Promote the native integration branch to master: M1-M4, #73, #76 (65892de1)

.github/CODEOWNERS

@@ -0,0 +1,8 @@
+# Native programme technical lead; policy is repository-wide so source, data,
+# fixtures, workflow, instruction and ownership changes cannot fall through a gap.
+# This file proposes ownership, not active branch protection. See:
+# docs/governance/REVIEW.md and issue #52.
+# GitHub uses CODEOWNERS from the PR base branch. This proposed file does not
+# protect its own bootstrap PR until present there. Do not add alternative owners
+# or overriding patterns that let another approval satisfy the lead requirement.
+* @fd-starscream-bot

.github/PULL_REQUEST_TEMPLATE.md

@@ -0,0 +1,54 @@
+## Task and scope
+
+Task: <!-- one accountable issue -->
+Epic / milestone:
+Base branch and SHA:
+Candidate SHA: <!-- update when requesting/re-requesting review -->
+
+Outcome and scope/non-goals:
+
+## Baseline and behavioural evidence
+
+Baseline revision / scenario / inputs:
+What was characterised before moving or changing it:
+
+| Evidence | Command / artifact | Actual result and scope |
+|---|---|---|
+| Baseline sociable checks | | |
+| Red reproduction, or reason not applicable | | |
+| Candidate sociable / regression checks | | |
+| Boundary / contract proof | | |
+| Integrated Lua-module path, when affected | | |
+| Tacview, when physical activity is affected | | |
+
+<!-- Keep collaborators real; control genuine environment boundaries. Tests must
+protect behaviour, not a preferred private class/ABI/crate layout. Pure refactors
+may begin with passing characterisation tests; prose edits need no fake red test.
+List not-run checks and missing prerequisites rather than calling them passed. -->
+
+Boundary discovered and evidence that it is actually used:
+Baseline differences and classification / approvals:
+Retained scenario, build, data, configuration, time/seed and artifact identities:
+Known gaps, risks and unavailable verification:
+Reproduction / rollback:
+
+<!-- For Tacview, distinguish generated, structurally validated and actually
+visually inspected. A recording alone is not campaign-correctness evidence. -->
+
+## Review and handover
+
+- [ ] Scope matches the Task; unexpected expansion was agreed with the lead.
+- [ ] Relevant baseline was protected before structural change.
+- [ ] Expected outputs/tolerances changed only with explained evidence.
+- [ ] Applicable checks pass; not-run checks and their impact are explicit.
+- [ ] Request formal review from @fd-starscream-bot using a different PR-author identity.
+
+<!-- These boxes are an evidence aid, not GitHub approval or merge authority.
+Do not merge without a current formal APPROVED review from fd-starscream-bot,
+passing applicable required checks and resolved blocking feedback. New reviewable
+changes require fresh lead approval. Do not self-merge or bypass protection.
+If this PR author IS fd-starscream-bot, keep it unmerged and resolve the identity
+conflict; no other approval or author-string change substitutes for the rule.
+See docs/governance/REVIEW.md. Non-default branch merges may not auto-close Tasks. -->
+
+Next bounded action / outstanding review feedback:

