# Current-contract Java DAC report gates

The Philadelphia-fast, QuickFIX/J and pinned Artio 0.182 DAC controllers now
apply the current scenario contract before promoting a fresh medium row.
Functional parity, process exits, physical TX counters, sample counts and
stability remain separate prerequisites.

The additional report proof is:

- An inherited quiet CPU lease on the selected benchmark host.
- Achieved throughput at least 98% of target in every run.
- A complete raw `scheduler_delay` series for each measured owner run,
  with zero drops and p99 no greater than the send interval.
- Explicit measured-window `admin_in`, `admin_out` and `resend_requests`
  counters for each role and run; any measured ResendRequest invalidates
  the report row.
- For tick-to-trade, a server declaration that its clock starts before
  the prepared wire send, alongside the existing server-owned raw series.

`report_blockers` is retained in CSV and JSON. Tiny and lower-rate runs
remain diagnostic; a medium command exits nonzero if its report gates
remain unmet. Local `summarize()` checks still report their functional and
latency result independently.

## Observed gap

The [paced Philadelphia-fast localhost bundle](evidence/philadelphia-paced-local-20261002/SHA256SUMS)
passes its existing market-data and mixed-flow summary gates at the 10k/s
rate. Its raw files include complete scheduler-delay series. Its
`counters.csv` contains only the header, however, and supplies no admin
processing observations. Regrading its market-data artifacts under the
additional policy records `admin_observations_missing`; it cannot be
promoted by supplying physical counters alone.

The controllers consume the existing hft-bench `counters.csv` layout,
`benchmark,run,counter,value`. Each named admin counter needs exactly one
nonnegative integer value for every measured run, including an explicitly
observed zero. Tick-to-trade consumes the server's raw file and a server
`manifest.json` metadata key `tick_to_trade_clock=before_prepared_wire_send`.
Absence of either proof stays diagnostic. Driver instrumentation and a new
pinned Java payload remain follow-up work; the current installed payloads
must not claim these observations by inference.

Twenty focused checks pass, including the retained real-data regrade,
missing/duplicate/invalid counters, measured resend, 97.9% throughput,
scheduler saturation, missing tick boundary, missing quiet lease and zero
primary server latency. The retained older campaign bundles and transcribed
coverage cells are unchanged; this gate controls fresh report promotion.