and collection.
Dependent agent work episodes
Status: v0.1.3 implementation plan; no measurements or passing claim.
Family ID:mixed_load_bearing. Four timed cases; no separate proofs.
Shared testing rules (docs/roadmap/0.1/0.1.3/testing-rules.md; planning file awaiting publication) own seeds, preparation, custody,
common lifecycle, timing, limits, and verification admission.
Question and membership
Can a complete agent workflow read, modify, reorganize, and publish a Workspace
while unrelated content remains exact? Every tier executes complete dependent
episodes. This replaces raw-operation prefixes, whose tier 1 was only a read
and whose cancelling mutations could disappear from a final-state-only oracle.
| Scenario ID | Complete episodes | Managed workload executions | Final LayerFS Commits |
|---|---|---|---|
agent-episodes-1 |
1 | 1 | 1 |
agent-episodes-10 |
10 | 1 | 1 |
agent-episodes-100 |
100 | 1 | 1 |
agent-episodes-500 |
500 | 1 | 1 |
These are the only four members: 12 performance samples, three per case,
with separate verification. Repeated public Exec, metadata, symlink errors,
and failure containment belong to
Workspace reliability (docs/roadmap/0.1/0.1.3/workspace-reliability.md; planning file awaiting publication), with no duplicate proof rows
or latency distributions here.
Bounded fixture
Prepare one fixed 64 MiB background using the shared tree generator, including
wide, deep, and regular placements. Freeze 500 independently named episode cells
outside background/. Each initially contains an 8 KiB source, an 8 KiB edit
target and a hard-link alias to it, and an 8 KiB replacement target. Counting
the alias twice, these cells contribute 16,384,000 bytes. N selects prefixes
of one deterministic 500-cell schedule; unused cells remain in the oracle.
Each episode may retain one 8 KiB output and one relative symlink whose target
is at most 256 bytes. It may temporarily hold one 8 KiB replacement file, one
4 KiB scratch file, and a 16-byte append visible through both hard-link names.
Only one episode runs at a time. A conservative whole-workload peak is:
64 MiB background
+ 500 * 32 KiB initial cell path bytes
+ 500 * (8 KiB output + 256 symlink target bytes)
+ 8 KiB replacement temp + 4 KiB scratch + 32 alias append bytes
= 87,729,184 bytes < 128 MiB < 1 GiB
Background files are at most 48 KiB; episode files at most 8 KiB + 16 bytes.
The bound includes temporary coexistence and alias path lengths, including files
removed before Commit. Receipts and expected copies stay outside the workload
with separately reported storage.
One dependent episode
One ordinary filesystem process performs these stages for each selected cell:
- Read the 8 KiB source and derive deterministic replacement bytes and an
output token from its content and the frozen episode identity. - Overwrite a 4 KiB range of the edit target. Read through its hard-link alias
and use the observed bytes to derive the next output; do not substitute
expected bytes without reading the filesystem. - Append 16 bytes, read the appended region through the alias, then truncate
back to 8 KiB. This pair has an intermediate observation despite cancelling
in the final length. - Move the cell directory to a prepared sibling destination, then read the
edited target through its new path. The alias relation must survive. - Write and sync an 8 KiB temporary file, close it, rename it over the
replacement target, and read the permanent name. - Create a relative symlink to the replacement and read through it. Create,
read, and remove a 4 KiB scratch file. - Retain one 8 KiB output derived from the earlier observations. Normalize only
declared changed-path metadata, complete the declared synchronization, and
proceed to the next cell.
Freeze names, offsets, bytes, metadata, operation order, and receipt shape
before admission. No per-episode Commit, remount, additional Exec, SDK edit, or
Git command belongs here. Reads that feed later operations stay in performance;
full hashes, manifests, and additional probes are verification-only.
After N episodes, the single managed execution exits with all handles closed.
One unpromoted Commit must return Created, followed by End and separately
collected fresh reconnect verification. Dedicated families diagnose component
costs; this family measures their dependence and composition.
Independent oracle and observations
Construct the expected manifest and retained output bytes from the input
fixture and specified transformations, not candidate-produced roots or reported
success. In verification mode observe each stage's reads, append visibility,
lengths, aliases, directory bindings, replacement bytes, symlink targets, and
scratch existence before removal. Preserve these intermediate receipts so
skipped append/truncate and create/remove pairs cannot pass via final equality.
After fresh reopen compare every path, including all background and unselected
cells: bytes, link classes/counts, symlink targets, modes, timestamps, head, and
qualified canonical root. Check the transient bound at mutation boundaries
and require complete mount/process/spool/reader/Workspace/lease cleanup.
Metrics, budgets, and grounding
Report episodes, per-class operations and bytes, complete workload and
Create/Commit/End walls, SDK/Exec counts, FUSE operations, candidate and
inserted/reused objects, transaction maxima, RSS, spool/Store growth, sync errors,
and cleanup. Sync success remains passive timing evidence; reliability
explicitly tests its error propagation.
The selected-case 1–5 second goal after cached preparation is provisional.
Freeze qualified per-case and verification walls, including 500 episodes. A
complete family campaign or extended reliability run is not a few seconds.
