reverse-engineering-anything/docs/testing-lanes.md

Testing lanes and provider verification

For test design and developer commands, see Testing REA.

Real-toolchain verification lanes

Linux Rizin debugger ownership changes require a separate real-provider gate on Rizin 0.7.4. Run npm run test:focused -- tests/boundary/providers/RizinInferiorTty.test.ts. This test verifies the caller-selected TTY against the actual inferior's file descriptors, confirms dbg.exitkills=false, detaches through the olj/o- protocol, and checks that the target survives debugger close and writes to the selected TTY. The suite is included in the reverse-tools Vitest and mutation typecheck selections; its real-provider assertions skip when Rizin 0.7.4 is unavailable. Run the focused command above on a host with that provider for the explicit real-provider verification gate.

Each real-toolchain command must require only the host tools needed to prove its stated claim. Use a host-native fixture for host/provider acceptance, and place optional cross-target formats or platform-specific runners in separate commands. Check prerequisites before starting expensive work and name the missing command, target, and lane in any failure message. A lane must not imply that a host or target is covered when it was skipped.

Provider admission accepts Ghidra 12.x (>=12.1) and the JDK range declared by that installation (application.java.min through application.java.max). Current 12.1 releases require JDK 21 or newer and set no maximum. The lanes below still prove behavior on the verified Ghidra 12.1.4 and JDK 21 build.

Ghidra lane	Supported runner/target	Additional local tools
npm run verify:ghidra	Linux x64/arm64 ELF or macOS x64/arm64 Mach-O	Host C compiler, Ghidra 12.1.4, and full JDK 21
npm run verify:ghidra:seeds	POSIX; raw x86 fixture; seed ordering, BSS labels, analyzer readback, not-applicable and unknown analyzer names	Ghidra 12.1.4 and full JDK 21
npm run verify:ghidra:swift	macOS x64/arm64 Mach-O	Host Swift compiler, Ghidra 12.1.4, and full JDK 21
npm run verify:ghidra:riscv	Linux/macOS x64/arm64 host; source-owned RV32/RV64 double-float and RV32 ilp32 ELF through CLI and MCP	RISC-V Clang, LLD, LLVM readelf and objdump, Ghidra 12.1.4, and full JDK 21
npm run verify:ghidra:switch	Linux x64 ELF; GCC/Clang optimized and stripped switch fixtures	GCC, Clang, GNU nm/objdump/strip, Ghidra 12.1.4, and full JDK 21
npm run verify:ghidra:aarch64-jump-table	Any supported Ghidra host; AArch64 ELF; signed/unsigned byte/halfword tables; host ARM64 Mach-O	Clang with AArch64 target support, Ghidra 12.1.4, and full JDK 21
npm run verify:ghidra:cross-format	Any supported Ghidra host; also analyzes AArch64 ELF, x86-64 PE, and x86-64 Mach-O	clang, LLD, and lld-link in addition to host-lane prerequisites
npm run verify:ghidra:windows	Controlled Windows x64 with native x86-64 PE; -- --x86 selects native x86 PE	Ghidra 12.1.4, full JDK 21, and the matching native artifact
npm run verify:ghidra:dol	Any supported Ghidra host; source-owned PowerPC GameCube DOL fixture built from a linker script (never executed)	Clang with PowerPC target, LLD, GNU readelf, Ghidra 12.1.4, and full JDK 21

The optional Swift lane compiles a source-owned fixture and checks modern and provider-demangled procedure names through real stdio MCP, with and without the dSYM companion. It checks unfiltered, structs, and literal-name filters, same-address alias evidence, deduplication, and explicit unresolved categories. It does not add a Swift prerequisite to the host C fixture lane.

The host lane also verifies namespaced C++ ABI symbols and annotation identity. That facet requires Ghidra's matching demangler_gnu_v2_41 native component in GPL/DemanglerGnu/os/<platform>/ or build/os/<platform>/, in addition to the native decompiler. The namespace verifier reports missing or non-executable components before testing the fixture. REA does not build or install them.

Windows native conformance runs with npm run verify:windows-native and does not require Ghidra or Java. It includes Job Object commit-limit controls (a limited child that reaches the limit, a control that does not, and invalid limit rejection). An optional independently compiled Windows fixture adds in-place reparse, breakaway, environment, and token observations. The separate scripts/verify-windows-native-inspection.mjs lane uses the .cache/windows-native-inspection-fixture artifact built on Linux with node scripts/build-windows-native.mjs --inspection-fixture. Its controlled first deletion failure checks retained NTFS ownership across a same-volume rename, later-query handle-based recovery, successful-admission caching, and environment-teardown cleanup. It neither establishes the cause of the intermittent #1701 host failure nor replaces verification with the shipped production artifact. npm run verify:ghidra:windows:package additionally packs and installs REA into an isolated prefix and runs ordinary-user CLI/MCP operations against their canonical schemas. Run the controlled fixture generator before that lane, or supply an installed package root and an explicit fixture as arguments. The host-native Ghidra lane also verifies native value tracing through the production CLI and a separate stdio MCP process. It compares complete dependency graphs, validates Evidence and upstream/workflow profiles, checks capability discovery, and closes the MCP session. No provider or transport is mocked. It also validates every advertised input/output JSON Schema and the exercised MCP outputs, probes address spelling and name/address ambiguity, and checks direct versus targetless calls, byte-read completeness, invalid input diagnostics, CLI/MCP parity, atomic annotation rollback, refreshed inventories, unchanged executable bytes, and discarded edits after reopen. A deliberately long temporary path exercises private Unix socket allocation and cleanup, including cancellation after a real headless process launches. Native annotation probes reject NUL and unpaired Unicode surrogates without partial edits or a broken bridge, preserve supported Unicode and control text, and check lossless malformed-text diagnostics. Memory-to-file mapping is checked against original artifact bytes. The fixture also stores a pointer one byte past a function entry; exact xrefs, raw procedure references, and CLI/MCP dossiers must retain that data edge. A valid legacy snapshot reconstructs the former omitted edge under its older profile; CLI and MCP must reject that binding with a mismatch reason and recovery advice. The rejected open must preserve the active live session. Exact external entries must resolve while retaining an empty body; unknown external addresses remain unresolved and external annotations are rejected. An adversarial regex over a full 12 KB literal must report stack exhaustion as a resource constraint, preserve live annotations, and allow complete literal searches afterward; CLI and MCP must agree on both results and recovery advice. Real snapshot lifecycle checks retain edited API results as Evidence while rejecting immutable snapshot saves and imports before and after a repeated open of the same target. They verify unchanged live annotations and run identity, absent rejected output files, an unchanged source snapshot, and successful snapshot import/save after closing and recreating the database. The pristine snapshot is written by an independent real CLI session. They also start a real annotation and snapshot close concurrently: the edit must succeed, the snapshot must be rejected without creating a file, and the edited session must remain usable until explicitly closed. Source-admission probes change a caller-owned fixture after open_binary but before the first Ghidra query. They require an actionable artifact_changed error preserving both digests and the selected path, unchanged provider availability, failed-copy cleanup, and successful recovery after reopening. After import, deleting that source must preserve the captured database identity. Instruction inspection and containing-function lookup also agree across hexadecimal case, leading zeros, and encoded default address-space spellings. The same source-acquisition workflow exercises missing and directory-replaced inputs, plus real read-permission denial on a host that enforces it. Permission fixtures must probe the relevant read, write or search operation before relying on POSIX mode bits: a non-root Linux process may bypass DAC through capabilities. requirePermissionDenial reports an explicit skip when the host permits the probe; it restores the probe's permissions before skipping. Such a skip leaves real permission coverage unverified. Filesystem integration separately checks selected-platform routing and exclusive creation. Namespace annotation probes compile a separate host C fixture with C++ ABI symbols, avoiding a C++ runtime prerequisite. Real Ghidra demangling supplies duplicate leaf names in two top-level namespaces and a nested namespace. The workflow verifies leaf and qualified renames, repeated reuse of fully qualified readback, lookup by the returned name, literal namespace-like leaf names, rejection of empty qualified leaf names without changing comments, CLI behavior, and independent CLI/MCP database ownership. Large-result probes compile initialized host-native data sized from the pinned MCP SDK receive budget. Real byte reads, annotation edits and function dossiers exceed that budget while preserving the connection and active analysis run. Successful-result and oversized-error delivery constraints must identify their successfully retained Evidence records; export must recover every source byte, complete annotation and original error diagnostic, with CLI parity and an unchanged executable. Formatter integration separately checks unacknowledged Evidence recording. Long ordinary procedure names and encoded address-space selectors must produce normal validation errors without exhausting Java's regex stack or losing the private bridge connection. Each rejection is followed by a real provider lookup; address-like literal names still resolve exactly after annotation. Malformed annotation readback, memory completeness, and inventory data remain separate SDK/provider integration cases; success from a real provider cannot establish rejection of a contradictory provider response.

