jakub
6a667d8deb
Include the current addon, terrain integrations, tools, regression tests and recorded development evidence.
15 lines
3.1 KiB
Markdown
15 lines
3.1 KiB
Markdown
# Mine-safety candidate helpers — pending evidence
|
|
|
|
These scratch scripts target `.build/playtest-source-2026-10-03-mine-safety` and a new `.build/playtest-2026-10-03-mine-safety`. They never modify the previous mounted-pair candidate. Root alone executes approved target-writing and native commands; no background launcher is scheduled here.
|
|
|
|
1. Freeze the current authored source, tools, tests and suite-bound documentation. Expected scope is 422 authored files and 299 registered regression programs. Create the fresh source ZIP using commands.json; never overwrite an existing unrelated archive.
|
|
2. Obtain `/tmp/erri-playtest-review-0644/preflight.json`. `preflight.py` recomputes source, suite registration/results and normal evidence. `suite_binding.py` requires exact authored/test/tool inputs and the independent review's exact documentation-only mismatch list. No automatic reuse from matching counts. Suite is `/tmp/erri-gameplay-root-0558`; normal evidence is `native-2026-10-03-engineer-movement`.
|
|
3. Root inspects host-visible processes with approved escalation before each native operation. A restricted sandbox check can miss an existing game. Preserve any existing session and stop if overlap is present. Use the canonical direct `/tmp/erri-package-mine-safety-0644` and `/tmp/erri-packed-chat-mine-safety-0644` paths; the packed `chat` prefix is required. Native remains hidden/muted and bounded, with separately installed authorized dependency closure.
|
|
4. Run native package, preserving its actual exit/log/report. `assemble.py` requires cleanup/input seals, native exit 0, exact PAK payload/source/ZIP membership and unchanged-source normal evidence before creating the new candidate. It retains a strict packaging refusal even when the independently checked package bytes are usable for this diagnostic handoff.
|
|
5. Run ordinary packed startup using the new runtime bytes, without smoke/trace/input/flight fixture flags. `verify.py` recomputes raw evidence and exact loaded bytes, preserves the strict failure and checks the historical installed-resource error set. Any new script failure, changed error set, cleanup ambiguity or input difference stops finalization.
|
|
6. Independent reviewer writes `final-review.json`. `finalize_docs.py --verified --review PATH --review-sha256 HASH` previews four documents in this scratch directory. Root reviews, then alone repeats with `--apply`. It refuses concurrent documentation edits or a changed candidate README; no historical package is rewritten.
|
|
|
|
`finalize_docs.py` without verification produces only a pending scratch preview. Counts are read from actual manifests/reports after verification. Source-only or failed packaging does not advance the latest candidate pointer: retain evidence and leave current docs unchanged pending a separate reviewed source-only note.
|
|
|
|
The candidate remains unqualified. Native mine placement/arming/detonation, complete human escape, manual UI, real multiplayer/JIP/reconnect/twelve-player operation, physical logistics/interception and comparative FPS remain unverified. The intermittent zero-radius diagnostic is retained as unresolved; opt-in instrumentation is not a gameplay fix or caller proof.
|