Historical report imported 2026-10-10 from docs/experiments/formal/readiness-continuation-results.md. Original SHA-256: 9c578baab21637409702e9f9c6c107f3efd2f3b975e6621d5bd1f02e5d84601d. Measurements and outcomes retain their stated revisions. Links and local environment labels are normalized; $FABRIC_DATA is the mounted experiment drive. See evidence policy.

Readiness continuation results

Execution of the registered queue (local checkout reference docs/experiments/formal/readiness-continuation-protocol.md; not present at origin/main).

Status: queue completed. The bounded writer is now the default. All 17 final unflagged fast checks and three manual documentation checks passed; both native trial freezes were archived, verified and removed. The results below retain their actual workload, revision and failure history.

Item	State	Evidence
Continuation instruction	Implemented	AGENTS.md now requires continuing actionable authorized queue items through verification. Policy commit 37d55db.
RC-FORMAL	Finite checks completed	Delivery, trace and transport TLC checks exited 0. Eleven registered Kani proofs passed and M-KANI-SEQ/DUP/RET-SATURATE were caught. The first fuzz attempt failed for missing C++; its corrected retry exited 0 across four 60-second targets and 11,370,454 executions. Corpus/logs were archived before scratch cleanup.
RC-MERGE	Passed finite checks	All 21 bounded-sealer tests passed, including 64 seeds × 5 byte limits and 13 fan-in populations, stable ties and altered-key/payload/order controls.
RC-RETENTION	Passed finite check	Positive B/B−1 byte boundaries, interrupted deletion/restart, long-lived readers and exact retained suffix passed against the unchanged Python query oracle.
RC-FAULT	Passed finite checks	All 17 native syscall faults recovered exact clean manifests, with ordinary-error cleanup checked before recovery. A further 13 named SIGKILL cuts plus a no-hit control passed exact custody and query-oracle chains before/after restart.
RC-GROUP	Builder gates passed; defaults adopted	The aligned-input/disk-PageStore campaign passed all eight cells and three pairs each, including exactness, pruning, ceiling, scaling, determinism and cleanup. Worst heap was 39,114,918 bytes at steady256 and 41,405,704 at bigrows64. Row-run spill replays and balanced page observations completed. Unflagged tests, loaded-binary equivalence, final fast and documentation checks passed. Earlier aligned-only results and failed screens remain historical.
Spool ENOSPC	Passed finite check	Actual scoped ENOSPC append plus no-hit control preserved exact prior unACKed bytes, sequence and ACK cursor; quarantine and reopen refusal remained effective after injection removal. Physical filesystem exhaustion and double faults remain untested.
Sealer mutants	Three named defects caught	M-SEAL-MERGE-ORDER, M-SEAL-SPILL-LEFT and M-SEAL-RETAIN-RUNS each had a successful baseline and an expected failing mutant; all three mutation commands exited 0.
RC-SOAK	All ten R2 gates passed	The native record preserves startup/smoke failures and the full frozen trial: 546,000 simulator batches and 825 companion sources ACKed/recovered exactly, ten Segments, clean exits and matching post-freeze checks. The frozen evidence is archived; historical run 01 remains failed.
RC-INSTALL	Finite installation acceptance completed	Fresh Debian 12 package build and cleanup exited 0. Its Debian 13 unified-cgroup VM passed all seventeen checks, including actual A12 OOM and service health; all three required mutation controls rejected their named defects. Original GLIBC rejection, cleanup failure and tail-stressor VM failure remain preserved.
Encoded-page probe	Counterexample and finite correction retained	The disk-store diagnostic reduced high-entropy heap from 143,075,502 to 42,011,899 bytes with exact files, filters and pruning. All 27 bounded tests and seven native page faults passed. A fresh screen and the full eight-cell/three-pair combined campaign passed. The measured builds used both selectors; the same mechanisms are now defaults with unflagged tests and loaded-binary equivalence verified. Earlier eight-cell results retain their original scope.
Companion shutdown	Regression checks passed	Three actual-supervisor controls accepted orderly child exit and rejected nonzero exit or forced deadline kill; all children were reaped. Both named shutdown mutants were caught by their specific regressions after passing baselines.
Dependency policy	Check completed	cargo xtask checks --only dependency-policy exited 0 using the private official cargo-deny 0.20.2 binary; no host installation.

Historical failed acceptance and soak records are retained unchanged.