The Linux switch lane checks dense, sparse-with-holes, shared-body, nonzero, negative, and nonexact JSON integer labels plus a comparison-only control. Independent source labels, ELF file bytes, table slots, and bounds branches define expected case/default destinations. Production CLI and MCP must agree; debug labels must retain their signed values. For stripped negative fixtures, the independently checked 32-bit dispatch permits equivalent unsigned labels only alongside the reported low-confidence undefined4 parameter; original source signedness remains unknown and the ABI residual must remain visible. unsafe labels remain unresolved and every recovered destination is retained. The compiler oracle also injects malformed records to check that its assertions reject missing/default-confused labels, wrong destinations, and numeric guesses. It also invokes the actual bridge methods on detached Ghidra model objects to check ambiguous dispatches, signed literals, precision bounds, shared targets, and conflicting labels. This reflection fixture depends on the pinned Ghidra model and does not claim a compiler produced those synthetic graph shapes. Pass --entrypoint /path/to/installed/rea-agents/scripts/rea.mjs directly to scripts/verify-real-ghidra-switch.mjs to verify an installed package through the same compiler oracles and CLI/MCP checks.

The cross-format Ghidra lane also analyzes an optimized AArch64 ELF switch fixture. It checks the recovered case values against the source cases and requires unresolved table bounds or case mappings to remain visible as residual unknowns.

npm run verify:inspector requires the supported Node.js runtime and installed REA dependencies. CI runs it on Linux and macOS x64/arm64 and Windows x64. It starts owned loopback Node Inspector fixtures and verifies discovery and passive observation through the CLI and stdio MCP, including special filenames, unresolved discovery locations, and independently resolved loaded scripts. Double-quote filenames are tested on POSIX only because Windows does not support them.

Node Inspector verification does not establish Electron GUI or other engine behavior. Browser and Inspector share CDP transport/value and file-location helpers; their producer workflows remain distinct.

Android APK analysis

The real Android lane also runs on Windows x64 with its matching bundled native controls. It verifies bidirectional protocol input, CLI/MCP result parity, real MCP cancellation/disconnect and Java exit after abrupt CLI owner termination. The forced Windows exit does not exercise the POSIX SIGTERM handler and can leave temporary workspace files. verify:windows-native separately checks binary input, backpressure, EOF and pending-write job closure.

npm run verify:android requires an existing Java 17+ and an explicit REA_JADX_MCP_JAR for jadx-headless-mcp 0.7.1. Set REA_ANDROID_TEST_APK to the fixed public ApiDemos v6.0.18 fixture. Obtain both with the explicit npm run fixtures:android command; files are SHA-256 verified and kept under ignored _reference/. No Gradle build, Android SDK, emulator or application execution is required. The lane compares real CLI/MCP package, class search, class inventory, method decompilation and incoming references. See Android analysis for boundaries and resource budgets.

The Real Android metadata bridge workflow compiles the Java bridge and runs its metadata regressions against the same checksum-pinned engine and public APK. It runs on branch pushes affecting its bridge, fixture or engine inputs, and can also be dispatched manually or called by another workflow. It uses the runner's JDK with two active processors and a 512 MiB heap; it needs no npm dependency installation. This lane checks exact raw/display class selectors, raw member names and field DEX types used by mapping lookups, and the existing cold-metadata regressions. It does not establish CLI/MCP parity or application execution behavior. The broader verify:android lane checks CLI/MCP parity; neither lane runs the APK.

verify:jeb requires a caller-started JEB client serving MCP at REA_JEB_MCP_URL (default http://127.0.0.1:8425/mcp) with a project already open, and exercises shared application/provider client inspection, unit listing coverage, and method decompilation. The script records the engine's exact response to open_jeb_project; the earlier licensed 5.48.0 headless verification did not advertise that tool. REA does not install or launch JEB; see JEB analysis for the bring-your-own boundary. The earlier licensed build was verified against JEB 5.48.0 serving jeb-mcp-server 1.3.0 at four scales: a compiled Java class fixture, a locally built signed probe APK with a launcher activity (manifest, v1/v2/v3 certificates, dex bytecode, filtered and paginated listing), the published Signal 8.30.3 universal release APK (109 MB, R8-processed Kotlin/Compose, four signature schemes, native arm64 ELF units) with MainActivity method decompilation through both CLI and MCP, and the official Flutter Gallery 2.9.2 release APK (112 MB): per-ABI libapp.so Dart AOT snapshot images analyzed as native code (75k methods on arm64) and decompiled through both CLI and MCP with explicit unit selection, plus the thin Java wrapper and decoded manifest. The dedicated Dart snapshot processor and run_script remained GUI-only surfaces in that headless build. The separate Linux CE native audit uses the official CE 5.48.0.202610071539 direct CLI launcher and real CLI/stdio MCP ELF workflows. That audit does not qualify CE Java/Android decompilation or GUI project opening, and it does not replace the licensed verification above. Authenticated IPA and macOS application inventory projection is documented in Apple application analysis.

npm run verify:harmony requires no SDK or engine. It projects the pinned harmony-next-pipeline 1.0.1 release HAP from an MIT-licensed source repository, fetched by npm run fixtures:harmony into ignored _reference/harmony-integration/, set as REA_HARMONY_TEST_HAP. The lane exercises inspect_artifact, projection determinism, Stage-model path hints, exact component sets, signing-entry limitations, and CLI/MCP parity. The release artifact has no independently established compiler/SDK or reproducible source build; the default lane does not verify an installed package. See HarmonyOS package inventory for those proof boundaries. HarmonyOS .har libraries are deliberately not suffix-classified because .har also names HTTP Archive JSON.

Synthetic producer regressions run independently:

npm run test:focused -- tests/boundary/process/jadxIntegration.test.ts tests/boundary/process/androidAnalysisMcp.test.ts tests/boundary/process/androidMappingMcp.test.ts

Apktool resource decoding

npm run verify:apktool -- --apk PATH exercises both Apktool operations against a real launcher (REA_APKTOOL_COMMAND or PATH) and one APK. No download or build step is involved. Record: apktool 2.7.0-dirty (Debian packaging) on Linux with OpenJDK 25, against a signed aapt2-built probe APK carrying two resource locales — launcher version, on-disk digest agreement with the reported target identity, apktool.yml metadata projection (1.2.3, SDK 24–34), manifest package agreement, and both the default and de string tables projected through the locale option. Parser goldens in src/apktool/ApktoolDecodeOutput.test.ts come from the same decode. See Apktool resource analysis for boundaries and budgets.

Flutter build identification

npm run verify:flutter -- --apk PATH [--plain-apk PATH] runs both pure static operations against a real Flutter APK, plus an optional non-Flutter APK for the negative check. No engine or external tool is involved. Record: Flutter Gallery 2.9.2 (112 MB, arm64-v8a / armeabi-v7a / x86_64) — every ABI reports the same agreeing snapshot hash (65817c30a78bb44c3dc3771876b6010a, two sections each) read 20 bytes after the Dart snapshot magic, per-ABI engine GNU build-ids, no labeled Dart version string (expected for a release engine build), and the Android clang/LLVM toolchain line; the AOT inspection locates all four snapshot symbols through the dynamic symbol table on both ELF64 and ELF32 images, validates the data-section headers, agrees with the identification hash, and counts 756 package: and 201 dart: URIs on arm64-v8a and x86_64, and 212 / 48 on armeabi-v7a, from each isolate data section. Responses retain at most 500 URIs per family and 2000 class-like tokens. Gallery's pool projection and overall AOT coverage are partial while its structural libapp.coverage is complete. A plain aapt2-built APK reports not-detected with complete coverage. The suite uses self-contained synthetic ELF images and recorded scanner prefixes, with no private local-file dependency. The verifier now compares compiled CLI and stdio MCP Evidence on every ABI, including data-header kind and declared lengths. Add --expectations scripts/fixtures/flutter-gallery-expectations.json for the documented Gallery APK to assert its pinned digest, hash, exact engine labels and per-ABI URI counts. The verifier requires complete structural snapshot coverage, reports bounded pool coverage as partial/unverified, and reports exact labels/counts on other APKs unverified without an expectations file. See Flutter build identification for boundaries and budgets.

ADB device acquisition

npm run verify:adb exercises ADB observation and transfer operations against a live device or emulator through the caller's adb binary (REA_ADB_PATH or PATH). No SDK installation or emulator management is involved. --install-apk is unavailable until the selected APK identity and device cleanup ownership can be established safely. The transfer probe creates and removes a temporary device file. --pull acquires a real package and re-digests the pulled files on disk against the returned SHA-256 values; --serial selects a device when several are attached.

