# Philadelphia-fast independent canonical host qualification

Both HP and SCAN independently build the complete stripped 838-file harness at
`bfb7d93e7`, against the unchanged public Philadelphia 2.0.0 core JAR. Each
passes 269 envelope/numeric checks, twelve actual canonical TCP session pairs,
49 copied-evidence refusals and SDK-free functional replay, with zero skips.
All producer source byte/mode inventories remain unchanged. The functional
build/suite runs under `isol`, with high priority on HP and normal on SCAN;
concurrent shared work is allowed and no timing grade is inferred.

## Full quiet local matrices

Both exact builds then complete twelve full pairs under `isol-no-shared` leases,
at 1,000 warmup cycles, 10,000 measured cycles per run, three runs and 10,000/s.
All original body, owner, administration and sample-count checks pass. Each
retains 120,000 primary and delay samples and 330,000 application frames per peer.

| Host | NOS/ER | Market data | Mixed | Tick-to-trade |
|---|---|---|---|---|
| HP | rate, delay and stability fail | rate and delay fail | rate and delay fail | rate and delay fail |
| SCAN | rate, delay and stability fail | rate, delay and stability fail | rate, delay and stability fail | rate and delay fail |

These results remain **local diagnostics**. No workload, rate or threshold is
relaxed, and no repeated run is used to obtain a favorable grade. All failures
are retained. Two independent relocated SDK-free full readers reproduce every
original receipt, raw quantile, true extremum and failed gate exactly, without
changing original files/modes or opening new sockets.

Across functional and full matrices: **48 actual pairs, 240,480 primary samples,
240,480 delay samples and 661,400 original application frames per peer**.

## Remaining gates

The separate immutable physical payload/controller and root namespace entry,
physical counter/placement admission, exact physical cold replay and matched
Java comparison remain required. The legacy installed Philadelphia generation
and historical cells retain their original identities and grades. Full-plan
and whole-harness release admission are not inferred from these diagnostics.

[Source, both independent builds, functional/full originals and exact cold replay](evidence/philadelphia-canonical-hosts-20261014/retention-seal.json)
are retained in a new immutable seal: 1,592 files, 367,606,880 bytes, SHA-256
`1f9eb66e71122519600fb413ebf91d83918efe08f18ce2a3eb86718084d5fd7e`.