Executed commands

All commands below used python3 -B tools/resource_group.py --; unit names identify the launcher JSON receipts in target/resource-containment/runs. Measured peaks are whole-job cgroup charge, not builder heap or server RSS.

Command after launcher	Exit	Receipt unit suffix	Peak charge
cargo xtask checks --only tla-delivery --receipts $FABRIC_DATA/results/readiness-continuation-tlc-01	0	365645144b0644c381cfe87cf46a8339	119.8 MiB
cargo xtask checks --only trace-model --receipts $FABRIC_DATA/results/readiness-continuation-tlc-01	0	1e78f6c3334e43d8b795260bc8e20a99	255.6 MiB
cargo test --offline --locked -p fabric-server --lib segment::bounded -- --test-threads=1	0	060e73cf4e0c4fcda4344ebcc64327ce	1.1 GiB
cargo test --offline --locked -p fabric-server --test completion_storage byte_ceiling_and_interrupted_retention_preserve_exact_suffix -- --exact --nocapture	0	4de390694d5c4af6957876523433706d	1.0 GiB
python3 -B tools/bench/labs/completion/test_faults.py	0	0125b76e0ecc485ea5a364175cba3ce4	18.8 MiB
python3 -B tools/bench/labs/completion/faults.py --id readiness-continuation-syscall-01	0	30f6616a64574dd0b94355712b6091d5	335.4 MiB
cargo test --offline --locked -p fabric-server --test completion_storage named_sealer_kill_cuts_recover_exact_custody_and_oracle_query_chains -- --exact --nocapture	0	c9f1c242135548bea5efd92d4733c6f2	1.0 GiB
python3 -B -m unittest discover -s tools/qualification -p 'test_soak*.py'	0	d093adb31d8c49628b4c98808e7668b7	22.1 MiB
python3 -B tools/bench/labs/completion/chunk_groups_screen.py --out $FABRIC_DATA/results/readiness-continuation-group-screen-03	0	4f72939790fc4c64a1a8633a99360014	1.6 GiB
cargo test --offline --locked -p fabric-server --lib segment::bounded::chunk_groups_tests -- --test-threads=1	0	8506b63b76eb485f9393c263aef314ac	981.9 MiB
python3 -B tools/bench/labs/completion/continuation_campaign.py --screen $FABRIC_DATA/results/readiness-continuation-group-screen-03 --out $FABRIC_DATA/results/readiness-continuation-group-01	0	a68baf85cd96446bb77842a28c5adb67	1.88 GiB
cargo test --offline --locked -p fabric_o11y --test spool_enospc native_enospc_preserves_unacked_bytes_and_requires_recovery -- --exact --nocapture	0	b5f6e235451f4efaa35c63cba8ca53c7	497.6 MiB
cargo xtask mutants --only M-SEAL-MERGE-ORDER	0	352e527836b94b1083b16b6c5fda8b5a	7.68 GiB
cargo xtask mutants --only M-SEAL-SPILL-LEFT	0	ce47b0bf3f9b4878a3bd5eeecabbd9cc	7.19 GiB
cargo xtask mutants --only M-SEAL-RETAIN-RUNS	0	0acadd731f43422f9df2d77f56d19d81	6.47 GiB
python3 -B tools/bench/labs/completion/spill_measure.py --campaign-root $FABRIC_DATA/results/readiness-continuation-group-01 --binary $FABRIC_DATA/results/readiness-continuation-group-01/frozen/completion_builder --out $FABRIC_DATA/results/readiness-continuation-group-01/spill-01	0	a91d936a7dee4e9eb73a151db16ccea9	822.2 MiB
cargo xtask checks --only kani-core --receipts $FABRIC_DATA/results/readiness-continuation-formal-01	0	ab4bd9bb6b384d6bace0fcab538af014	941,514,752 B
cargo xtask mutants --only M-KANI-SEQ	0	21bb48b268f34099bec35d33a7573008	7,044,755,456 B
cargo xtask mutants --only M-KANI-DUP	0	4b99960338e24ad3a57f217c473bb539	6,194,630,656 B
cargo xtask mutants --only M-KANI-RET-SATURATE	0	4dfd1db5a7b941c0b09048f9cce84c19	6,408,376,320 B
cargo xtask checks --only tla-transport --receipts $FABRIC_DATA/results/readiness-continuation-formal-01	0	47799151774442efb7c38a415d6f6f5e	263,237,632 B
cargo xtask checks --only fuzz-smoke --receipts $FABRIC_DATA/results/readiness-continuation-formal-01	1	9e840dfca5794dc0b94d06c03eebd0ef	2,423,889,920 B
python3 -B - (private C++ compile/runtime check)	0	fe67837836f747e4b566cc9a376886e0	59,056,128 B
python3 -B tools/qualification/install/build-isolated.py --run-id readiness-01	2	07aaa965f6de49acbcb09998c5aac732	3,235,823,616 B
python3 -B -m unittest discover -s tools/qualification -p 'test_soak*.py'	0	8ac105b29cc24786beb5a366da8c396f	24,633,344 B
python3 -B tools/qualification/test_verify_soak_freeze.py --out $FABRIC_DATA/results/readiness-continuation-freeze-controls-01	0	4a73c176af684cf88fd659ac00394358	30,871,552 B
cargo xtask checks --only fuzz-smoke --receipts $FABRIC_DATA/results/readiness-continuation-formal-02	0	ccdce3fad04f4d0cba570520886ae89f	3,901,775,872 B