Record: adb 34.0.5-debian on Linux against an Android 14 (API 34) x86_64 emulator. The lane covered the complete observation surface — 337-process listing, 287 binder services including AIDL /-suffixed names, 92 features including hex GL versions, display size/density, window focus (legitimately null on headless devices), settings get global adb_enabled, bounded logcat, directory listings — plus a real two-APK split set (base.apk plus split_probe.apk, built and installed through install-multiple) pulled with byte-exact digests, a push/pull roundtrip verified by the device's own sha256sum, a 1.3 MB screen capture with PNG dimensions, dumpsys package projection, and the full lifecycle: unique resolution before launch, launcher-activity start through cmd package resolve-activity with am start -W, the started app visible in the process listing, force-stop, uninstall, and zero matches after removal. System-package pulls whose APKs keep non-base.apk names report the unknown role with the file-name basis, verified with com.android.settings (single 73.9 MB APK). Parser goldens for adb devices -l, getprop, pm list packages -f, and pm path come from the same device. See ADB device analysis for boundaries and budgets.

Optional NativeAOT Ghidra analysis

The source parser lane is independent of the optional annotation JAR. Its synthetic RTR 9.1 PE golden checks captured-byte digest admission, rehydration into an immutable parser overlay, MethodTable/slot relationships, and frozen literal extraction; provider tests exercise inspect_native_load_image and address-based inspect_native_data_type with the database result still unavailable. These tests do not establish Windows Ghidra host behavior.

This lane is separate from the default native lane. build:fixtures:nativeaot requires an existing .NET SDK 8.0.416, the platform NativeAOT compiler/linker and runtime pack 8.0.22. It builds benign sources into ignored _reference/ and records independent symbol/directory/SHA oracles; it never executes the target. Linux also builds stripped, ordinary-native and small malformed/ambiguous/layout negative inputs. The optional Windows fixture workflow builds a PE on a Windows runner; analyzing that PE on Linux does not verify a Windows Ghidra host. On a local Windows host, the .NET 8 ILCompiler finds the linker only through Visual Studio 2019 or 2022 (vswhere -version [16,18)). With only Visual Studio Build Tools 2026 installed, open its x64 Developer Command Prompt. In that same cmd.exe session, run set IlcUseEnvironmentalTools=true and then npm run build:fixtures:nativeaot. Running vcvars64.bat as a child of PowerShell does not preserve its environment in the parent PowerShell session.

Build the clean pinned upstream adapter with build:ghidra:nativeaot, then set REA_GHIDRA_NATIVEAOT_JAR. Run verify:ghidra:nativeaot -- symbols, -- stripped, -- ordinary, -- unsupported, -- malformed, -- ambiguous, and -- loader-failure, and -- default-native separately. The loader-failure mode source-builds a deliberately failing JDK 21 initializer and checks the actual loader cause and cleanup. The default-native mode verifies ordinary analysis with the optional extension disabled. The real MCP lane checks source identity, inline format discovery, metadata relationships/slots against independent compiler symbols, frozen strings, pseudocode and owned cleanup. Set REA_NATIVEAOT_PROOF_CLI=1 for one equivalent CLI type inspection; this costs an additional full import. Select an unpacked installed package with REA_NATIVEAOT_PROOF_PACKAGE_ROOT, a fixture directory with REA_NATIVEAOT_PROOF_FIXTURE_ROOT, and optional evidence capture directory with REA_NATIVEAOT_PROOF_CAPTURE_DIR (absolute paths).

Keep builds/imports sequential on small hosts; scope GHIDRA_HEADLESS_MAXMEM (e.g. 768M) to this command and use CPU affinity if needed. REA does not install or upgrade Java, Ghidra, .NET or native toolchains. See the supported layout and provenance.

IDA MCP adapter

npm run verify:ida -- --target /absolute/path/to/program --procedure main uses the existing REA_IDA_MCP_CONFIG registration. It installs no engine, Python package, or compiler. The target must already be open in the GUI for the legacy or native attached profile; the database-supervisor headless profile opens a digest-verified private copy. A caller-supplied fixture keeps prerequisites limited to the selected engine and host. tests/conformance/ida/inventory.c provides an optional small native fixture source with exported rea_fixture_add and rea_fixture_subtract functions. For their switching check, use --second-procedure rea_fixture_subtract --second-marker='- 13'.

The lane invokes the production CLI dispatcher and connects the pinned MCP client SDK to the production REA server. It verifies function Evidence and CLI/MCP parity, inventory/search, pseudocode, instructions, xrefs, malformed input, original-input preservation, and lifecycle cleanup. The headless profile also checks containing procedures, the admitted program as the document, address context, and feature tracing. For headless analysis it confirms that the owned database IDs disappear from upstream discovery and private workspaces are removed. For attached analysis it confirms the existing GUI target remains reachable with the same input identity. --package-root selects an installed/extracted REA artifact. --report writes private local observations with mode 0600; the console summary contains no target paths or upstream output.

Adapter and composition tests cover producer parsing, pagination, canonical entries, external callees, target switches, cancellation draining, snapshot replay exclusion, ownership failures, and incomplete cleanup. They do not establish real IDA operation. The initial real workflows cover legacy upstream 1.4.0 on a Windows GUI and the modern supervisor at upstream commit c133c3853faa111a9b00ee615c013b720d0c4acd with Windows x64 IDA 9.3, and the headless profile on macOS arm64 with IDA Professional 9.1 (build 9.1.250226) at the same upstream commit, using tests/conformance/ida/inventory.c built with clang -O0 -arch arm64 and the procedure _rea_fixture_add (Mach-O C symbols carry a leading underscore). Before the headless session waited out the transient busy health a worker can report right after opening, that lane failed intermittently on that host. Linux headless, modern attached GUI tools, other engine versions and architectures remain unverified; see the provider guide.

Native IDA MCP verification uses "protocol": "native" with an already-running native plugin and requires --second-procedure other_function --second-marker unique_second_body_text to exercise GUI function switching and a return to the first function. These options remain optional for other profiles. The literal marker must appear in the second function's pseudocode and be absent from the first. Native cloud/widget failures are reported without hidden retries.

The initial Linux x64 IDA Free 9.4 run used native MCP 1.1.2 with a local navigation correction. Windows x64 verification uses IDA Pro 9.3, the two-function public fixture, and the shipped Windows whoami.exe application. The native plugin was built from upstream c9d15d98e8a143fc74f6e14037c48f63889b6d6c, including the merged navigation fix, with a local loopback-listener change. Production CLI/MCP parity, switching, identity and cleanup passed. Correctness and performance measurements are recorded in the Windows native verification notes. Stock 1.1.2 switching, native macOS, and Windows IDA Free remain unverified. See the native bridge prerequisites.

Kuna lane

npm run verify:kuna requires a caller-supplied kuna in REA_KUNA_COMMAND and a host C compiler (REA_CC, default cc) with strip and nm. It builds a source-owned fixture with non-inlined calls, with symbols and stripped, then verifies through MCP that callers, callees and exact incoming xrefs agree with those calls, graph/path composition reaches the callee, reference-only queries leave project/staging entries unchanged, repeated relationships reuse the graph, and the first cold pseudocode answer comes from a single-function run or an export that already finished, assembly waits for and serves from that export, and pseudocode and both dossier facets match their recorded file spans and digests, a fresh server reuses the export without rebuilding it, staging is empty after a completed export, and no owned process survives. Pass --targets DIR --functions a,b to add caller-built binaries laid out as DIR/<flavor>/<name> beside <name>.stripped; addresses come from the symbol-bearing build. --compare-ghidra with GHIDRA_INSTALL_DIR runs those caller-built functions through Ghidra (it refuses to run without --targets and --functions, an empty matrix, or names missing from every supplied target) and reports latency, sampled process-tree CPU and peak memory, UTF-8 response size, and goto and warning-marker counts. Those counts are not a correctness judgement, and model tokens are not measured. Reference timings and response sizes are reported separately; reference caches expire when the server closes. --report PATH writes the JSON report.

The first matrix run used Kuna 1.833 and Ghidra 12.1.4 with JDK 21 on macOS arm64. It covered zlib 1.3.2 minigzip, coreutils 9.9 sort and the sqlite 3.54.0 shell, built at -O2 for ELF x86-64, ELF AArch64, Mach-O arm64 and Mach-O x86-64, each with symbols and stripped. A separate Linux x86-64 run verified the fixture and system zlib 1.2.13 / SQLite 3.40.1 libraries through MCP, including cached byte spans and a Ghidra comparison. See the Linux report for measured scope and limitations. Windows, PE and universal Mach-O remain unverified by this lane.

Kuna 1.986 passed the fixture and a separate small, source-owned Mach-O with symbols and stripped on macOS arm64, including the Ghidra comparison and owned cleanup. Linux and other hosts have not been verified with 1.986; this check is not a repeat of the larger 1.833 matrix.

Kuna 1.988 passed the real-provider lane on macOS arm64 with the generated Mach-O fixture and caller-built symbol-bearing and stripped x86-64 ELF fixtures. Linux and other hosts have not been verified with 1.988; this is not a repeat of the larger 1.833 matrix.

