acdream/docs/research/2026-07-24-slice-f5-render-scene-shadow-gate.md

Slice F5 Render-Scene Shadow Gate

Date: 2026-07-24

Candidate: 81e2f1a57577457436f244f8c9d083e14a9de362

Result: accepted; Slice F is complete

Purpose

Slice F built the future render scene without allowing it to draw. The physical gate compared its active projections with the accepted current renderer after streaming, network, teleport, animation, attachment composition, and the final update-boundary journal publication. This was an exact state referee, not a screenshot approximation.

The retained scene can contain inactive identities that have temporarily left spatial presentation. Therefore the binding count is the active matched projection count, not total retained entity capacity. Logical teardown still removes those identities through OnResourceUnregister.

Exact physical evidence

Route	Report	Final comparisons	Mismatch	Pending	Rejected	Exit
Capped nine-stop	logs/connected-r6-soak-20260724-232827.report.json	1,663 / 1,663	0	0	0	graceful
Uncapped nine-stop	logs/connected-r6-soak-20260724-233826.report.json	1,677 / 1,677	0	0	0	graceful
Uncapped dense Arwic	logs/connected-dense-town-20260724-234807.report.json	81 / 81	0	0	0	graceful

Every route reported:

• Passed=true, ExitCode=0, and GracefulExit=true;
• source commit, embedded binary commit, and candidate commit equal to 81e2f1a57577457436f244f8c9d083e14a9de362;
• shadow comparison enabled at every canonical checkpoint;
• active matched projections equal to the current-path oracle;
• firstMismatch=null;
• cumulative rejected-generation, out-of-order, stale-incarnation, and missing-record counts all zero.

The two nine-stop routes covered Caul, Sawato, Rynthid, Aerlinthe, Holtburg, same-location Sawato revisit, Caul return, Sawato plateau, and Caul plateau. The route exercised movement, jump, combat, portal churn, repeated materialization, generation replacement, and graceful shutdown. Dense Arwic covered 15,040 active projections and 111 animated entities during the turn sample.

Shadow footprint

The final checkpoint's estimated retained shadow footprint was:

• capped Caul plateau: 21,891,296 bytes for 21,216 retained identities;
• uncapped Caul plateau: 21,894,656 bytes for 21,218 retained identities;
• dense Arwic: 20,001,016 bytes for 15,791 retained identities.

This footprint is intentionally additional during Slice F because both the current and candidate representations coexist. Slice G must remove the old whole-world partition ownership after production parity; Slice F numbers are not the post-cutover steady-state memory target.

Rejected evidence and root corrections

Earlier physical runs were deliberately rejected:

• 056bbd4e: ordinary roots could not be recovered after same-location rehydration because derived synchronization walked only retained records;
• 91463db5: the ordinary journal drain was skipped while simulation was quiesced instead of being a final update-frame commit;
• 9fab1feb: destination pose could not restore an equipped projection after presentation withdrawal;
• bb1f4a64: equipped spatial withdrawal was incorrectly treated as logical identity destruction.

The final model separates logical lifetime from spatial presence. Ordinary roots reconcile from LiveEntityRuntime's canonical active workset. Equipped children remain callback-owned, retain their exact inactive scene identity during spatial withdrawal, reactivate only from an authoritative visibility or pose edge, and unregister only on logical resource teardown.

Rollback

No production draw consumer changed in Slice F. The exact pre-F/G runtime rollback anchor remains e7d9d6fa. A later visual failure in Slice G must revert the exact production cutover commit recorded in the plan and render pipeline digest; it must not discard this referee or evidence work.