# Changelog

## Unreleased

### STEP20C experimental evidence

- Local offline nested geometry/fixture review protocol: separate SHM, assigned/rendered, scene, XDG and PNG spaces; exhaustive assertion/source/test register and independent review required. No graphical launch authorized; final offline verdict INCONCLUSIVE.

- Preserved static nested-grab fixture failure: sole unchanged-binary run confirms identity/output/camera/tiled clip and exact whole-H2 pre-popup references, then fails before any click/grab on a native-SHM versus painted-height assertion. INCONCLUSIVE / UNDECIDED; additional IPC-y/titlebar assumption invalid, offline assertion audit insufficient. Raw runner exit1 and compositor exit0 retained; popup/standard gates NOT TESTED, no retry/build/production change. Historical runner and sealed artifacts preserved.

- Preserved independent static RGB/input/standard retest: corrected shared predicate and real replay pass; exact whole-H2 RGB, H1 retained/H2 forbidden pixels, real target/reference clicks and targeted splith clip removal/input pass. FEASIBLE for this matrix. Terminal exit IPC makes runner exit 1 after measured gates and compositor exit 0; raw failure retained and offline diagnosis passes. No new compositor/library build or production change.

- Preserved static focus/input/standard fixture failure: exact static binary and loaded pair verified; one run stopped before captures/input on an incorrect border-width fixture assertion. INCONCLUSIVE / UNDECIDED; camera and clip traces retained, RGB/clicks/splith NOT TESTED, no retry or production change.

- Approved private static candidate decisions (2026-10-13): separate notices, project maintenance and exact-pair qualification. Current package unchanged; private real `/usr` static build/model/staging/library identity/headless startup pass. Whole-H2 reference fixture failed after a focus change; input and splith return NOT TESTED. PARTIALLY_FEASIBLE / UNDECIDED, preserved historical evidence; the original manifests and fixture bytes remain unchanged.

- Preserved static distribution contract: one hash-qualified static executable staged through private Meson metadata at /usr/bin/swayfx-mosaic, exact-one-file shape PASS without recompilation. Existing scene interposition is characterized, version/security/notice obligations and an unsubmitted upstream API dossier are explicit. PARTIALLY_FEASIBLE, contract decision required; no recipe, packaging contract, production or system change. Historical verdict retained; later user decisions authorize a separate private gate only.

- Preserved SceneFX/wlroots ABI/package audit: current stock build and exact-one-file staging PASS; experimental shared clip functions absent from stock despite common SONAME. Private static build/loader passes, but wlroots interposition, notices, security and version-maintenance gates remain. PARTIALLY_FEASIBLE; architectural decision required. No integration or system change; the historical ABI evidence is published separately from any later distribution-contract investigation.

- Preserved optimized-edge gate: four completed controls reproduce the same boundary differences in pristine SceneFX, private splith and Mosaic without OPT traces. Same-binary Mosaic trace-off/on and pristine/private standard match full RGB at six states; central freshness, H2 stability and sampled clipping hold. PARTIALLY_FEASIBLE, architecture UNDECIDED: reduced-domain FBO history has source/model support but no native causal observation. Separate ABI/package audit justified; optional bounded GPU study remains unexecuted. No production change; historical graphical campaign closed.

- Preserved optimized blur cache probe: private splith/Mosaic used a real background source and output-scoped GPU cache, naturally reconstructed A/B/A and restored exact central filtered pixels with stable whole H2 RGB. Boundary equivalence still differs by up to six RGB levels in both private modes. Pristine C0 failed before captures; its duplicate initial image was charged after observation, with no retry or numerical budget overrun. PARTIALLY_FEASIBLE, architecture UNDECIDED; neutrality and whole-cache correctness unqualified. No production change; the historical campaign is closed.

- Preserved saved-target natural timeout gate: one Mosaic run used a static SHM H2 reference with four identical whole-RGB captures. The measured target saved node crossed x=0, rendered 182,214 retained-color pixels on H1, contributed no H2 pixels and was removed after a real 2500 ms timeout while the client held serial 50 alive. Two sampled pointer points found the H2 reference and visible H1 saved node. PARTIALLY_FEASIBLE because complete damage, broader input and GPU/texture lifetime remain unqualified; no production change.