Kuna 1.989 passed the real-provider lane on macOS arm64 with the generated symbol-bearing and stripped Mach-O fixtures, including cold and warm reads, references, and cleanup. Linux, other hosts, and the larger 1.833 target matrix were not verified with 1.989.

DOS Ghidra analysis

npm run verify:ghidra:dos requires the supported Ghidra and JDK installation on Linux x64 or macOS x64/arm64. It generates a source-owned MZ fixture without a DOS emulator or compiler, then checks real 16-bit decoding, segment relocation, near/far calls, decompilation, disjoint function body ranges, stable CLI/MCP observations, unchanged source bytes, and owned process/project cleanup. Raw p-code address-space selector tokens are reported separately from the stable observation comparison. Linux x64 and macOS arm64 are verified; macOS x64 remains unverified. This lane is separate from host-native and optional cross-format verification. See DOS analysis.

npm run verify:ghidra:com uses a generated headerless fixture with no compiler, DOS emulator or game data. It exercises explicit admission, BinaryLoader entry preparation, measured register context, whole-file byte readback, source offsets, unmapped PSP/partial reads, actual decompilation, CLI/MCP parity and owned cleanup. Both segmented-address lanes reject oversized default, explicit-space and encoded-space coordinates through real reads, function queries and annotation attempts. Rejected annotations must preserve the live function dossier; CLI and MCP must report the truncation constraint, while leading-zero coordinates still resolve correctly. It has the same Ghidra/JDK prerequisites as the MZ lane. Neither lane claims DOS runtime or PC-98 device execution.

npm run verify:ghidra:6502 uses ca65 and ld65 to build the source-owned NMOS 6502 fixture from tests/conformance/ghidra/6502.s and 6502.cfg. Set REA_CA65 and REA_LD65 for non-PATH installations (tested with cc65 V2.19). With an installed Ghidra 12.1.4 and full JDK, it verifies real CLI/stdio MCP mapping parity, exact source bytes, functions, a direct call reference, decompilation, 16-bit overflow rejection without replacing the active session, and runtime cleanup. Windows additionally requires the matching native addon; its verifier lineage remains explicitly unavailable outside the provider's retained Job Object, as in the other Windows lanes.

npm run verify:ghidra:m68k needs no assembler: its 22-byte big-endian 68000 fixture is source-owned annotated hex in the verifier, loaded at 0x1000. With an installed Ghidra 12.1.4 and full JDK on Linux or macOS, it verifies the 68000:BE:32:default BinaryLoader import, real CLI/stdio MCP load-image parity, exact source bytes, both functions, the bsr.b call reference, decompilation through address and file offset, rejection of a misaligned entry and of a mapping past 0xffffffff without replacing the active session, and runtime cleanup. Windows Ghidra P0 does not admit m68k, so this is not a Windows lane. The lane has been run on macOS; Linux has not been exercised.

npm run verify:ghidra:raw uses a generated ARM little-endian headerless image through the compiled CLI and stdio MCP server. With the same Ghidra/JDK prerequisites, it checks preparation and bridge startup, verified load mappings, instruction and byte readback, canonical-entry session reuse, replacement when the entry changes without provider_id, unchanged source bytes, and owned cleanup. Six-byte x86 and x86_64 controls also verify supported compiler specifications, independent mappings, unchanged bytes, and decompilation of a known return value. AArch64 remains covered by deterministic tests rather than this real-provider lane. See raw binary analysis.

npm run verify:ghidra:address-context runs on Linux with a caller-installed Ghidra, full JDK, and an x86-64 ELF GCC toolchain (CC selects another compatible compiler). It builds tests/conformance/c/pie-address-context.c as a PIE ELF without executing it. The compiled CLI and real stdio MCP workflow must expose the measured image base, retain segment address spaces, agree on the overview, resolve explicit and numeric file offsets, and reject unmapped analysis addresses with actionable guidance. It checks the reverse address-to-file-offset mapping, file-coordinate Evidence locations, and the suggested CLI search and load-image commands. The verifier compares mapped bytes with the source file and checks source identity, session preservation after failed lookup, owned runtime cleanup, and verified empty process lineage. This lane does not establish Windows or macOS real-provider coverage.

For source-identity boundaries that a normal single-file import cannot create, tests/conformance/ghidra/ReaFileOffsetSourceProbe.java exercises the production bridge against real Ghidra FileBytes. It rejects auxiliary and partial sources, keeps distinct outcome codes for uninitialized and unmapped addresses, checks the original target's round trip, and retains ambiguity when duplicate target bytes have distinct loaded mappings. Injected decompiler failures and cancellation verify that completed source/function resolution facts and backend failure codes survive; these injections do not establish natural engine failures. On Windows, after creating the source-owned PE fixture, run from the repository root:

node scripts/create-ghidra-windows-fixture.mjs
$proofRoot = Join-Path $env:TEMP 'rea-file-offset-source-proof'
New-Item -ItemType Directory -Path $proofRoot -Force | Out-Null
Copy-Item bridge/ghidra/ReaGhidraBridge.java, bridge/ghidra/ReaGhidraExtensions.java, tests/conformance/ghidra/ReaFileOffsetSourceProbe.java -Destination $proofRoot
& "$env:GHIDRA_INSTALL_DIR/support/analyzeHeadless.bat" $proofRoot source-proof -import "$PWD/build/fixtures/rea-ghidra-windows.exe" -scriptPath $proofRoot -postScript ReaFileOffsetSourceProbe.java -deleteProject

Require the REA_FILE_OFFSET_SOURCE_OK marker and no script errors; Ghidra may exit successfully even when a post-script fails. The probe edits only its temporary analysis database and does not execute the fixture. This producer regression supplements public CLI/MCP coverage and does not establish Linux host behavior or native process-supervision controls.

Apple Interface Builder archives

npm run verify:interface-builder compiles the source-owned AppKit XIB into a real .nib with Xcode ibtool, wraps it in a temporary app bundle, and runs the compiled CLI and stdio MCP server with production providers. It compares their decoded results and checks the view hierarchy, outlet, action, evidence coverage, and truncation status. It also recompiles the XIB with toggled control booleans and requires both decodes to match ibtool --export for each source. Storyboard compilation additionally requires an installed iOS platform.

Keep the provider-specific acceptance path independent from optional cross-compilers. Cross-format failures belong to the cross-format lane and must not make native host acceptance unavailable.

Native platform baseline in CI

Pull requests that change only root README*.md files, docs/, AGENTS.md, CONTRIBUTING.md, CHANGELOG.md, SECURITY.md, native/windows/README.md, or the GitHub issue and pull-request templates run formatting and generated-document validation, including the small onboarding example consistency suite, without the source-test shards or native package lanes. Classification compares the PR head with its merge base, so later base-branch changes do not expand that scope. Static and portable-test aggregate jobs remain present and fail if classification fails. Every code change retains the complete portable Linux suite; individual test files are not selected by imports or filenames.

Ordinary PRs also produce an observational test-selection plan. The candidate leaf-adapter ownership table lives in scripts/ci/test-selection.mjs, separately from provider-job scopes. It proposes keeping affected owners plus every unowned test, including all CLI/MCP, acceptance, composition, conformance, and process tests. Shared inputs, unknown paths, missing owner tests, and removed tests fall back to the complete inventory. Renames account for both paths. Main, scheduled, manual, and ci:full runs always propose the full suite.

The source group and every built shard together still execute the complete suite. The existing report merge compares the proposal with actual module outcomes and retains .cache/ci-test-selection/report.json in test-results, including selected and omitted paths, changed input paths, missed failures, unobserved files, scheduling weights, PR head and executed merge SHA. This adds no test reruns or runner jobs. Missing, duplicate, skipped, interrupted, or unhandled results cannot establish a complete comparison; even a complete passing comparison is not proof of ownership completeness. Before enabling omissions, review representative full-run comparisons and cross-owner consumers, including failures, and measure completed CI feedback and runner demand. There is no automatic promotion or switch to skip tests. The initial proposal measures only leaf producer-test omissions; it does not promise the savings of a broader affected-test planner.