python3 -B - (fuzz evidence archive and cleanup)	0	e1b67ddd2f2044edbb55807a474eb9a8	25,501,696 B
python3 -B tools/qualification/install/build-isolated.py --run-id readiness-debian12-02	2	cca20bd898b84d8da4cd28797a60215e	2,702,073,856 B
python3 -B - (package scratch cleanup follow-up)	0	c5f921a226a24f35ae813f6fee2e2806	16,519,168 B
python3 -B tools/qualification/install/run-qemu.py --deb $FABRIC_DATA/results/installation-package-build/readiness-debian12-02/fabrico11y_0.1.0~alpha.1_amd64.deb --source-commit bc858a7580a4e89994d744d4b3780b2979a3520f --run-id readiness-debian13-01	1	402ef0c958644c1faa0a9f7b5c84e02e	5,089,480,704 B
env PATH=$FABRIC_DATA/toolchain-cache/cargo-deny-0.20.2:$HOME/.cargo/bin:/usr/bin:/bin CARGO_HOME=$FABRIC_DATA/toolchain-cache/cargo RUSTUP_HOME=$FABRIC_DATA/toolchain-cache/rustup RUSTUP_TOOLCHAIN=fabric-local cargo xtask checks --only dependency-policy --receipts $FABRIC_DATA/results/readiness-continuation-dependency-01	0	b7eac9d6453e4461a6d547b5d72e3821	328.9 MiB
python3 tools/bench/labs/completion/entropy_probe.py --binary $FABRIC_DATA/cargo/release/examples/completion_builder --out $FABRIC_DATA/results/readiness-continuation-encoded-page-01 --build-command 'env -u FABRIC_RUN_MIB_EXPERIMENT -u FABRIC_EARLY_ROW_RELEASE_EXPERIMENT -u FABRIC_SPILL_WORKSPACE_EXPERIMENT FABRIC_ROW_GROUP_CHUNKS_EXPERIMENT=1 cargo build --offline --release -p fabric-server --example completion_builder --features responsibility-alloc-probe'	0	a239e0059636442480ad6831068b2368	987.7 MiB
python3 tools/bench/labs/completion/entropy_probe.py --binary $FABRIC_DATA/cargo/release/examples/completion_builder --out $FABRIC_DATA/results/readiness-continuation-encoded-page-02 --build-command 'env -u FABRIC_RUN_MIB_EXPERIMENT -u FABRIC_EARLY_ROW_RELEASE_EXPERIMENT -u FABRIC_SPILL_WORKSPACE_EXPERIMENT FABRIC_ROW_GROUP_CHUNKS_EXPERIMENT=1 FABRIC_PAGE_STORE_EXPERIMENT=1 cargo build --offline --release -p fabric-server --example completion_builder --example completion_fault --features responsibility-alloc-probe'	0	b346b7c15ad84eddab1f3975f86461d9	985.3 MiB
env -u FABRIC_RUN_MIB_EXPERIMENT -u FABRIC_EARLY_ROW_RELEASE_EXPERIMENT -u FABRIC_SPILL_WORKSPACE_EXPERIMENT FABRIC_ROW_GROUP_CHUNKS_EXPERIMENT=1 FABRIC_PAGE_STORE_EXPERIMENT=1 cargo test --offline -p fabric-server --features responsibility-alloc-probe segment::bounded --lib	0	8eb1d179f9a345df810f3682efa42019	1348.8 MiB
python3 tools/bench/labs/completion/page_store_faults.py --builder $FABRIC_DATA/results/readiness-continuation-encoded-page-02/completion-builder-frozen --helper $FABRIC_DATA/cargo/release/examples/completion_fault --out $FABRIC_DATA/results/readiness-continuation-page-store-faults-01	0	7ebc39e8a33141b69e425e64b39ef8b7	215.1 MiB
cargo test --offline --locked -p fabric-server --bin fabric-server shutdown_tests -- --test-threads=1 --nocapture	0	8ed9f39c35a74536920d48afcfee0792	859.5 MiB
python3 tools/bench/labs/completion/chunk_groups_screen.py --out $FABRIC_DATA/results/readiness-continuation-page-store-screen-01	0	07cb099b12b44a18b8113c688ff36380	1687.0 MiB
python3 tools/bench/labs/completion/continuation_campaign.py --out $FABRIC_DATA/results/readiness-continuation-page-store-group-01 --screen $FABRIC_DATA/results/readiness-continuation-page-store-screen-01	0	e615ef879f284c07a4355328f474e8d9	1922.1 MiB
python3 tools/bench/labs/completion/spill_measure.py --campaign-root $FABRIC_DATA/results/readiness-continuation-page-store-group-01 --binary $FABRIC_DATA/results/readiness-continuation-page-store-group-01/frozen/completion_builder --out $FABRIC_DATA/results/readiness-continuation-page-store-row-spill-01	0	082bdce8a93241589d899031f73f2761	802.0 MiB
python3 -B tools/bench/labs/completion/faults.py --id readiness-continuation-page-store-syscall-01	0	943d515bf89a442ebc5ba2f0f10f1498	333.2 MiB
env FABRIC_ROW_GROUP_CHUNKS_EXPERIMENT=1 FABRIC_PAGE_STORE_EXPERIMENT=1 cargo test --offline --locked -p fabric-server --test completion_storage named_sealer_kill_cuts_recover_exact_custody_and_oracle_query_chains -- --exact --nocapture	0	5eb1138889a14f66ae4c34ac7b293955	1909.8 MiB
cargo xtask mutants --only M-SELF-SHUTDOWN-STATUS	0	32860cdddee14fca92bdbd7120aaaec9	6575.7 MiB
cargo xtask mutants --only M-SELF-SHUTDOWN-DEADLINE	0	c3918ad6b8b84e1897b1d3c91724e366	6727.8 MiB

