[tetrisgm]

HD native texture authority previously refused render passes even though the transaction restores authoritative CPU VRAM and GPU presentation surfaces. This change permits that path. In explicitly opted-in forward netplay, a temporal pass receives the same presentation-only local-view scope as the base own-view draw. Rollback resimulation remains refused.

The final diff is five files. Input replay, scheduling, PGXP, rewind, render-thread, and tracing work from the earlier draft has been removed.

Validation:

• render_pass_sandbox_test passes on Windows and macOS. Positive opted-in local temporal passes, discarded passes, device-read abort, rollback refusal, and a nonparticipating peer preserve the shared RAM/register/clock/interrupt digest.
• The real-GL HD fixture enables native HD authority, draws in a kept and a discarded pass, and verifies authoritative CPU VRAM plus restored presented-pixel digests. Focused passes mode passed 14/14 checks at 1x and 9x on macOS; Windows sandbox passed after the fixture change.
• An R4 two-peer Link race resumed into a loaded state at effective 18x (4320 internal lines). Host 120 Hz presentation with the title's local HD HUD pack active (36 replacements indexed), guest 30 Hz with HD disabled: 50 own views each, six actual host temporal passes and none on the guest, zero aborts, 16/16 before/after shared-digest receipts per peer, four matching common netplay digests, zero dispatch/segment misses. The short sample did not apply an HD replacement draw (applied_draws=0), so it proves pass admission with HD authority but not replacement-pixel use. R4 consuming-title receipt.
• Matched built-performance check: upstream master d19d8c7e versus this PR's renderer, same hidden real-GL 9x scenes, five interleaved runs per scene and OS. Windows capture medians: 329.25 ms base / 324.63 ms PR; lines: 435.23 / 450.08 ms. Mac capture: 306.12 / 290.59 ms; lines: 405.36 / 403.73 ms. Digests match within each OS, all checks pass, and the variation is larger than the patch's observed timing differences. These are fixture process-wall measurements, not title output FPS.
• Same-binary R4 title feature-cost A/B at 18x and the same checkpoint: 30 Hz host setting made zero passes with 6.56 ms average local draw; 120 Hz setting made six passes with 6.31 ms average local draw. Both had matching common digests and no misses/aborts. This isolates enabling temporal passes, not upstream-master versus PR.

No sustained 120 FPS claim is made. Desktop Duplication delivered no updates on this PC; a 12-second Windows.Graphics.Capture probe saw changing window pixels but was capture-bound (13.15 captured surfaces/s versus 349 GL source captures and 167 swaps), so it cannot establish physical display cadence. The code change is restricted to permission and local-view scope; ongoing high-refresh throughput work is separate.

[mstan]

Reviewed head 1cd67fc.

The stated render-pass/HD/local-view compatibility is meaningful, and simulation correctness must remain unconditional inside an enabled enhancement. This draft is much broader than that contract: it also adds production input replay/benchmark controls and host-span tracing and changes host time, scheduling, PGXP, rewind and render-thread behavior. Please isolate the render-pass permission/transaction change from that independent tooling and performance work, or identify a minimal explicit dependency stack. Preserve canonical shared state and rollback-resimulation refusal; qualification must compare the shared digest before/after an actual local-view pass with HD authority, including abort/failure and a disabled/nonparticipating peer. The synthetic GL fixture is useful but does not clear the source ownership and scope questions. Keeping this as draft, with TechnicallyComputers assigned for the engine netplay/local-view policy. Tracking: beads-8zz2.

[mstan]

Rebased onto frozen master, preserving current bounded sleeper/FLS code and projective transaction/failure guards, and pushed ce40165. Added the missing emulation-side render-pass flag double to the standalone HD GL fixture. Its existing four-wave timer test now has a 45-second CTest envelope consistent with four internal ten-second watchdogs; the internal watchdogs remain unchanged.

The combined 21-test native gate and real GL scale/thread/quality matrix pass; the HD fixture also passes 209/208 checks and native reference 19 after the preload/STP expectation fixes on their owning branches. A generic refusal test with no configured pass is not positive HD transactional-pass qualification. The draft still needs its consuming-title positive pass/opted-in netplay contract and matched built-performance acceptance before promotion.