scripts/ci/plan.mjs selects lanes using the ownership table in scripts/ci/scopes.mjs. Each lane record owns its path and test selection rules and required fast/full jobs; workflow contract tests verify the mapped jobs. Provider implementation changes run their real-provider verifiers and the Linux x64 installed-package/Inspector baseline. The apple scope owns the Apple artifact lane; the package scope owns the multi-host package matrix and the Windows curated lane, including Windows native controls, Inspector, setup, configuration and packaging entrypoints. Runtime sources under src/ or bridge/ without an owner, and the authored skill, run the Linux suite and the Linux x64 installed-package/Inspector baseline only. Test-only changes run the Linux suite without native package verification, except test files that the Apple or Windows curated lanes invoke explicitly, which also select that lane. Packet-capture and signal-recording tests and their producers select the corresponding real-tool lane; both also run in full qualification. The shared Linux resource-limit source and its tests select both. Each lane requires its Linux kernel resource-limit cases and all seven descriptor-cleanup cases to pass in the focused Vitest report. A missing report, missing case, skipped case, or duplicate identity fails the lane; unrelated host-specific skips remain allowed. These tests use synthetic producers. The separate real TShark or sigrok verifier establishes the actual-tool observation. Edits to verifier entrypoints run only by manual, licensed or self-hosted workflows run static checks and the Linux suite. Website changes run website checks. Provider workflow changes validate the workflow and run that provider's verifier. Changes under bridge/ghidra/ also select the ghidra_bridge lane, which compiles the packaged bridge sources against the pinned Ghidra 12.1.4 jars with npm run verify:ghidra:bridge-sources. That compile step never launches Ghidra or imports a target, so a missing import fails the PR instead of every Ghidra session. The launcher, compiler and package checks share named required source roles. The manifest must match the complete Java cohort, so deleting a helper or adding an unregistered dependency cannot silently change the shipped set. The preflight reuses the provider's Java Properties parser with the pinned Node toolchain's native type stripping, without an npm install or runtime build. The shared release policy owns the verified version and JDK major. CI checks the archive pin before downloading and verifies the extracted SDK version; changes to these source owners also select the compile lane. After compiling, the lane launches headless Ghidra on source-owned instruction bytes with -noanalysis. The ARM/Thumb probe sets each instruction's mode explicitly and checks the bridge's decoder-context readback. The decode probes check that ARM:LE:32:Cortex, ARM:LE:32:v8-m and Xtensa:LE:32:default decode their bytes from each language's default context. The bytes are never executed, and edits to those probe scripts also select the lane. Manual-provider/Android workflow changes run workflow validation only. The four-version SQLite compatibility matrix has its own sqlite scope for SQLite implementation, contracts, CLI/MCP registration, fixtures, and shared runtime boundaries. SQLite and server-lifecycle test edits and the SQLite npm scripts also select it. Unrelated code edits retain the complete portable suite on the pinned Node version without starting four additional compatibility jobs. Full qualification retains every SQLite version. Shared process/contract/CLI boundaries, dependencies, other packaging inputs, unknown paths, and changes to the planner select the fast Linux x64 baseline: portable code checks and the installed-package runtime lane, without provider lanes or the multi-host package matrix. A shared file with exactly one owner still selects that owner. Full qualification still selects every lane. CI timing collection, timing history, and test-report helper edits retain static checks and the complete portable suite without selecting unrelated host/provider lanes. Classification, qualification, gate, release receipt, and unknown CI scripts keep that fast baseline. Package script-only edits select their owning verifier when its entrypoint is identifiable; unknown or shared script changes select the fast baseline. Android build-script edits retain the Linux suite and workflow validation; real Android verification stays manual. Renames account for both paths.

The agentenv scope owns the environment application and server modules, the environment-serve and environment-export command registration, the shared streamable HTTP listener, scripts/verify-agentenv-environment.mjs and its workflow real-agentenv.yml. The lane exports a bundle from the source-owned Grafana-shaped capture through the built CLI, packs the checkout into the bundle's image and runs agent-env run on the hosted runner's Docker with the pinned agentenv-framework release; it passes when the smoke task passes and no container remains. It claims nothing about any hosted platform or another framework version.

This narrower selection trades PR-time coverage for runner capacity. A change to unowned shared runtime code that fails only on arm64, macOS, Windows, the Apple artifact lane, or a provider lane that does not own the file is not caught on the PR; scheduled full qualification reports it after merge. Ownership globs do not follow imports. Use the ci:full label when a change may have native-host or provider effects outside its owners.

Ordinary main pushes use the same owner scopes as PRs. Their diff starts at a successful ancestor push run of this CI workflow, rather than the preceding push: changes from failed or cancelled intermediate runs remain in scope. Missing, untrusted, or non-ancestor history selects the fast Linux x64 baseline: portable checks and the Linux runtime lane. The lookup is bounded to the last 100 successful push runs; lookup failures broaden selection. Every code change still selects the source group and all eight fast built shards, without coverage instrumentation in this fast tier. Ordinary PR matrices cancel remaining shards after the first failure; passing PRs execute the complete suite. Main, ci:full, scheduled, manual, and release qualification retain every shard's results after a failure. Available failed-shard reports remain diagnostic only: failed, cancelled, skipped, or missing selected shards cannot pass the gate or publish successful timing history. This deliberately avoids relying on an import graph for generated files, filesystem inputs, subprocesses, or bridges. The portable report rejects an entirely skipped module inventory and lists fully skipped modules in its summary. A mixed passed/skipped inventory is allowed: Vitest's skip state does not establish whether a skipped module was required on that runner. Review the listed skips against platform and provider prerequisites; the inventory alone does not prove every required case ran.

Scheduled qualification is admitted hourly at minute 23. A forced full run at 03:23 UTC refreshes runner and dependency drift every day. Other hourly runs can reuse a successful full-qualification receipt for the exact source SHA from the same repository's trusted CI workflow on its default branch. Both the receipt and original workflow run must be less than 24 hours old, so retrying an old gate cannot renew stale proof. Missing, expired, failed, foreign, or uncertain proof runs full validation; API lookup failure also runs full. Lookups examine at most 100 named receipts and three candidate runs within a 20-second budget. Reused runs select no validation lanes, identify the original run, and produce no new receipt, so reuse cannot indefinitely extend proof freshness. They make no fresh provider observation. Manual dispatch, ci:full, and exact release qualification always run full validation.

Full qualification includes aggregate coverage, every portable provider lane, and all native hosts below. Adding ci:full immediately starts the separate Full PR qualification workflow against the label event's merge revision; subsequent PR revisions also select the full tier in normal CI while that label remains. Unrelated labels do not start validation jobs or replace the main CI checks. Manual dispatch also requests full validation. Fast-tier broad fallbacks keep the portable suite and the Linux x64 runtime baseline. They do not start every owner lane or the three-host package matrix. Intel macOS and coverage remain full-tier obligations. Real provider lanes stay path-gated, and a shared registration file listed by many lanes does not start those lanes on a fast PR. Each classification job records its selection and broad-fallback reasons in the run summary. Representative Git-diff cases live in tests/boundary/filesystem/ciChangeScope.test.ts.

CI required checks classification and every selected lane, rejecting failed, cancelled, or unexpectedly skipped work. It rejects missing or unknown tiers and requires every canonical scope to be explicitly true or false; incomplete classification cannot silently omit verification lanes. The aggregate remains cancellable so superseded runs release their slots. Branch protection should require this aggregate, Static checks, Portable tests, and Conventional PR title. Title validation runs separately on PR opening, reopening, commits, and edits; correcting a title refreshes only this lightweight check. It reads title text through an environment variable, executes no PR code, and has no token permissions. Its separate concurrency group cannot cancel test runs. Fast PR runs cancel superseded feedback for the same PR. Main push runs keep one active run and the latest pending run for that branch. Scheduled/manual qualification has a separate concurrency group: active work finishes and only the latest pending run is retained. Release qualification uses its exact source SHA and cannot be cancelled by newer main pushes. Provider and website lanes inherit their caller's scheduling; reusable workflows have no independent cancellation group. Direct manual lane dispatches run independently on disposable hosted runners. Publication requires that qualification to succeed and consumes its exact verified tarball, checking source SHA, workflow run, run attempt, package/Node version and SHA-256 before npm publication. Earlier attempts from the same workflow run are accepted when rerunning failed publisher jobs; future attempts and other runs are rejected. The public-registry canary still verifies the published package. Licensed/self-hosted provider workflows and the Android smoke test retain their explicit triggers.

CI exercises the pinned Node.js runtime on native hosted runners:

Host	Runner	Baseline checks
Linux x64	ubuntu-latest	Installed package and real Node Inspector CLI/MCP
Linux arm64 (aarch64)	ubuntu-24.04-arm	Installed package and real Node Inspector CLI/MCP
macOS 15 arm64	macos-15	Installed package and real Node Inspector CLI/MCP
macOS 15 x64	macos-15-intel	Installed package and real Node Inspector CLI/MCP (hourly, dispatch, release and ci:full)
Windows x64	windows-latest	Curated capabilities, native controls, installed package and real Node Inspector CLI/MCP

macOS x64 is the slowest package host. Ordinary PRs and main pushes omit it, so darwin x64 installed-package and Inspector regressions surface in full qualification. Schedule admission can be delayed by GitHub; the hourly trigger does not guarantee an hourly completed result. Release publication waits for its own complete result at the exact released revision.

Package lanes assert the actual Node platform/architecture before verification and record those values with the Node version. The package matrix runs up to four jobs concurrently on independent hosted runners with explicit timeouts and per-runner Node heap/thread limits; Windows retains its curated capability lane. Package checks cover installation, CLI/MCP discovery, target-free analysis, configuration backups/recovery, Evidence and owned lifecycle; Inspector checks execute source-owned loopback targets and special filename cases.