The Kani commands and failed fuzz attempt used an env prefix with PATH=$FABRIC_DATA/toolchain-cache/cargo/bin:$HOME/.cargo/bin:/usr/bin:/bin, CARGO_HOME=$FABRIC_DATA/toolchain-cache/cargo, RUSTUP_HOME=$FABRIC_DATA/toolchain-cache/rustup and RUSTUP_TOOLCHAIN=fabric-local. Kani also set KANI_HOME=$FABRIC_DATA/toolchain-cache/kani; fuzz set CARGO_TARGET_DIR=$FABRIC_DATA/cargo-fuzz. The named launcher receipts preserve the complete argument vectors. The successful fuzz retry additionally set the private CXX compiler under $FABRIC_DATA/toolchain-cache/cxx-16.2.1-2.fc44, with its recorded CXXFLAGS and LIBRARY_PATH; the original failed invocation remains a separate receipt.

Combined writer campaign

The encoded-page record retains the combined FABRIC_ROW_GROUP_CHUNKS_EXPERIMENT=1 / FABRIC_PAGE_STORE_EXPERIMENT=1 campaign, all 24 pairs, source/binary hashes and scratch observations under results/readiness-continuation-page-store-group-01. All implemented builder gates passed; full_bs_acceptance remains false. Steady64-to-steady256 heap grew 5.88%, within the unchanged 10% threshold. Seven cells took 1.7–11.9% longer; bigrows64 took 2.7% less time. These are measured comparisons on a shared desktop host, not isolated causal timing claims.

All eight supplemental row-run replays matched inputs/manifests and cleaned scratch. Page observations across the 24 pairs had balanced puts/takes, zero outstanding blobs and successful cleanup; malformed, unbalanced and cleanup-false controls were rejected. At steady256, 336,302,946 row-run bytes plus 193,741,173 page blob bytes totaled 530,044,119 logical scratch bytes. Page cache is charged to the cgroup; counted heap savings do not establish server RSS bounds. The original soak remains failed; the companion-compatible R2 trial passed all ten gates.