.github/workflows/eech-campaign.yml

@@ -0,0 +1,109 @@
+name: eech-campaign
+
+on:
+  push:
+    branches: [master]
+    paths:
+      - "eech-campaign/**"
+      - "aphavoc/**"
+      - "modules/**"
+      - ".github/workflows/eech-campaign.yml"
+  pull_request:
+    paths:
+      - "eech-campaign/**"
+      # the native module compiles the original C sources
+      - "aphavoc/**"
+      - "modules/**"
+      - ".github/workflows/eech-campaign.yml"
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: eech-campaign-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  test:
+    runs-on: ubuntu-24.04
+    timeout-minutes: 30
+    defaults:
+      run:
+        working-directory: eech-campaign
+    env:
+      # tools/Dockerfile's GCC 12: GCC 14 makes the original sources'
+      # incompatible-pointer warnings errors
+      CC: gcc-12
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          # the intended native target
+          - target: x86_64-unknown-linux-gnu
+            env: ""
+          # EECH's historical platform: platform parity of the same module
+          - target: i686-unknown-linux-gnu
+            env: ""
+          # i686 with en_creat.c's original va_list reinterpretation: patch P1 is behaviour-neutral
+          - target: i686-unknown-linux-gnu
+            env: EECH_ORIGINAL_STACK_ATTRIBUTES=1
+          # an optimised kernel reproduces the C reference too (docs/fpu.md)
+          - target: x86_64-unknown-linux-gnu
+            env: EECH_C_OPT=2
+    steps:
+      - uses: actions/checkout@v4
+      - name: Toolchain
+        run: |
+          sudo apt-get update
+          # the host links PUC Lua 5.1 (crates/eech-world/build.rs), for the target's arch
+          if [ "${{ matrix.target }}" = i686-unknown-linux-gnu ]; then
+            sudo dpkg --add-architecture i386
+            sudo apt-get update
+            lua=liblua5.1-0-dev:i386
+          else
+            lua=liblua5.1-0-dev
+          fi
+          sudo apt-get install -y --no-install-recommends gcc gcc-multilib gcc-12 gcc-12-multilib $lua
+          rustup target add ${{ matrix.target }}
+      - name: Test
+        run: env ${{ matrix.env }} cargo test --locked --target ${{ matrix.target }}
+      - name: Lints
+        if: matrix.target == 'x86_64-unknown-linux-gnu' && matrix.env == ''
+        run: |
+          cargo fmt --check
+          cargo clippy --locked --all-targets -- -D warnings
+
+  windows-cross:
+    # LLP64 (the DCS platform): MinGW cross build, run under Wine
+    runs-on: ubuntu-24.04
+    timeout-minutes: 40
+    defaults:
+      run:
+        working-directory: eech-campaign
+    steps:
+      - uses: actions/checkout@v4
+      - name: Toolchain
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y --no-install-recommends gcc-mingw-w64-x86-64 wine wine64
+          sudo update-alternatives --set x86_64-w64-mingw32-gcc /usr/bin/x86_64-w64-mingw32-gcc-posix
+          rustup target add x86_64-pc-windows-gnu
+      - name: Lua 5.1.5 as lua.dll
+        # the vendored DCS import lib (crates/eech-dc/lua5.1/lua.lib) is not in
+        # the repository, so build the same lua.dll as tools/Dockerfile does
+        working-directory: ${{ runner.temp }}
+        run: |
+          curl -fsSL -o lua.tar.gz https://www.lua.org/ftp/lua-5.1.5.tar.gz
+          echo "2640fc56a795f29d28ef15e13c34a47e223960b0240e8cb0a82d9b0738695333  lua.tar.gz" | sha256sum -c -
+          mkdir -p lua-mingw/include && tar xzf lua.tar.gz && cd lua-5.1.5/src
+          x86_64-w64-mingw32-gcc -O2 -DLUA_BUILD_AS_DLL -c $(ls *.c | grep -v -E '^(lua|luac|print)\.c$')
+          x86_64-w64-mingw32-gcc -shared -o ../../lua-mingw/lua.dll *.o -Wl,--out-implib,../../lua-mingw/liblua.dll.a
+          cp lua.h luaconf.h lualib.h lauxlib.h ../../lua-mingw/include/
+      - name: Test under Wine
+        env:
+          WINEDEBUG: -all
+          CARGO_TARGET_X86_64_PC_WINDOWS_GNU_RUNNER: wine
+          LUA_LIB: ${{ runner.temp }}/lua-mingw
+          WINEPATH: ${{ runner.temp }}/lua-mingw
+        run: cargo test --locked --target x86_64-pc-windows-gnu