Keep each episode complete at every tier.
Relevant sources are cow_tree.rs
for links/rename/unlink, file_io.rs
for ordinary writes/truncate, filesystem.rs
for FUSE operations, and execution.rs
for managed execution. Existing SDK alias and mixed-edit checks in
Workspace tests
remain regression coverage; they do not replace this ordinary-tool route.
Activity
yifanxuaaa commentedon Sep 4, 2026
Performance-results checkpoint for #29: 12 independently validated active VM8 performance samples. The report validates 0 retained full proof(s) for this family at this checkpoint; further full checks are running. This comment does not claim global Phase1 terminal pass.
Published report, exact rows, source identities and verification status, phase/counter evidence.
VM8 performance source:
7948df2de269e5ffd47a232ffd8091ff83f8869f; fixed 8-CPU Docker VM, unchanged 2-CPU/2-GiB benchmark-container caps. No VM4/VM8 timing pooling. Every listed performance row passed the independent authenticity, custody, observation, resource and cleanup validation; workload admission still requires its separate full proof.Values below are medians across the prescribed three seeds. Product time is the declared cumulative public-operation time; preparation and complete sample wall are separate. Store growth is the observed physical Store-file delta, not a deduplication ratio or verified canonical byte count.
agent-episodes-1agent-episodes-10agent-episodes-100agent-episodes-500All prior actual failures, source incompatibilities, invalidations and preparation interruptions remain in the linked report and raw evidence. The fast development verifier contributes zero full-verification credit. Completed family tickets will close only after their remaining mandatory full checks pass; #21 remains open for later phases.
yifanxuaaa commentedon Sep 4, 2026
The user has amended Phase 1 acceptance: qualified fast verification is sufficient for routine benchmark verification. The published amendment supersedes earlier comments requiring every routine exhaustive proof before closure. The reference qualification contract preserves independent expected states, affected-content/metadata/alias/absence checks, authenticated references and deterministic witnesses.
The posted performance results remain valid at their original source/environment identities. Completed full proofs remain accepted and are reused. At the scope checkpoint,48 routine slots already had full proofs; the305 unrun routine exhaustive counterparts are now deferred to Phase2 while their qualified fast checks are implemented and collected. Fast evidence keeps its actual assurance and explicit skipped coverage; it will not be labelled
fully_verified.Required targeted reliability checks, the CDC boundary proof, exact expected-error oracles, resource and cleanup gates still apply. Observed failures still require focused repair confirmation. The already-qualified600-second proof is retained rather than repeated. This amendment does not manufacture missing evidence or itself establish terminal pass.
Remaining routine families will close after their accepted fast/full coverage and applicable gates pass. #21 remains open for later phases and release.
yifanxuaaa commentedon Sep 4, 2026
Read-only pin repair performance results for #29 — 12 corrected qualified PASS; 0 corrected non-PASS outcome(s). The corrected passing samples have completed independent validation with zero issues or resource violations. It does not claim terminal pass.
Corrected runtime source:
30d13deeec72b46ff7bc411f1ec08a46990541e1. Other rows retain their original qualified checkpoint at7948df2de269e5ffd47a232ffd8091ff83f8869fwith their original numbers and source. Exactly 45 historical slots across #23/#25/#29/#33 were affected. The correction produced 42 independently qualified PASS slots; directory-content-scan-100 seed 1 exceeded the cumulative 15-second limit and its seeds 2/3 were not run. That case is now suppressed for all Phase 1 seeds/sources: 15 disabled case IDs, 370 active performance slots (328 retained plus 42 corrected). No old/new timings are pooled.The changed product branch now acknowledges the existing read-only regular-file Pin before exposing the Open handle. Historical affected rows remain actual observations of their old source; they are superseded for the repaired candidate, not deleted or called failed. No speedup claim is made.
Product ms, preparation s, complete sample s and physical Store-file delta MiB use the prior table definitions. Corrected medians use only the listed completed qualified PASS seeds; incomplete three-seed groups are explicitly partial. They are medians of independently qualified samples. The full-proof column is copied from the prior checkpoint and is not a refreshed current proof count. Current routine acceptance follows the authorized fast-verification amendment; a fast result remains fully_verified=false. Targeted reliability gates are unchanged.
