swayfx-mosaic/tooling/probes/step20c-static-e2-completion

Static E2 completion fixture

EXPERIMENTAL / PREPARATION; runtime NOT RUN, review PENDING. Ticket SWAYFX_MOSAIC_STEP20C_E2_COMPLETION_AND_PRIVATE_INTEGRATION_BATCH_V1. No production source, UX, packaging or cumulative patch change. STEP20C stays NEXT; production multi-output is NOT YET SUPPORTED.

derivation.json identifies initial exact copies from the frozen step20c-static-nested-geometry-replay fixture at c5f96a34e3924275179365938d02f3a318c0c4bf. Model, IPC adapters, observations, pixels, predicates and core geometry/protocol/terminal tests are retained. reusable_tests.py preserves the identified AST method bodies. The historical readiness aggregate and its ACCEPTED review grant no acceptance here.

transition.py is shared by raw historical replay and the new E2-only runner. It examines the entire selected H1 callback window: old applied Mosaic, targeted accepted splith, first coherent applied standard, then at least two coherent standard observations. It verifies source-backed current layout, actual owner scene, exact workspace/output/XDG identities, unique enabled root, clip removal, capture intervals, and no reactivation through standard B input. PI_POPUP lacks owner IDs and reports the global popup root on both outputs. Unique live XDG hierarchy plus exact IPC ancestry supplies bounded correlation; independent observations do not establish an atomic transaction generation.

Historical request-start time was not recorded. replay.py uses the previous actual capture event and first monotonic trace after actual command handling as an explicit request interval. It preserves the old applied root after handling and before reply, followed by two cleared roots; no row is discarded. The historical cleanup was compositor shutdown, corroborated by seat capabilities0, then destruction. It does not qualify normal owner close. Replay cuts only at that explicit shutdown boundary, retaining any earlier dismissal/destruction.

The new runner records exact IPC request start and actual reply separately. Its transition window starts at a complete source-backed Mosaic commit after A/B are fully mapped and before the request; earlier pre-root callbacks stay in separate opening evidence. Absence of an eligible callback is INCONCLUSIVE. Only E2 runs: real left TOPLEVEL opens A, real left A opens B, right B in standard, explicit client close B, then a fresh measured A right click. Each point comes from a measured solid interior with an8px margin. Every click records fresh client protocol deltas, full button pairs and wire/callback coordinates. Configured/applied destinations remain distinct from native420x320 SHM; an old Mosaic observation is not reparsed with a later standard configure. Client close requests are recorded before flushing stdin. B destruction and COMMAND close_b DONE must match exact objects, without popup_done. A preserves map/IDs, exact pixels on both outputs, zero B colors and input. Normal owner close must be corroborated by command, child-first destruction and wire destroy requests. Final whole H2 RGB equals the initial #303038 reference without mask/tolerance; final H1 has no popup colors. Raw terminal IPC/compositor/client codes and rescue are retained separately. Terminal IPC1 alone does not undo completed science.

Budget: run1, chain1, grabs2, complete clicks4, camera1, transition1, PNG12, positions16. Maximum150s; active work watchdog cuts at143s to leave cleanup margin. Two full H2 references are identical. Runtime is0700, private two-output headless H1(-800,0)/H2(0,0), 800x600 scale1 normal, with explicit private swaymsg socket. No pristine control, compositor/SceneFX build or desktop operation.

Offline preparation

All outputs below are in a new directory outside Git. Commands never reserve a real runtime budget, launch a client/compositor or change historical LAB bytes.

F=tooling/probes/step20c-static-e2-completion
O=/home/fy59/swayfx-mosaic-e2-batch-hyrd8qm1/implementation-final-offline
mkdir "$O"
PYTHONDONTWRITEBYTECODE=1 python "$F/tests_offline.py" --output "$O/test-receipts.json" > "$O/tests.log" 2>&1
PYTHONDONTWRITEBYTECODE=1 python "$F/replay.py" --output "$O/replay.json"
PYTHONDONTWRITEBYTECODE=1 python "$F/audit.py" "$O/source-audit.json"
PYTHONDONTWRITEBYTECODE=1 python "$F/gate.py" "$O"

The gate requires nonempty exact test inventory, actual executed PASS receipts, every temporal negative rejection, final source hashes, raw replay equality and an exhaustive current static assertion inventory. Audit PASS means classified source inventory; graphical execution coverage remains NOT TESTED. Negative handoff tests simulate records only in temporary isolated directories; these are explicitly synthetic and never mint a real independent review.

handoff.py prepare "$O" creates the sole fresh registered LAB, exact source manifest and prospective handoff. It reuses the historical exact C17 clients and their actual adapted static_cross/protocol sources byte-for-byte; no build or client execution is required. Static binary identity remains f82557c1a9978a5e9665b9b4291b74880253725359e3df1c5b8d554921f7603b, build ID942db64269640778ee433e5970ba1100bc60f94c, mapped wlroots identity 7edca9a7d65855addda68e29ba5db94d57c803eb5fa7d0f86db3136b90ca9837. The original source directory is part of the frozen runtime manifest, alongside installed fixture, binaries/clients, actual private trace source and receipts.

Independent review before dispatch

The implementation author does not generate acceptance. A genuine independent read-only reviewer must produce a report and results/handoff-review.json with: status ACCEPTED, independent=true, read_only=true, scope STATIC_E2_COMPLETION_FINAL_BYTES, blockers=[], fixture_sha256, source_sha256 (all final source files), runtime_handoff_sha256, authorization_sha256, registration_sha256, gate_sha256, report_file and its report_sha256. These bind the actual final files, not a historical gate.

After parent approval, handoff.py preflight LAB publishes an exact preflight. Only then may the parent run launch.py LAB once. Exclusive O_EXCL consumed token precedes dispatch. Runner repeats reviewed source preflight and checks the exact consumed token before its lock or run debit. Wrong/old LAB, token, binary, wlroots, stale review, changed manifest/runner, existing run/socket and prior consumption fail closed. Failure never refunds the ticket; no automatic retry. No runtime dispatch has been performed during implementation.