POSIX package verification resolves the runner's configured npm cache before creating disposable client homes and passes that location to its child installs. This reuses the download cache restored by the shared npm-cache action while keeping client configuration isolated. Installs retain their existing registry and integrity checks.

The Apple artifact CLI/MCP lane uses one macos-15 job to avoid adding demand for scarce macOS runner capacity. It runs the artifact-format and metadata verifiers, followed by one consolidated focused invocation containing all existing native and process-boundary regression paths. The real native provenance verifier checks CLI/MCP agreement on complete per-slice encryption absence. The encryption CLI regression builds a universal image, marks its owned arm64 command with cryptid=1, and compares the reported range with otool; this verifies declared metadata, not actual FairPlay data.

The Linux source-test shards download a shared runtime and generated test metadata snapshot produced by the build job with npm run test:prepare. The snapshot includes dist/, generated skills, the MCP tool catalog, product catalog, and managed-conformance metadata. Each shard still installs the locked dependencies and runs its assigned tests; full qualification enables coverage. It no longer regenerates the same metadata independently.

The build job finds the newest trusted timing artifact for its tier (test-timings-fast or test-timings-full) from a successful default-branch push, schedule, or manual CI run, and includes that same report in each shard's snapshot. The bounded lookup verifies the producer. Pull-request runs and other branches are not producers. A tier with no qualifying artifact leaves history unset, and Vitest uses its own assignment. A later schedule that reuses qualification without uploading timings leaves the previous artifact in place.

On 2026-10-21 the newest full history was run 39545047294, recorded before compiled-CLI session reuse. Fast run 39552964873 packed its eight built shards from that history, and those shard jobs executed for 319–591 seconds. Replaying that run's file durations through the sequencer's two-worker schedule balances every shard at about 404 seconds of file time. The same replay with the stale history weights leaves the slowest shard near 547 seconds and the fastest near 293. That spread is file-schedule time, not runner queue, which is why each tier keeps its own history. CI assigns discovered files by historical duration, with a deterministic fallback weight for new files, rather than dividing only by file count. Each file keeps its full duration on one worker. Extra workers overlap other files in the same sequence group, and process-boundary runs after that group. History published before the competing-loader split still names only javascriptLiteralPercentPaths.test.ts. Until an artifact records javascriptLiteralLoaderOwnership.test.ts, the sequencer assigns that recorded duration by the pre-split case counts: 54/192 stays with the percent-path file and 138/192 goes to the consolidated loader file. History that names the retired Early, Middle, and Late shards sums those durations onto the consolidated file instead, and scales a partial set as three equal shares. That weight is the old dual-surface cost. The demoted file runs every loader once through the shared stdio MCP session, so the first schedule overestimates it. Every discovered file still belongs to exactly one shard; worker limits and project ordering remain unchanged. Fast qualification keeps eight built shards without coverage. The merge job rejects successful reports with missing, unexpected, or duplicate portable files. Coverage instrumentation and aggregate thresholds remain enforced in full qualification. Missing or unusable history falls back to Vitest's default assignment. Only timing data is reused; live provider results are never cached. Turbo outputs are restored by task hash, with trusted default-branch runs as cache writers and PRs as restore-only consumers. The static job also restores its Turbo snapshot; each task revalidates its own input hash before reusing outputs or diagnostics. The Node pin and npm install settings are global hash inputs. Knip runs only in full qualification, with --cache and a GitHub cache of node_modules/.cache/knip. The saved key includes the commit SHA. Restore uses the configuration prefix, so the next trusted full run can publish a refreshed snapshot. That cache is an AST cache keyed by file mtime and size, so a fresh checkout often misses it; Turbo still hashes the Knip task inputs. Knip includes its ignore rules and VitePress entry, so changes to those inputs cannot reuse an older successful check. Jobs that run Turbo export TURBO_TEAM. A called workflow sees the caller event, so workflow_call is not an event name inside CI. TURBO_TOKEN is set when release qualification passes remote_cache_write, or when the ref is the default branch and the event is push, schedule, or workflow_dispatch, and only when that secret is available to the run. Release qualification passes TURBO_TOKEN and remote_cache_write into the called workflow. Labeled pull-request qualification does not. Pull requests receive an empty token and TURBO_CACHE=local:rw. Remote reads require TURBO_TOKEN, and this repository does not pass a separate read-only credential into pull requests. An empty token leaves the local Turbo cache in place.

The shared npm-cache action caches downloads, not node_modules, and every consumer still runs npm ci. Keys cover the runner OS, actual Node architecture, root lockfiles, and .npmrc. An exact-key miss restores the newest cache for that OS and architecture, so a lockfile change reuses unchanged tarballs while npm ci still verifies the lockfile. Only push, schedule, and manual events on the repository's default branch may save shared npm caches, and only from the opted-in writer for that host: Static checks for Linux x64, the package matrix for every other included host, and the Windows curated lane. Every other job restores without saving, so parallel jobs do not publish duplicate copies. Pull requests restore only. Static snapshots are saved only after the static job's validation succeeds. Cache deletion is a separate workflow, because reusable qualification calls this workflow and cannot raise its read-only token. After a push, schedule, or manual CI run on the default branch finishes, including when a later lane fails, times out, or is cancelled, that workflow runs on ubuntu-latest and deletes older rea-static-v1, rea-static-deep-v1, and rea-portable-build-v2 entries for that runner's OS and architecture, plus older knip-meta-v1 entries for that OS. It keeps two generations plus the key that run saved. A fast run that did not publish a deep or Knip snapshot still leaves the two newest entries in those families. It deletes an npm download record only when another record has the same key, cache version, and git ref, and it keeps the most recently accessed copy. A different version or ref is a different restore target and stays. One cache id is one record: a repeated id keeps the later access time when its key, version, ref, size, and creation time agree, and an inconsistent repeat deletes nothing. An incomplete or unexpected cache list deletes nothing. Distinct lockfile generations of the download cache stay, so the restore prefix still has a seed. Main warms the new cache namespace after merge; a cold cache falls back to the normal install. This prevents each PR from saving another branch-scoped copy of the same download cache. It does not remove runner assignment waits.

Portable owner-based test selection remains observational. Its reports compare proposed omissions with the full run; incomplete or skipped observations do not establish that fewer tests would preserve regression detection.

The automatic browser, binary-layout, EVM, JavaScript-recovery, packet-capture, signal-recording, recorded-crash, web and environment-bundle provider lanes download rea-build-cache from their own run after the build and static jobs succeed. Their existing build commands still validate task hashes and rebuild mismatched inputs; provider observations always run. Provider-workflow-only edits also admit the build without enabling source test shards. Direct provider dispatches retain the cross-run restore-only cache. This removes repeated compilation on fresh source revisions, but adds the build as a prerequisite: a delayed build can delay provider admission. Compare queue and execution times before claiming an end-to-end speedup.

These native baseline checks complement the Linux source-test shards and the separate Apple-artifact and real-provider lanes. Actual Hopper, Ghidra, IDA, browser and managed-tool claims require their corresponding verification lanes. Runner labels follow the GitHub hosted-runner reference.

Termux Android browser smoke test

The Real Termux browser verification workflow is manual-only while its emulator setup is being validated. After the workflow reaches the default branch, select a revision under Actions → Real Termux browser verification → Run workflow. It boots an Android 11 (API 30) x86_64 emulator, installs the checksum-verified Termux 0.118.3 APK, and lets the app initialize its bootstrap. The host stages the selected revision using git archive HEAD; no workstation build outputs or dependencies are copied into Android.

scripts/verify/termux/emulator.sh sends a RUN_COMMAND intent to Termux's app service. Root access is limited to staging files, sending the intent, and reading diagnostics; installation, compilation, Node, and Chromium run as Termux's app UID. The in-app script installs Termux's Android-linked Node distribution, checks its version against .nvmrc, pins npm from packageManager, runs npm ci, and builds with build:termux. Termux repository packages, including Chromium, are resolved at run time; their installed versions are logged.

The verifier leaves Playwright's cache environment overrides unset, requires a healthy public MCP doctor result, and captures URL plus DOM from a loopback HTTP fixture through capture-browser-scenario. It checks completed steps, the exact URL, a DOM marker, and reported browser cleanup. This lane covers one Android emulator and Termux combination, not physical ARM devices, Electron, or the complete desktop browser suite. A workflow definition alone does not establish passing Android coverage; inspect its execution receipt.

The termux-browser-diagnostics artifact retains the Termux command log, exit status, and Android logcat on success or failure. The emulator is disposable.

Packet capture analysis

npm run verify:pcap -- --pcap PATH exercises caller-supplied tshark and capinfos against a saved capture. Linux verification uses Wireshark 4.6.6 and captures containing DNS, TLS and HTTP traffic. Native Windows and macOS engine behavior remains unverified; process fixtures do not establish it. See Packet capture analysis for contracts and budgets.