- Local target-crossing saved-buffer probe: one measured signed camera delta placed the real target source and saved node across x=0. The target node rendered within H1's clip and retained its SHM pixels; H2 showed no target color but had a repeated 253-pixel reference difference. Mosaic E2 timeout stopped through stale fixture pointer correlation. PARTIALLY_FEASIBLE, architecture UNDECIDED; no production change, commit or push for this campaign.

- Freeze audit of that target crossing: the repeated H2 difference is one filled 11×23 rectangle with Kitty cursor/background colors in pristine and Mosaic runs, strongly compatible with cursor blinking but not causally proven. The E2 visible bound at x=-30 matches the applied tiled clip narrowed by gaps. Timeout/input remain untested.

- Preserved saved-transaction-buffer probe: real held XDG configures produced nonempty saved scene buffers; private saved nodes entered rendering and sampled ACK/timeout paths removed their trees. Mosaic's target was outside H1 after the applied camera reached 1760, so partial-boundary clipping is NOT EXERCISED and the verdict is INCONCLUSIVE. Three distinct executed runner copies and the private-standard fixture timeout are retained; no production change.

- Local static-R90 partial-popup probe: a legal nested B popup crossed the Mosaic viewport by 72 logical pixels; 46,455 B pixels and a real button survived on H1, while H2 was pixel-identical to its reference and delivered no B button. Both standard controls showed 35,360 B pixels and a B button on H2. A crossed visually, but its forbidden-side input was not independently clicked. PARTIALLY_FEASIBLE for the tested matrix; no production change or phase-B commit/push.

- Preserved rotated-output probe: actual 90° output keeps an 800×600 logical viewport while the physical mode becomes 750×1000 and `grim` remains logical-oriented 1000×750. Private standard shows 15,860 B-popup pixels on H2; private Mosaic shows zero, but its B popup ends before the boundary, so partial clipping is unchallenged. Fully hidden owner, pointer, measured physical damage and R0/standard returns pass sampled checks. Pristine R90 capture stopped on a fixture IPC-direction assertion; no replay. PARTIALLY_FEASIBLE, architectural conclusion UNDECIDED; no production change.

- Preserved dynamic-output ownership probe: real 800→960→800 logical resize and H1→H2→H1 workspace migration preserve IDs and sampled applied clips. Exact vacated-output backgrounds and measured physical damage are compatible; two standard-control final calibrations fail through reference occlusion. PARTIALLY_FEASIBLE; maximum clamp, migrated clipped-popup input, opaque spill and effects remain open. Six reconfigurations/six migrations, zero extra camera pairs; no production change. The original temporary evidence manifest was verified before an offline analyzer rerun replaced it; the report records the provenance limit.

- Preserved fractional-scale probe: actual 1.25 uniform/mixed and 1.5 mixed outputs, calibrated logical input, sampled root/nested popup and effects clipping, two exact camera returns and physical settled-damage coverage. PARTIALLY_FEASIBLE; opaque tiled controls, interior shadow, restored Mosaic hit and broader gates remain unqualified. No production/package change.