agent-episodes-130d13deeagent-episodes-1030d13deeagent-episodes-10030d13deeagent-episodes-50030d13deeHistorical affected rows (preserved exactly from the prior qualified table; excluded from corrected medians):
agent-episodes-1agent-episodes-10agent-episodes-100agent-episodes-500Exact corrected outcomes and historical supersession mapping:
agent-episodes-1seed 1:agent-episodes-1-s1-performance-43e514de7dd6(independently qualified PASS), replaces timing selectionagent-episodes-1-s1-performance-70b7424e73ee.agent-episodes-1seed 2:agent-episodes-1-s2-performance-582d00f14fbb(independently qualified PASS), replaces timing selectionagent-episodes-1-s2-performance-a55dcfb319e9.agent-episodes-1seed 3:agent-episodes-1-s3-performance-c81989243d67(independently qualified PASS), replaces timing selectionagent-episodes-1-s3-performance-221b9184c232.agent-episodes-10seed 1:agent-episodes-10-s1-performance-ecf03e8e91fe(independently qualified PASS), replaces timing selectionagent-episodes-10-s1-performance-07ea2847ac8c.agent-episodes-10seed 2:agent-episodes-10-s2-performance-8abaffb14009(independently qualified PASS), replaces timing selectionagent-episodes-10-s2-performance-7754c6039e76.agent-episodes-10seed 3:agent-episodes-10-s3-performance-beda7acdca71(independently qualified PASS), replaces timing selectionagent-episodes-10-s3-performance-db9d5b3ac6ca.agent-episodes-100seed 1:agent-episodes-100-s1-performance-f389f72f3df7(independently qualified PASS), replaces timing selectionagent-episodes-100-s1-performance-9bbf9891ec06.agent-episodes-100seed 2:agent-episodes-100-s2-performance-96e79feb4693(independently qualified PASS), replaces timing selectionagent-episodes-100-s2-performance-64b5e33812b6.agent-episodes-100seed 3:agent-episodes-100-s3-performance-9df0fd916af6(independently qualified PASS), replaces timing selectionagent-episodes-100-s3-performance-54cec18ebd7a.agent-episodes-500seed 1:agent-episodes-500-s1-performance-552ff9c5a6e5(independently qualified PASS), replaces timing selectionagent-episodes-500-s1-performance-61d5ba300f5d.agent-episodes-500seed 2:agent-episodes-500-s2-performance-02f6454e12b0(independently qualified PASS), replaces timing selectionagent-episodes-500-s2-performance-515649f13d40.agent-episodes-500seed 3:agent-episodes-500-s3-performance-7a3e18cf5fd6(independently qualified PASS), replaces timing selectionagent-episodes-500-s3-performance-3e09d5e01c04.All attempt paths are under
benchmark-results/fs-bench-pro/phase1-v013/attempts/. Source, custody, observations, resources and cleanup were independently checked; matching routine correctness status is reported separately. Original failures and suppressions remain preserved; central #21 stays open.Published corrected timing checkpoint and per-case medians.
yifanxuaaa commentedon Sep 4, 2026
Phase 1 #29 is complete under the approved fast-verification amendment: 12 independently qualified performance samples and 12 accepted routine fast-verification results across all four cases and three seeds. There are no suppressed cases in this family, no missing family slots, and no validation issues or resource violations in the selected evidence.
All 12 performance samples were recollected on corrected source
e0922904e2bb607a138157755dab9613b441d5b9after the Rename cache repair. The first 11 fast results retain their actual30d13deeec72b46ff7bc411f1ec08a46990541e1source; the finalagent-episodes-500seed 3 result is one0922904. Exact source compatibility and unchanged workload/oracle definitions were checked, and every performance/verification pair matches its input and VM8 environment identity. No old/new timing samples are pooled.The table reports three-seed performance medians for each exact corrected case. Routine assurance is
fast_iteration_verified;fully_verified=false. There are 0 exhaustive full-verification results credited for this family.agent-episodes-130d13deeagent-episodes-1030d13deeagent-episodes-10030d13deeagent-episodes-50030d13dee; seed 3:e0922904The frozen ordinary Exec/FUSE workload remains intact: source reads feed writes; reads through hard links and symlinks feed later output; append, truncate, rename, temporary-save publication and scratch cleanup retain their original counts. The fast verifier independently checks the final namespace and selected changed, uncertified and witness content, with metadata and alias checks. Certified unchanged bodies and untouched native metadata reads are omitted exactly as the retained receipts state; the full canonical census is deferred. These omissions are not described as exhaustive verification.
Completion evidence: sealed family receipt with all 12 performance/fast identity joins, final seed 3 fast validation, ten retained fast verdicts and omissions, first retained fast verdict, and corrected performance rows/attempt seals.
The source-selection map preserves 337 other campaign performance observations at their original sources. This family completion follows the Phase 1 fast-verification amendment; it does not claim a product speedup or cost equivalence for either repair. All historical affected timings and actual failed proofs remain retained. The targeted hard-link repair has its separate full qualification.
Close #29 for the amended Phase 1 scope. Central issue #21 stays open for remaining family work, Phase 2 optimization and later exhaustive/release qualification.
yifanxuaaa commentedon Sep 4, 2026
Phase 1 is closed under the user's explicit withdrawal of all remaining verification tests. This is a scope withdrawal, NOT a claim that the original verification terminal criteria passed.
The assistant designed an excessively expensive verification suite and spent too long expanding it after the user requested a much faster finish. That design is withdrawn and must be reassessed in Phase 2. No more verification work is scheduled.
Final performance results:370 independently qualified active observations, with actual source identities and unoptimized timings. All15 suppressed case definitions, original failures and incomplete attempts remain preserved. Completed29 targeted proofs,48 routine full proofs and qualified fast evidence remain historical evidence; unrun verification is withdrawn/deferred, never relabeled passing.
Explicit withdrawal and accountability record. Separate Phase 2 backlog. Central#21 stays open for Phase 2/release. No release tag was created.