Normal-writer adoption

The default run record retains the unflagged build and verification evidence under $FABRIC_DATA/results/readiness-continuation-default-01. Aligned sorted groups and disk-backed completed pages are now enabled without experiment flags. With all five selectors unset, the 27 bounded tests passed, exit 0, unit 7d4f4c00fadc4075b4657ea8680dfa9d; the 13 named kill cuts plus no-hit control passed exact custody and unchanged query-oracle chains, exit 0, unit 9fe4e4065c8149fdaec846229aaca429.

The release build exited 0 in unit abcffb68bfb241b9af20cff00c36b1ca. Four artifacts match the accepted frozen full-file hashes. The server's full ELF hash differs, but loaded-elf-equivalence.json verifies all program headers, loaded segments and allocated section bytes after validating and normalizing only the section-table offset and GNU build-id descriptor. The checker rejected a .text byte-flip control, exit 0, unit 3e1ff3b5898a47419968a2548cf4731f. This establishes loaded-content equivalence, not whole-file identity.

A repository-target smoke placement violated the data-only scratch requirement; its interrupted evidence was preserved before cleanup. The replacement frozen native smoke used the data drive. The subsequent runner guard rejects output outside FABRIC_SCRATCH_ROOT before creation or launch, while preserving marker, symlink and oracle checks. All 66 runner/workload/delivery-oracle tests passed, exit 0, unit 056b12832b1f43bfb1f9410cb418980c. Both owned freezes were archived and removed after verification: full-trial cleanup unit 39397bc72cc54e8090642f456ff3311b and default-freeze cleanup unit 5adaf862e45648cfb4dea2daa5e837da each exited 0.

Final unflagged cargo xtask checks --profile fast --receipts $FABRIC_DATA/results/readiness-continuation-fast-03/checks passed all 17 checks, exit 0, unit 37b7e81db5c54e72ba7c998140e4bebe. All five experiment selectors were explicitly removed. Elapsed time was 329.399 seconds; CPU 728.407 seconds; whole-job cgroup peak 15,354,585,088 bytes; zero memory events and no swap. Its group stopped and scratch was removed. Manual documentation verification through python3 -B tools/bench/labs/cross_system/sweep_documentation.py --out $FABRIC_DATA/results/readiness-continuation-docs-01 passed all three checks, exit 0, unit 502bf34625e04cde9a575eeaf23a3ddb. Its private authenticated Bun copy was removed; CI/hooks remain disabled. The default run retains artifact identities, the interrupted attempt, completed smoke and cleanup receipts.

After record reconciliation, the same three documentation checks passed again with output results/readiness-continuation-docs-02, unit 485c0270232740cb9213d5853616ddd3, exit 0. Final consolidation through python3 -B $FABRIC_DATA/results/readiness-continuation-default-01/closeout.py exited 0, unit 5bc9e64eab684b4399ec0eb8bd7702f5 (2.031 seconds, 29.1 MiB peak). It copied 93 referenced launcher receipts, confirmed all 196 compilation inputs still match the unflagged build, verified trial cleanup and absent tagged processes, and counted 66,039,800,484 apparent project bytes before receipt copying, below the 100 GB ceiling. Its compact report and receipts are under $FABRIC_DATA/results/readiness-continuation-closeout-01.

Historical aligned-only writer campaign

Frozen source/binary identities, all eight cell commands and three pairs per cell are retained under the mounted drive's results/readiness-continuation-group-01; summary.json gives the reductions and result.json explicitly records full_bs_acceptance=false. The selector FABRIC_ROW_GROUP_CHUNKS_EXPERIMENT=1 aligns encoder input batches while preserving reference physical groups. Production defaults were unchanged during that historical campaign.

Representative median wall times in seconds; heap is incremental builder heap, not whole-job cgroup charge:

Cell	Reference wall s	Opt-in wall s	Change	Opt-in heap MiB
steady16	0.349	0.365	4.5% slower	37.2
steady32	0.695	0.722	3.9% slower	37.6
steady128	2.744	2.696	1.8% faster	39.7
steady256	5.368	5.223	2.7% faster	39.7
bigrows64	0.898	0.885	1.6% faster	63.4

Timing is reported, not gated; small cells regressed while these larger cells improved modestly. All measured builder gates passed, including the original 80 MiB ceiling and 10% steady64-to-steady256 scaling threshold. This finite campaign does not establish arbitrary-input, live-service or whole-server bounds. The separate eight-cell spill observer replay exited 0, preserving each frozen input and manifest, with a known-write control and full owned-scratch cleanup. Raw observations and source/library hashes are in results/readiness-continuation-group-01/spill-01. This is one supplemental logical-write observation per cell, not three-pair spill/I/O or timing/heap measurement. BS-7 had not run at the time of that historical campaign.

