# Explicit Philadelphia-fast eager JVM experiment on both hosts

The canonical builder now accepts an explicit `--jvm-profile=eager`. Its
declared options are `-Xms128m -Xmx128m -Xbatch -XX:-TieredCompilation
-XX:CompileThreshold=100 -XX:CICompilerCount=2`; the build receipt retains them.
The default launcher remains byte-identical to the installed generation:
SHA-256 `fccb59120796c253609da3b36593dd6cd336a3058e0c98e34b778be1a6c00e07`.
The existing runtime generations and queued physical collection use that default.

Both independent stripped builds pass 269 capture/numeric checks, twelve actual
functional pairs and 49 cold refusals, with zero skips. Full quiet local matrices
then retain twelve pairs per host, 120,000 primary/delay samples and 330,000
original application frames per peer. All body, owner and administration checks
pass. Every flow on both hosts still fails achieved rate; the original delay
and full stability decisions pass for this experiment. Large actual extrema
remain visible. No grade threshold or warmup/sample/rate profile is changed.

Both relocated SDK-free readers reproduce all original receipts, raw quantiles,
true maxima and original gate decisions exactly. Archived runtime-product
inventories match the full matrices byte-for-byte and mode-for-mode. SDK/runtime
binaries remain separately archived inputs; receipts retain their exact hashes
and the actual launchers. This is localhost two-process S4 diagnostic evidence,
with no physical run or Java comparison promotion.

The complete 856-file source snapshot has SHA-256
`5a9f561e4b23ef25e3c0b8b797f6d6cc3c378a714cf748b1dca5c8dd714974d8`.
Its explicit identity includes the working-tree suffix and full inventory;
it does not claim a clean Git revision. The builder change is separately
published at `225425c0a`. The snapshot contains no engine, internal or Git tree
and remains unchanged after both host qualifications and replays.

The seal retains 735 inputs, 356,773,340 bytes; SHA-256
`134f00b774947fa93497ca71bed4252c165107f765d8847a5a6e035cb9f069c6`.
Across both hosts, 48 actual functional/local pairs retain 240,480 primary/delay
samples and 661,400 application frames per peer. All failed rate gates remain.

[Full originals, independent replays, source archive, runtime receipts and flags](evidence/philadelphia-eager-canonical-hosts-20261014/retention-seal.json)