Apple native metadata and UI

tests/boundary/mcp/nativeDispatchBoundTarget.test.ts uses the production stdio MCP server on macOS to bind an inert, self-authored Mach-O, then replace its path with missing, directory and symlink-loop states. Each refusal must report artifact_changed; restoring the same bytes must succeed in the same session. A separate non-root case checks real read permission denial as access_denied. These tests read the target without executing it. Controlled provider tests retain the physical acquisition failure through descriptor close denial and confirm cleanup retry releases the original owner.

npm run verify:apple-dispatch compiles the Objective-C fixture (classes, protocols, a property and an NSString category) and the Swift conformance/vtable fixture.

• Each fixture is linked with legacy LC_DYLD_INFO binds and with chained fixups; on Apple silicon the ObjC fixture is also built as arm64e, which uses authenticated pointers.
• The lane inspects the bytes and repeats after stripping local symbols.
• It requires the bound NSObject superclass, the external category, and the matching pointer_fixups coverage. It requires macOS and the host Xcode toolchain; targets are not executed. npm run verify:native-ui launches exactly one source-owned fixture window and requires successful selected-window capture and selected actions. An OS permission denial fails the positive lane. npm run verify:native-ui:permissions allows a host-permission-boundary-only result and explicitly reports positive_e2e: false; it must not be reported as capture/action proof. Both commands reject a changed executable digest and clean up the fixture process and helper. These lanes require an interactive macOS desktop. See native investigation.

npm run verify:native-calls needs macOS with Command Line Tools (clang, lldb, codesign, nm) and actual debugger access to the owned fixtures. The lane reports Developer Mode status without treating it as proof of access or denial, and does not change host settings. LLDB and target entitlements establish the tested permission boundary. It compiles tests/conformance/native/calls.m and runs observe-native-calls through the CLI and stdio MCP. It checks:

• the receiver class, selector and argument registers of every entry, and that breakpoint addresses equal nm's symbol addresses;
• overlapping breakpoint selections retain each selection's hits and share the aggregate event limit; callback capture reads the stopped frame before LLDB resumes, avoiding stale frame metadata from asynchronous stop events;
• captured stdout and an environment override, plus a 2 MiB flood on each output stream with bounded retained prefixes and exact drained-byte counts;
• the event-limit and duration outcomes, with the process confirmed gone;
• that a hardened-runtime copy is refused with debugger-attach-denied, and that the same copy signed with get-task-allow is traced.

Firmware adapters

Linux-owned provider launches require a procps-compatible ps on REA's PATH so ownership can be inspected before launching a child. A missing or incompatible command fails with an actionable capability error before the launch.

npm run fixtures:firmware uses existing Python 3 and a host C compiler to make an ignored gzip/USTAR firmware fixture and independent offset/hash oracle. npm run verify:firmware requires caller-supplied Binwalk 3.1.0, Unblob 26.6.4 and util-linux prlimit on Linux, with 7z on PATH for the gzip/USTAR fixture's Unblob extractor. The provider also accepts other 3.1.x and 26.6.x builds and reports them as unverified; this lane proves the audited releases. It verifies CLI/MCP parity, selected ranges, unknown chunks, depth limits and extracted child digests. The optional REA_FIRMWARE_VERIFY_EXT4=1 lane requires existing mke2fs/debugfs; the separate REA_FIRMWARE_VERIFY_GHIDRA=1 lane checks a selected host ELF through real Ghidra. Neither optional toolchain is a base-lane prerequisite. See firmware analysis for limits and unverified formats.

npm run verify:esp-images requires a Clang with the RISC-V target and ld.lld (REA_RISCV_CLANG, REA_RISCV_LLD) and a Python that runs esptool and espsecure (REA_ESPTOOL_PYTHON). It builds ESP32-C3 app and bootloader ELF files from tests/conformance/esp/, converts them with esptool elf2image, including --secure-pad-v2, --ram-only-header and --pad-to-size, signs one with RSA, ECDSA P-256 and P-384 keys, and compares REA's facts with esptool image-info and espsecure signature-info-v2. It compares extracted segments and signature blocks with the producer's bytes, and derives malformed images and ESP8266 v1 and v3 images. It checks each tool before building and warns when esptool is not the pinned v5.6.0. REA_XTENSA_ESP32_GCC adds an ESP32 Xtensa build. See ESP images.

npm run verify:uf2 needs only a Python that runs esptool, named by REA_ESPTOOL_PYTHON (default python3). It writes inputs at known addresses, builds UF2 files with esptool's merge-bin --format uf2 (v4 merge_bin) with and without MD5 and with 256-byte payloads, and concatenates two chips' outputs. Inspection and extraction through the built CLI must return each input at its address under the familyID esptool itself reports. It is not part of CI. See UF2 and DfuSe images. npm run verify:esp-flash needs the same tools and an ESP-IDF checkout (REA_ESP_IDF_PATH) for gen_esp32part.py and otatool.py. It builds partition tables, assembles dumps with esptool merge-bin, writes OTA records with otatool through tests/conformance/esp/flash_file_tool.py, and compares REA's flash facts with image-info, the decoded table and otatool read_otadata. See ESP flash images.

JavaScript source recovery

Build once with npm run build:cached, then run npm run verify:javascript:recovery. This focused lane requires Linux x64, util-linux prlimit, REA_WAKARU_COMMAND pointing to the official Wakaru 1.14.0 Linux x64 binary, and REA_JAVASCRIPT_FIXTURE_TOOLS pointing to an isolated npm prefix containing esbuild 0.25.10 and webpack 5.101.3. No global installation is required. The lane compiles source-owned fixtures, exercises CLI and stdio MCP, verifies published bytes and provenance, feeds recovered modules into existing analysis, and compares a finite set of known fixture results. A webpack fixture with separate runtime, shared, entry and lazy chunks verifies multi-input recovery through both surfaces, per-file provenance and resolved cross-chunk imports in the downstream application graph. The provider also accepts other ^1.13.0 releases and reports them as unverified; this lane proves the audited release. It does not establish arbitrary recovered-application equivalence. CI installs these prerequisites only in .github/workflows/real-javascript-recovery.yml; the existing real-browser lane uses real Chrome for browser capture and website workflows.

Captured website source-map lane

npm run verify:browser:source-maps checks actual Chromium capture/export and source-map point tracing through CLI and stdio MCP. It requires absolute REA_BROWSER_EXECUTABLE and REA_WEB_SOURCE_MAP_COMPILER pointing to esbuild 0.25.10's lib/main.js in a caller-owned isolated installation. Preflight reports missing prerequisites for this lane. The compiler is used only to generate the source-owned fixture. No Hopper, Ghidra or application dependency installation is required.

REA_BROWSER_EXECUTABLE=/absolute/path/to/chromium \
REA_WEB_SOURCE_MAP_COMPILER=/absolute/path/to/fixture-tools/node_modules/esbuild/lib/main.js \
npm run verify:browser:source-maps

scripts/verify-browser-source-maps.mjs /absolute/path/to/installed/rea.mjs checks an installed package after building the verifier dependencies. The separate conditional real-web-source-map CI job supplies Chrome and an isolated pinned fixture compiler; static/unit checks do not acquire a browser. See the source location guide for the verified decoder profile.

JavaScript large-output lane

npm run verify:javascript:output exercises the CLI JSON result surface beyond the running Node engine's single-string limit. Shared input leaves keep the fixture's graph small; the verifier writes one temporary output file, checks its complete byte count and an independent digest, then removes it. It requires only Node and the built REA runtime, with space for the output plus a 1 GiB reserve. Run npm run verify:javascript:output -- jsonl for compact JSONL coverage. Its row collection must match independent row and final-metadata byte/digest goldens; its token count is checked against the actual file one line at a time, even when the complete export exceeds the engine string limit. This opt-in lane is separate from routine tests and the canonical hash check npm run verify:javascript:digests. It verifies serialization rather than an arbitrary third-party application's parsing cost or MCP client capacity.

Website runtime attribution lane

npm run verify:browser:runtime uses caller-supplied REA_BROWSER_EXECUTABLE and an owned synthetic site/profile. It exercises public CLI and stdio MCP for precise execution and native listener source locations, including actual armed progress, Unicode/CRLF digests, repeated source URLs with distinct script IDs, zero branches and function-only unknowns on repeated coverage, request initiators and an externally owned page that remains open.

An optional entrypoint argument to scripts/verify-browser-runtime.mjs runs the same checks through an isolated installed package. The conditional real-web-runtime CI job runs only for relevant changes and needs no fixture compiler. Ordinary unit/static gates acquire no browser. See website runtime attribution for effects, resource bounds and coverage limits.

