Include the current addon, terrain integrations, tools, regression tests and recorded development evidence.
23 lines
1.4 KiB
Markdown
23 lines
1.4 KiB
Markdown
# Reforger 1.8 workaround probe: ai_query_worker
|
|
|
|
Runs batches of 256 native world-collision traces through the worker RPC. Phase 1 compares server/worker results; phase 2 performs the batches only on the worker and applies a returned candidate as an authoritative waypoint; phase 3 deliberately silences the worker to exercise timeout fallback.
|
|
|
|
This is an isolated experiment, **not a production Escape feature**. It deliberately
|
|
reuses the ownership probe addon GUID and script layout for its resource index.
|
|
Load exactly one probe variant at a time, with no Escape addon.
|
|
|
|
Follow [the original probe's hidden-run prerequisites](../ai_delegation/README.md).
|
|
Index a disposable copy of this variant with matching Workbench; then run this
|
|
folder's `run_local.py` with `--runtime`, `--indexed-addon`, `--compat`, and `--run`.
|
|
The indexed script must match this fixture exactly. Keep tests hidden and leave the
|
|
normal game closed; the runner refuses an existing game. It uses an isolated
|
|
loopback dedicated server and a private client profile. No published mission changes.
|
|
|
|
The logs contain `AID` markers. `AID PROBE COMPLETE` means the observation phases
|
|
finished, not that native AI offloading succeeded. Preserve native counters and
|
|
actual returned flags when interpreting a result. All runtime scripts must identify
|
|
as 1.8; these fixtures were tested on 1.8.0.13.
|
|
|
|
See [the workaround report](../../../docs/ai-delegation-workarounds-1.8.md) for measured
|
|
results and limitations.
|