AGENT.md

@@ -0,0 +1,135 @@
+# Agent operating policy
+
+Canonical shared instructions for the EECH native campaign programme. Read this
+file before work. `CLAUDE.md` imports it; `AGENTS.md` directs other agents here.
+The enduring charter and linked roadmap are in [PROJECT.md](PROJECT.md).
+The review procedure is in [REVIEW.md](docs/governance/REVIEW.md).
+
+## Start each session
+
+Read PROJECT.md, the active Task and parent Epic, the latest accepted baseline,
+and outstanding PR review feedback. Inspect the actual branch, base SHA and
+working tree; do not overwrite another session's work. The native integration
+branch selected for this bootstrap is `claude/eech-campaign-native-spike-tjkj6g`;
+verify the current agreed target rather than assuming `master` contains it.
+
+Work on a dedicated Task branch and PR. Continue that PR across sessions. Resolve
+the earliest unmet acceptance gate; justify any later-stage probe by the risk it
+resolves. Governance Task [#51](https://github.com/flying-dice/eech_source_code/issues/51)
+must be accepted before campaign/world restructuring. Read its live blockers;
+the presence of these files alone does not mean the gate is accepted.
+
+## Architecture to protect
+
+- One native campaign Lua module, EECH_DC [1], initially original C behind a Rust
+  public boundary. This is not a campaign rewrite into Rust or TSTL.
+- EECH World [2] and DCS World [3] are alternative hosts, each owning its Lua
+  environment and physical world. The reference host may reuse EECH C physics.
+- The campaign owns state, decisions, task policy, logistics and progression.
+  Hosts execute physical attempts and supply observations. Campaign algorithms
+  remain core even when they consume environmental data.
+- Exercise host-owned Lua -> public native module -> world interactions for
+  product acceptance. Private Rust/C tests cannot substitute for that path.
+- A whole-engine DLL is useful transitional evidence, not proof of replaceable
+  world authority. Identify coupling honestly; do not split it before evidence.
+- Do not add a second authoritative physical world inside the campaign, move
+  campaign policy into adapters, fabricate success, or substitute a sidecar or
+  second campaign implementation without an explicit architectural decision.
+- Keep legacy pointers and layouts private. Discover semantic boundaries from
+  source and runtime behaviour; do not impose a speculative trait/ABI/crate plan.
+
+## Behaviour-led red / green / refactor
+
+**Characterise first.** Before changing, splitting or moving behaviour, identify
+and exercise it on a named baseline revision. Establish sociable tests around
+real collaborating campaign components before restructuring them. Control genuine
+environmental inputs, not the campaign decisions being asserted. Record passing
+behaviour, known defects, unsupported cases and unavailable prerequisites.
+
+**Red where meaningful.** For a defect, reproduce the failing behaviour before
+fixing it. For a new capability or discovered boundary contract, show the relevant
+check fails for the missing behaviour, not a setup accident. Do not manufacture a
+red test for a pure refactor or an administrative edit: characterisation tests
+may correctly pass before the move. No mutation-testing framework is mandated.
+
+**Green with a bounded change.** Preserve existing behaviour while satisfying the
+new contract where applicable. Separate structural moves from intentional rule
+changes, bug fixes and balancing. Avoid mock-per-module tests that reproduce a
+chosen implementation rather than exercise the real system.
+
+**Refactor and prove the boundary.** Keep sociable tests green across the move.
+Assert externally meaningful state, invariants, transitions and interactions;
+assert ordering where it has contractual meaning, not a private call graph.
+After separation, demonstrate that host observations actually influence the
+campaign and host actions are actually used. Interface declarations and similar
+recordings do not prove this. Reuse the same behavioural scenarios across old
+and new paths; explain necessary changes to the test entry point.
+
+TDD is a means of protecting behaviour, not crystallising an untested design.
+Do not write tests requiring a preferred private type, directory, crate layout or
+algorithm solely to make an implementation choice permanent. Focused isolated
+tests are appropriate for genuine local contracts and numerical edge cases.
+
+## Baselines and evidence
+
+Run the affected sociable/contract tests and applicable retained regression and
+conformance checks. Include integrated scenarios and Tacview evidence when world