Mutation baseline and defect logs are retained on the data drive under scratch/semantic-mutants/logs. Each named command's zero exit means the verifier caught its injected defect, not that the mutant passed. The retain-runs mutant failed the registered steady128 heap ceiling test as intended. Their launcher receipts confirm stopped groups, temporary-directory removal and no OOM events.

Formal tooling and installation counterexamples

The high-entropy probe retained its frozen sources and exactness results under $FABRIC_DATA/results/readiness-continuation-encoded-page-01. pair.json records heap_ceiling=false and all_checks=false; the probe command exited 0 after recording this counterexample. The candidate used 136.45 MiB incremental heap against the unchanged 80 MiB ceiling. The separately frozen combined candidate, built with FABRIC_ROW_GROUP_CHUNKS_EXPERIMENT=1 and FABRIC_PAGE_STORE_EXPERIMENT=1, used 42,011,899 bytes and passed all comparator checks. Moving 101,019,985 completed log blob bytes to scratch closely accounted for the 101,063,603-byte heap reduction. Twenty-seven bounded tests and seven native page-fault cases passed. The fresh screen at results/readiness-continuation-page-store-screen-01 records screen_passed=true and acceptance=false: heap was 36,943,714 bytes at steady64, 41,406,229 at bigrows64 and 39,115,218 at steady256, with 5.88% scaling and owned cleanup complete. The combined repeated campaign and supplemental scratch observations completed; the subsequent native R2 trial passed all ten gates. The original failure and aligned-only eight-cell result retain their scopes.

With both selectors enabled, the seventeen original scoped syscall cases also passed, including cleanup before ordinary-error recovery. Thirteen named kill cuts plus no-hit control passed unchanged query-oracle chains before and after restart. Exact binary hashes, command logs and shutdown mutation baselines/defect logs are archived under $FABRIC_DATA/results/readiness-continuation-verification-01. The status and deadline mutants were each caught by their named regression.

The fresh combined-selector fast profile retained a failure: readiness-continuation-fast-01 exited 1 in unit 689b255731214655969ef89007b4f2a9. Its workspace tests and fifteen other gates passed; Clippy rejected a nested cleanup condition and a cloned test slice. Direct Clippy reproduced both diagnostics (unit aa22816237fd4108956f4634960a3763, exit 101). The correction only collapses the cleanup condition into an equivalent let-chain and borrows the test slice with std::slice::from_ref; it changes neither cleanup ordering nor expected bytes. Formatting and direct Clippy then exited 0 (units 65d47c143b924ebab5d68d00bbdac413 and 0dc7b3b4e02240b49df3c4ebf349be2b). The measured release binaries retain their recorded hashes and revision scope. The fresh full fast profile under readiness-continuation-fast-02 passed all seventeen gates, exit 0, unit 4b038818d4df46d095603bc45a20fa3a: 517.994 seconds elapsed, 1,059.754 CPU seconds and 12,763,164,672 bytes cgroup peak, with zero swap. The command was env FABRIC_ROW_GROUP_CHUNKS_EXPERIMENT=1 FABRIC_PAGE_STORE_EXPERIMENT=1 cargo xtask checks --profile fast --receipts $FABRIC_DATA/results/readiness-continuation-fast-02, passed after the resource launcher. The disposable mutant source tree was archived and removed, exit 0, unit bef8f5f51c1a4f36ac7c12b287a58a69; shared build caches and historical logs remain. Cleanup and exact command receipts are in readiness-continuation-verification-01.

The R2 review found that production shutdown could ignore a nonzero companion exit or kill it at the deadline while still returning server success. The supervisor now returns failure for both outcomes and for premature successful HTTP termination with a live child, while preserving earlier errors and reaping the child. Three regressions ran the actual TLS supervisor with controlled child exit 0, exit 7 and ignored SIGTERM. They passed with a private 200 ms grace and confirmed ESRCH after reaping; the production grace remains 12 seconds.