- Added a source-only private effects/damage probe with a real translucent SHM buffer and striped background. Both standard controls show 27,594 shadow-overflow pixels on the neighboring output; private Mosaic excludes that sampled spill, renders a visibly filtered blur within its viewport and restores exact settled pixels after three camera pairs. Sampled private damage covers changed pixels, while interior shadow, optimized cache, popup effects and saved buffers remain unqualified. The result is PARTIALLY_FEASIBLE; no production or package files changed.
- Recorded an effects-probe campaign deviation: replay/trace variants ran 15 camera pairs against the requested maximum of three. All raw runs are retained; the final runner now reserves a finite campaign budget before launch and rejects excess repetitions.
- Preserved and independently replayed a bounded event-loop pending-input probe. The client flush and server application were observed after decoupled request submission, but no input mutation fell between a matching native headless commit and present in five attempts per private mode. The temporal case remains INCONCLUSIVE / NOT EXERCISED; this research line is closed for the tested backend and geometry. No production or package files changed.
- Preserved and independently replayed a commit/present/input ordering probe. A passive private top-level listener and native headless traces found 20 applied input-region mutations but no natural mutation between an accepted output commit and its matching present; sampled hit and client delivery followed effective input. The temporal verdict is INCONCLUSIVE and the generation-contract decision remains UNDECIDED. A complete outside-Git raw evidence manifest now covers all 94 run/result files. No production or package files changed.
- Preserved a reproducible mutation-coverage fixture and source audit. A fresh-clone replay rebuilt private SceneFX/Mosaic and verified pinned wlroots sources in its own lab. Bounded trace-only runs separated camera/clip, popup lifecycle, same-size client content and input-region hits; the latter changed hit testing at an unchanged partial epoch. A real private output-destroy path released one held reference, while backend rejected-commit cleanup remains untested. The architectural generation decision is UNDECIDED; STEP20C stays NEXT.
- Preserved a private applied-scene/input probe. A separate corrective patch scopes retained output buffers to their owner, samples three buffers through present, and logs partial applied-scene epochs at build and real pointer hits with client delivery kept separate. The tested matrix is PARTIALLY_FEASIBLE; STEP20C remains NEXT and multi-output Mosaic unsupported.
- Preserved a private output-buffer identity fixture and protocol. In fixed headless geometry, 32 GBM buffers were read through SceneFX FX and linked to native commits and presents; three second reads matched. Four references were released early by a cross-output scope defect in the private instrumentation, not by demonstrated same-output buffer replacement. Readback changed timing, so transaction/input coherence and production readiness remain unproven. STEP20C stays NEXT and multi-output Mosaic unsupported.
- Added a source-only private presentation-identity probe at fixed negative-origin headless geometry. Native wlroots output present events paired with 470 accepted commits by output and `commit_seq`; applied scene/clip and sampled input agreed at the target states, but exact commit-buffer pixels remain unproven. STEP20C stays NEXT and production multi-output Mosaic unsupported.
- Added a source-only temporal damage fixture at the fixed negative-origin geometry. Ten XDG open/close cycles in each of three modes and sampled camera transitions restored popup regions without a detected ghost pixel; private damage/commit traces remain insufficient for a presented-frame guarantee. STEP20C remains NEXT and multi-output Mosaic unsupported.
- Added a source-only negative-origin nested-grab probe. Two outputs at `x=-800/0` verified signed pointer conversion, real standard popup crossing, sampled private Mosaic clipping, hidden-owner return, standard-layout return and a positive horizontal regression. Production STEP20C support remains open.
- Added a source-only stacked-output nested valid-grab probe. Both standard controls showed real popup overflow and pointer delivery below `y=600`; the private guarded Mosaic build clipped A/B at sampled lower points, preserved hidden-owner and standard-layout return, and reproduced the horizontal control. STEP20C production support remains open.
- Added a source-only nested XDG grab fixture and a comparative headless probe. Two real button serials produced an interactive toplevel → popup A → popup B chain; sampled guarded Mosaic clipping, child-first dismissal, hidden-owner camera return and standard-layout return were feasible. STEP20C production support remains open.
- Preserved the valid-serial XDG grab / hidden-owner probe's original C and Python sources, private patches and protocol XML with hashes and a disposable headless replay procedure. Its result remains **PARTIALLY_FEASIBLE**; no production source, package or UX policy changed.

### Patch export tooling

- Fixed cumulative Mosaic patch export to diff against the verified pristine SwayFX 0.6 import, preserve the real Git index, validate clean-base application and source equivalence, and replace the patch atomically only after successful checks.
- Added offline regression coverage for complete and repeatable export, working and staged edits, source additions/deletions and mode changes, index preservation, conflict refusal, and artifact preservation on failure or interruption.

### Documentation

- Established a public landing page, original source-controlled SVG banner, canonical documentation index, UX contract, packaging contract, known limitations, and documentation-first roadmap.
- Clarified the current single-row implementation, historical STEP evidence, package coexistence, and upstream forward-port policy.