+activity is affected. Select validation by risk: do not demand a full war for a
+prose edit or call a narrow unit run proof of the entire engine.
+
+Record exact revisions, commands, results, scenario/data/configuration, time and
+random inputs, retained artifacts and limitations. Use the optional
+[baseline record](docs/governance/BASELINE_RECORD.md). Distinguish reported,
+reproduced, inferred and unavailable evidence. A generated, structurally checked,
+and visually inspected Tacview recording are three different claims. Recording
+must remain observational; missing samples are not authoritative events.
+
+Never silently re-record golden outputs, widen tolerances, delete failing tests,
+suppress unsupported paths or tune scenarios just to make new output pass.
+Classify differences as regression, approved defect correction, intentional
+semantic change, or expected input/world variation. Preserve old evidence and
+obtain explicit review of changed expectations. Characterisation does not approve
+a legacy defect forever. Exact equivalent-input campaign conformance is separate
+from comparable outcomes across different physical simulators.
+
+The abandoned `eech-core-ts` port is reference material, not an active roadmap.
+Reuse its source archaeology, C harness, scenarios and numerical findings without
+resuming TSTL work or overstating its coverage or independence. Preserve declared
+compatibility decisions; unknown historical behaviour must remain unknown.
+
+## Task, review and merge rules
+
+Each completed Task produces its own reviewable PR/MR, including investigations
+and administration. If too large, agree linked smaller Tasks before bundling work.
+Milestones describe outcomes and evidence, not mandatory implementation edits.
+Seek technical-lead agreement before broadening semantic scope, changing accepted
+contracts or revising milestone intent. Resolving an uncertainty is valid progress.
+
+The designated technical lead is **@fd-starscream-bot**. A current formal GitHub
+**APPROVED** review from that identity is mandatory before merge, alongside passing
+applicable checks and resolved blocking feedback. Comments, review requests,
+other approvals, CI success and agent self-reviews do not replace it. New
+reviewable changes require fresh approval; material base changes require renewed
+validation and review. Use the [PR template](.github/PULL_REQUEST_TEMPLATE.md).
+
+Implementing agents must use a different PR-author identity from the technical
+lead. Never self-approve, self-merge, bypass rules, dismiss blocking lead feedback,
+impersonate reviewers or use another account's credentials. An identity conflict
+is a blocker, not an exemption. Rules and templates are not server enforcement;
+verify effective protection as described in REVIEW.md. Missing permissions or
+unverified protections remain blockers. Governance itself is reviewable work.
+
+Do not close a Task when code is merely pushed or approved. Link the approved
+merge and acceptance evidence first; non-default-branch merges may require an
+explicit issue-state update. Epic and milestone acceptance requires its own review.
+Do not merge this bootstrap or mark #51 complete while its approval, protection
+or instruction-discovery requirements remain unmet.
+
+## Handover at every session boundary
+
+Update the Task/PR with: branch and commit; scope and milestone gate advanced;
+behaviour characterised; red/green/refactor evidence or justified non-applicability;
+boundaries discovered/asserted; actual test results; retained artifacts; baseline
+differences; limitations/blockers; review status; and the next bounded action.
+Keep continuity in GitHub and reviewed repository records, not only chat history
+or ephemeral paths. A failed experiment is evidence, not a completed capability.

AGENTS.md

@@ -0,0 +1,9 @@
+# Agent entry point
+
+Read and follow [AGENT.md](AGENT.md), the canonical shared operating policy,
+before doing work. Read [PROJECT.md](PROJECT.md) for the enduring charter and
+linked milestones. Do not duplicate those policies in this file.
+
+Paths above are relative to this repository root, including when your working
+directory is `eech-campaign/`. Inspect the active Task and review feedback as
+directed by AGENT.md; this entry point does not select a Task for you.

CLAUDE.md

@@ -0,0 +1,7 @@
+# Claude entry point
+
+@AGENT.md
+
+Read [PROJECT.md](PROJECT.md) and the active GitHub Task/Epic before changing files.
+AGENT.md is the single operating policy; do not maintain a separate Claude policy.
+The import above is active and relative to this file, not to the working directory.