NFS follow-through: duplicate WASM pool cannot establish disk floor

The distinct generated-WASM pool is insufficient. A bounded, read-only inventory of scratch/runs/**/*.wasm completed across 52,470 directories: 210 paths, comprising 190 regular files and 20 symlinks. Symlinks were excluded without following them. The initial 50,000-directory bound stopped; the completed inventory used explicit limits of 250,000 directories and 45 seconds. No other file pool was inventoried or cleaned.

Full SHA-256 and size grouping of the regular files found 53 distinct inodes and 17 distinct byte sequences. A second complete hash pass verified those groups and stable inode, size, mode, nanosecond mtime and link count. Most apparent duplicate modules already share hardlinked allocations. Summing each pathname's blocks would misleadingly count the same allocation many times. No hardlinks or symlinks were created or substituted.

Even assuming every regular path were closed and had no runtime dependencies, the maximum reclaimable allocation is only 4,521,984 bytes, while preserving an existing original of every byte sequence. This upper bound excludes inodes with links outside the inventory from reclaimable allocations, and retains one inode per hash/size group when all its links are inside the inventory. Closure and dependency exclusions could only reduce the result. Two full-sized duplicate allocations account for 4,382,720 bytes; small generated fixtures account for the remainder.

Available disk at the pool measurement was 1,492,353,024 bytes. Adding the entire optimistic upper bound gives 1,496,875,008 bytes, still 3,124,992 bytes below the strict 1,500,000,000-byte floor. This is newly bounded evidence from the authorized pool, not a repeated native-launch decision. No closure, FD/maps scan or deletion eligibility is claimed. No exact removal paths were claimed, no files were unlinked, and no recovery index or restore implementation is needed. All original modules, unique bytes, existing retained dependencies and recovery recipes remain untouched.

Evidence: scratch/runs/20261113-nfs-duplicate-wasm-1700/, including metadata, full-hash groups, validation, Git SHA and relevant binary diff. The twenty O_NOFOLLOW refusals in the first hash report correspond exactly to the twenty metadata-classified symlinks; validation confirms all 190 regular paths were successfully hashed and reverified. No source/module archive, media copy, raw-log encoding or other cleanup occurred. Restore earlier raw logs before historical consumption using their existing recovery recipes.

Both NFS1550 and NFS1640 notes were read. The bcd413390 actor resolver, unchanged chooser export and local media were not modified. No heavy lock, runtime child, watchdog, input or box was created because this pool cannot meet the prerequisite. Actual UI-actor inspection/bind, two reviewed Pause/Continue frames/worldtimer, and a reviewed >=1500 ms held driving frame with independent UP/Escape remain unproved. No guest/config/clock/AI writes, autopilot, UNKNOWN replay or forced presentation occurred.

Next requires a separately authorized material resource change that can establish sustained headroom. No unchanged-floor retry is justified by this pool. No production change or build was required; validation covered the inventory, complete hashes, allocation arithmetic and own-note diff. Box ID NONE; state NOT_CREATED; expiry null; stop receipt N/A.