### Existing implementation

- Native Mosaic layout with workspace-local state and real tiled Row 0.
- Insertion/reflow/focus model tests and `layout mosaic` command.
- Workspace-local `layout toggle mosaic`, logical scrolling, and first viewport scene translation.
- Mosaic `get_tree` IPC state and headless compositor validation.
- Isolated Arch packaging path for `/usr/bin/swayfx-mosaic`.

### Safety and limitations

- Ordinary SwayFX remains separately installed and independently updatable; no automatic session replacement.
- Multi-output clipping, multiple runtime rows, camera animation, Session V2 restore, final cell sizing, and daily-driver qualification remain open. See [known limitations](docs/KNOWN_LIMITATIONS.md).

### Preserved offline nested geometry gate — EXPERIMENTAL / INCONCLUSIVE

- Replayed true static failure SHM/configure/destination/titlebar/paint data and
  nine preserved private/control popup campaigns, without any graphical run.
- Prepared typed shared geometry, actual-configure/lifecycle/input validators,
  exhaustive145 historical assertions/2288 AST-site register and195 passing
  offline tests; compiled only three C17 client utilities, never executed.
- Preserved the independent deep review and bounded repair/delta review:
  input correlation and terminal IPC evidence fixes accepted, readiness receipt
  semantic-completeness defect remains CORRECTION_REQUIRED. No READY claim.
- Preserved the [geometry gate](docs/design/STEP20C_STATIC_NESTED_GEOMETRY_REPLAY_GATE.md)
  and experimental source fixture with its historical INCONCLUSIVE verdict; STEP20C NEXT, architecture UNDECIDED and
  production multi-output NOT YET SUPPORTED. No production/package/UX change.

## Local receipt semantic integrity gate — offline only

The [R02 receipt-integrity gate](docs/design/STEP20C_STATIC_NESTED_RECEIPT_INTEGRITY_GATE.md)
is local and experimental. The real writer/aggregate share semantic validation
through the existing immutable-data replay and R01 input correlation; typed
receipt content, scenario-specific omissions, AST/test inventories and final
review bindings are checked. Historical geometry evidence remains INCONCLUSIVE;
this correction cannot retroactively qualify static popups. The final local gate
is READY_FOR_NEW_RUNTIME: 206 offline tests pass, 13478 recorded invalid mutations
are rejected, and independent review accepts the final 18-source/15-receipt
bindings after one bounded B-01 writer repair. The actual writer returns READY
with five PASS markers and no blockers. R02 is RESOLVED for this offline fixture;
partial AST coverage and runtime limitations remain explicit. Phase B is
uncommitted and unpublished, and readiness only permits requesting a separate
future authorization. No compositor/runtime authorization,
production/UX/package change or STEP20D work follows. Architecture UNDECIDED;
STEP20C NEXT; production multi-output NOT YET SUPPORTED.


## Static E2 completion and private integration evidence — 2026-10-13

- Recorded the independently accepted shared applied-standard-transition gate:
  104 targeted offline tests PASS, historical replay and 277 classified assertion
  sites, with the separate older-suite 145 PASS/61 setup errors preserved.
- Recorded the sole unchanged-static-binary E2 run as EXPERIMENTAL / FEASIBLE
  for two genuine grabs, Mosaic pixels, applied splith return, standard B/A
  input, explicit B close/A survival and normal owner close/full-H2 restoration.
  Raw terminal IPC1, startup entries and historical INCONCLUSIVE findings remain.
- Recorded a private initial integration candidate BUILDABLE from one SceneFX
  build, one Mosaic build/model/ELF check and one exact five-file DESTDIR stage.
  Its distinct binary is graphically NOT TESTED. Independent P2 comment repair
  is accepted as COMMENT-ONLY / NOT REBUILT; initial recipe/build evidence stays
  unchanged, recipe execution NOT RUN and legal compliance NOT APPROVED.
- Added the [completion report](docs/design/STEP20C_STATIC_E2_TRANSITION_AND_COMPLETION.md)
  and [current fixture results](tooling/probes/step20c-static-e2-completion/README_RESULTS.md),
  preserving the original README as a sealed pre-execution protocol. Final batch
  audit is PASS / ACCEPTED with no BLOCKING finding. No production source, UX, package, public ABI
  or cumulative patch change; STEP20C NEXT, multi-output NOT YET SUPPORTED.