Local exact-gzip/XZ bound: insufficient material capacity
The bounded local opaque-byte codec experiment passed exact reconstruction, but cannot supply the material headroom required for original NFS. No original file was encoded on disk, unlinked, replaced or restored. Actual reclaimed space is zero. The private file API's earlier 401 was never retried or worked around; no network transport, worker or box was used.
Evidence: scratch/runs/20261113-local-exact-xz-1800/. The read-only candidate
superset is all 33 original canonical session .gz paths in the 1730 inventory
that have successful result/notification, nonempty final, reviewed completion
ledger entries and existing review evidence. Current/pending registry matches
are excluded. The 1,157,441,767-byte inventory is not an eligibility claim.
Nonledger/coordinator sessions, existing XZ, game captures, source/media,
saves, ISOs, databases, authentication and remote-only backup are outside
this bound. No unchanged raw-log or module cleanup was retried.
The 33 gzip files contain 42,019,436 encoded bytes, occupying 42,102,784 allocated bytes. All original bytes were SHA-256 hashed, decoded as opaque bytes, and hashed again without payload parsing. Every admitted stream has a fixed 10-byte original gzip header, verified CRC/length trailer, one completed raw-deflate member and no unused/unconsumed input. The original header and trailer are retained verbatim in the proof. Current zlib 1.3 reproduced every original encoded stream byte for byte using a single whole-stream compress followed by Z_FINISH, raw window -15, memLevel 8, default strategy: 29 use level 6 and four use level 9. No flush schedule was guessed. The probe refuses optional headers and unmatched parameters/forms rather than declaring them reconstructable. Per-file evidence records settings and both full hashes.
Each individual raw-byte XZ passed full decode identity and reconstruction of
the original gzip from that decoded raw data. A combined preset-9-extreme XZ
over all 78,178,176 opaque raw bytes was 21,714,788 bytes, with whole decode
identity. Its SHA-256 was
902060e3e11beb043bde80358ac86a823db3f0a8a0b6e1d2ccc89776e0d9dc9a.
XZ bytes existed only in process RAM and were discarded at successful process
exit; there is no new durable recovery dependency.
Even optimistic 4 KiB rounding gives only 20,385,792 bytes savings before any index, protected reconstructor, RECOVERY or evidence costs. This is a feasibility calculation, not measured allocated recovery. The measured free space after evidence was 1,463,386,112 bytes; adding that optimistic saving gives 1,483,771,904, still below 1,500,000,000 and below the desired 30 MB material saving. Full transitive dependency/foreign-claim eligibility was deliberately not asserted: exclusions could only reduce this superset. No partial small cleanup followed. This bounded experiment does not prove that every possible local log pool is exhausted.
Privileged before/final current-inode FD/maps and producer-cmdline censuses passed without hits or errors. Stable lstat signatures and complete original gzip hashes were rechecked after the experiment. Every original gzip, old helper, index and receipt remains intact. There was no path mutation claim, filesystem encoding or unlink stage, so no new protected restore layer, actual RAM-file restore/metadata/refusal test or old-helper-chain test is claimed. Those remain mandatory before any separately authorized removal. An initial summary validation incorrectly required all compressors to be level 6; the per-file evidence was correct, and the corrected 29/4 summary passed.
The 1550/1640 notes and actor fix bcd413390 were read. Original live UI-actor
binding, native Pause/Continue and finite >=1500 ms held driving with an
independent UP remain unproved. No fresh runtime admission, floor retry,
watchdog, heavy lock, input replay, guest/config/clock/AI write or forced
presentation occurred. Next requires a separately authorized material resource
change yielding sustained headroom, then the unchanged chooser export/media,
current freeze, heavy lock/actual-child checks and independently prearmed
observable guard before inspecting the live actor and issuing reviewed input.
Only this note is committed; no production code changed, so no production build is required. Validation covers the real codec experiment, its successful parent wait, privileged census, complete unchanged-original rehash, metadata stability, current/pending exclusions, script syntax and evidence hashes. Git SHA and owned new-file patches preserve source provenance without archives. Box ID NONE; state NOT_CREATED; expiry null; stop receipt N/A.