Dependency-policy checking exited 0 using official cargo-deny 0.20.2, whose prebuilt archive SHA-256 is 9f12ed4c49936e09b48bf862b595cde2fe64fcbd9d74dfacac6131ca824c8d5f. The private installation record is $FABRIC_DATA/toolchain-cache/cargo-deny-0.20.2/install.json; check receipts are under results/readiness-continuation-dependency-01.

Kani's eleven checks are bounded proofs of the registered core harnesses; TLC explores the registered finite models. Neither establishes arbitrary Rust execution. Their receipts live under results/readiness-continuation-formal-01. The failed fuzz attempt and missing C++ evidence remain in that directory's fuzz-missing-cxx; the subsequent private-toolchain compile/runtime check remains a prerequisite result. The subsequent four-target fuzz retry completed 60 seconds per target without a crash, with 11,370,454 total executions; this time-boxed search does not prove absence of defects. Its raw logs/corpus archive is $FABRIC_DATA/results/readiness-continuation-formal-02/fuzz-evidence.tar.gz, SHA-256 ac348a4342e72c38d9536055efee6d51990727b103bc1ce7654ebe0b5afb647c. fuzz-archive.json records 2,931 members, 451,836 bytes and removed scratch.

The rejected package build is preserved under results/installation-package-build/readiness-01, including compatibility-counterexample.json: Debian 13-built fabric-server contained weak undefined pidfd_spawnp and pidfd_getpid references marked GLIBC 2.39. The unchanged gate refused them above GLIBC 2.34. The Debian 12 sysroot build itself exited 0 and produced $FABRIC_DATA/results/installation-package-build/readiness-debian12-02/fabrico11y_0.1.0~alpha.1_amd64.deb, SHA-256 ea4239365e0d96d38a4aae91d16307693481beb4485b9b7828ba59faaa439ea5. Its outer command exited 2 because Podman teardown recreated empty host-owned store locks after the user-namespace cleanup. A separately recorded follow-up exited 0 and removed remaining owned scratch; cleanup-followup.json preserves both earlier exits and the package identity. The original nonzero outer result has not been changed to success.

The actual Debian 13.7 QEMU trial with systemd 257 and cgroup v2 follows the installation plan (local checkout reference docs/experiments/formal/installation-acceptance-qemu-plan.md; not present at origin/main). Its $FABRIC_DATA/results/installation-qemu/readiness-debian13-01/acceptance.txt reports fails=1 not_run=0: both serial tail memory hogs timed out below their hard caps. The failed run remains unchanged. The fresh installation record follows source snapshot 717fcca151487dbe1f510b6e99296d8caf18d749: its Debian 12 build and cleanup exited 0 (unit f5d197e1facb4dc78d1384740c595f3e), package SHA-256 ac71995b3c8b88cbb6f7dd05d9854e347b7a783caadc1b602f6590d6460f4a2f. The fresh Debian 13 VM exited 0 (unit bcc7714a7f464e7bae24f0919df7c86e), all seventeen checks passed and fails=0 not_run=0. The registered parallel A12 fixture produced actual kernel OOM events and systemd oom-kill under the original hard caps; both services stayed healthy and delivery continued. Cleanup removed the VM and owned scratch. Required root-user, no-collision-check and no-memory-max VM controls then rejected A6, A2a/A2b and A8 respectively. Every mutated guest exited 1 with no unrun check; each driver exited 0 for the confirmed rejection. Secondary A7/A10 and A3 failures remain in their reports. The installation record links archived mutated packages, exact receipts and successful owned-scratch cleanup.

The native syscall evidence is retained in readiness-continuation-syscall-01 (local evidence omitted from stage). The fixture cleanup receipt confirms removal. Generated test fixtures remove their owned state after success. Screens 01/02 results live under the mounted drive's results/readiness-continuation-group-screen-01 and -02; the invalid first screen's readback and byte hashes were preserved before removing 167.8 MB of scratch. Screen 02's unequal Parquet files and all file hashes were preserved before removing its 168.8 MB scratch; screen 03 removed its successful fixtures.

At the last storage census the project used 45,963,306,089 bytes, below 100 GB. The toolchain install first completed nightly extraction but returned 1 when rustup attempted self-update in the isolated Cargo home; rerunning with --no-self-update confirmed the installed toolchain and exited 0. No check result was inferred from the initial nonzero install.