Offline binary layout

npm run verify:binary:layout requires Linux x64, GCC/binutils, the aarch64-linux-gnu, i686-linux-gnu, arm-linux-gnueabi, mips-linux-gnu and mipsel-linux-gnu cross gcc/readelf toolchains, absolute REA_PWNTOOLS_PYTHON with pwntools 4.15.0/pyelftools 0.33/Unicorn 2.1.2 and absolute REA_VERIFY_STRACE_COMMAND. It compiles ephemeral source-owned ELF fixtures and checks public CLI/MCP, lossless addresses/names, file ranges, mitigation inferences, malformed/unsupported input, original file hashes and released process ownership. Cross-compiled ELF64 AArch64 and ELF32 executables and relocatable objects check the reported class, machine and byte order, and every section name, address, offset and size against GNU readelf. Exec syscall tracing must identify only the declared Node/Python launchers; no target binary is executed. Core/debugger claims need separate verification lanes. Pass an installed package entrypoint as the script's first argument to verify packaging independently of the checkout. The valid SHN_XINDEX fixture has 65,281 full section rows; CLI is checked in the ordinary lane. Its large MCP transfer is opt-in with REA_VERIFY_LARGE_ELF_MCP=1 (or the workflow dispatch large_mcp input), an explicit 256 MiB SDK receive buffer, matching server REA_MCP_MAX_RESPONSE_BYTES, and five-minute request timeout. Ordinary MCP fixtures retain the pinned SDK defaults.

Offline EVM interface

npm run verify:evm:interface requires Linux x64, an absolute REA_VERIFY_STRACE_COMMAND, caller-supplied util-linux prlimit and REA_VERIFY_SOLC_MODULE selecting the absolute module path for solc 0.8.30. It compiles source-owned plain/optimized/via-IR Cancun fixtures in private storage and checks actual CLI/MCP selector evidence, raw/hex identity, unknowns, malformed carriers and independent cleanup. An optional positional entrypoint verifies a fresh installed package. It acquires no engine, compiler or chain dependency and does not execute a contract on a chain.

Recorded crash evidence

npm run verify:recorded:crash is a separate Linux x64 lane. It requires GCC, GDB, absolute REA_PWNTOOLS_PYTHON with the offline ELF profile above, REA_PWNDBG_GDBINIT and REA_PWNDBG_VENV_PATH with unchanged pwndbg 2026.09.15, and REA_VERIFY_STRACE_COMMAND. Its disposable CI runner installs GDB, checks out the exact upstream commit and installs its frozen lockfile in isolated runner storage. No developer host configuration or core-pattern setting changes.

Fixture generation explicitly runs an owned source-built two-thread program under GDB to create a recording. Subsequent public CLI/MCP inspection verifies lossless high registers, signed signals, note source bytes, malformed/missing notes, unfamiliar owners, optional core-only mapping context and actionable missing/unsupported plugin errors. A historical-PID collision fixture references an owned live sentinel; inspection syscall traces reject process attach/memory access, provider lookups of that PID's /proc files and attempted Internet sockets. Traces admit the observed upstream startup helpers (iconv -l, the selected checkout's Git version lookup) and REA ownership inspection separately from target execution. This is fixture evidence, not a sandbox claim. Inputs remain unchanged and the sentinel must stay alive; owned cleanup and empty verifier descendants are required. Pass an installed package entrypoint as the script's first argument for package coverage.

Exported environment bundles

npm run verify:agentenv:environment is the real-agentenv lane. It requires the host's Docker CLI named by absolute REA_DOCKER_COMMAND, Python 3.12 with agentenv-framework==0.9.1334 installed in a virtual environment whose agent-env binary is named by absolute REA_AGENT_ENV_COMMAND, and outbound network for pip on the runner and for npm inside the image build. Both variables are required; neither upstream executable is resolved from PATH, and the verifier installs nothing. Packing uses the shared npm helper, honoring npm_execpath from the caller's selected npm. Direct Node invocations without that value use the helper's platform fallback, so configure the pinned toolchain on PATH for those runs. It checks the supplied runner's --version against the pin before using Docker and retains the observed version in both passing and failing receipts. Its disposable CI runner creates the virtual environment in runner storage, exports both paths and uses the runner's own Docker; no developer host configuration changes. agent-env run attempts share one 22-minute budget inside the 30-minute job (the first attempt at most 15 minutes), so a hung or retried run ends with the verifier's failed receipt rather than the runner's kill. The framework builds its gateway and service-db images from public.ecr.aws on every fresh runner, and hosted runners share that registry's anonymous pull quota, so the lane first pulls the base images the pinned framework's Dockerfiles name from mirror.gcr.io, which serves the same Docker official images under the same digests, and names them as those Dockerfiles do (on Docker's containerd image store, so digest-pinned names resolve), as the framework's own CI does. The quota handling below covers anything still pulled from that registry. The receipt's agent_env_run records the last run whether it passed or failed, with public_ecr_quota set to data_limit or rate when that registry's 429 Too Many Requests ended it. The data-transfer cap (toomanyrequests: Data limit exceeded) is counted per source IP, so it fails the lane at once: a pause on the same runner cannot lift it. When a rate refusal ends the run and at least five minutes of the budget would remain after the pause, the verifier waits 90 seconds and runs once more, recording attempts: 2 and retried_for in the receipt. Any other failure ends the lane at once.

The verifier writes the Grafana-shaped HAR the view tests own, packs the checkout with npm pack, builds the bundle from the HAR with rea environment-build --package-tarball through the built CLI, the command users run, so the image installs this checkout rather than the registry package, and runs agent-env run on the bundle. It checks that the build reports the 28-record capture, the default template rules, the planned file set with the shipped manifest naming the reported capture, a coverage summary whose read and served write tools add up to the served tool count, and a bundle served from the derived state, which serves fewer tools than the manifest because write tools the capture did not confirm are not served, that the runner exits 0 with tasks/smoke.json reported as passed and the verifier's description naming that served tool count, and that no container exists after the run that did not exist before; the container inventory runs even when the run fails or times out, so a failed receipt still lists what it left. The one exception is the framework's own image-store helper, agentenv-registry (or agentenv-registry-<port>), a registry:2 container the local provider starts on first use and keeps running across runs; a first run on a fresh host creates it, so the receipt lists it under image_store_helpers instead of counting it as a leak. Images built by the run remain and are counted in the receipt. Pass an installed package entrypoint as the script's first argument for package coverage.

Agent evaluation and conformance records

npm run verify:agent evaluates investigation scenarios through a real local Codex CLI. The lane runs sequentially; select scenarios with REA_AGENT_EVAL_SCENARIOS and choose the installed client/model with REA_CODEX_CLI and REA_AGENT_EVAL_MODEL. The detailed agent evaluation reference contains isolation, authentication cleanup, grading, and conformance boundaries.

The Notes evaluation packages website/public/examples/notes-electron/. When changing those example bytes, run tests/boundary/mcp/agentKnownAnswerMcp.test.ts and check its rubric in tests/evaluation/model-evals/AgentEvaluationFixtures.ts. Preserve exact source facts and the coverage reported by the analysis. If a source change crosses an unresolved static-analysis boundary, keep that uncertainty and a negative control rejecting claims of complete coverage.

Frida instrumentation

npm run verify:frida builds REA, spawns a short-lived local Node.js fixture, loads a harmless marker script before resuming it, captures the message, and detaches. This lane requires Linux on the REA host and the optional Frida binding; it does not need a Frida server or a caller-selected process. To run the separate remote attach lane, pass -- --address HOST:PORT --pid PID for an explicitly selected process. Optional remote flags are --token, --certificate, --origin, and --keepalive-interval (seconds); authentication values are not written to files or included in the report. Mock tests do not establish real device/server compatibility.

Offline WASM artifacts

npm run verify:wasm:artifact -- --require-tools uses an absolute REA_WABT_BIN_DIRECTORY containing WABT 1.0.42, including wat2wasm for fixture generation. It installs nothing. Without configuration, the default invocation reports a named skip; --require-tools makes missing configuration fail.

The verifier creates and validates real modules with custom sections, multiple import/export kinds, escaped names, an empty module, and a bulk-memory DataCount section. Debug-name fixtures cover compiler-generated module names and preserved name-section payloads with empty, multiline, Unicode, long and NUL-containing names. It compares exact WAT, section output and selected/tool digests through the compiled CLI and real MCP SDK, validates advertised schemas, checks identical CLI/MCP Evidence IDs and session bundle readback, retains invalid/truncated module diagnostics, preserves all inputs and checks workspace cleanup. The lane also injects one cleanup denial for exact private runtime roots and input descriptors, checks recovered CLI errors, and verifies retry on the same MCP connection after repeated runtime-removal denial. These are controlled faults, not naturally occurring permission failures; no WASM target is executed. Pass an installed package's scripts/rea.mjs as the first argument to run the same lane against the package. npm run verify:wasm:package packs and installs REA into an isolated temporary prefix with lifecycle scripts disabled and runs the same required real-tool lane. It requires WABT configuration and makes no host tool installation.

Source-owned provider tests separately exercise cancellation, deadline propagation, malformed producer output and cleanup uncertainty; shared owned-command tests exercise actual subprocess timeout and cancellation.