[thehobbit85]

An oversized positive DM_PRESENT_LATENCY request makes link_sync compare CAMetalDisplayLink's bounded readback with an unsupported value, repeatedly writing and logging every frame. Normalize the request to the supported maximum of 2 before creation and comparison. Valid overrides 1/2 and main's automatic policy—2 in variable-refresh fullscreen, 1 otherwise—remain intact.

Addresses #603, finding 2. The separate ProMotion late-present and fixed-rate display-link delay findings remain open.

Current integration and validation

Rebased onto pinned main c1df4ddaaedc3c18228065ea5b909688b79884e2; head e698f325f004290576a6d96a9ab8770cda480654. The same existing PR remote branch and its canonical local tracking branch are retained. All paginated issue comments, reviews and inline feedback were read; no new feedback was found.

The independently reviewed override normalization, automatic variable-refresh policy and repeated-write regression cases are unchanged. Main's compact NVN command records do not replace the explicit-override bound.

• Fresh: 11 documentation tests and the environment-switch index test passed.
• Historical: t_present_rescue --metal passed 80 checks with MTL_DEBUG_LAYER=1 at 1d882a01, including oversized 3/INT_MAX requests across both screen types and window/fullscreen states. Its presentation-policy/fixture sources are unchanged; the run predates the newer NVN command-record changes and is not a final-head native run. That build reported 52 existing graphics protocol warnings.

Range-diff marks the feature patch unchanged, and the old-to-new tree delta exactly matches the upstream increment. Ancestry and git diff --check pass. The previous d138-based head 3e18f0c3548da726ec24fa8fed85f7a25ff15c32 completed all 7 CI checks successfully; that is historical evidence. CI for this new head is pending. Central shared validation passed on 4820ba509adb90ee7cca8c5a65a51e0a4dc12e74 with GAME=010015100B514000: t_fs (102 checks), t_fs_case (51), t_fs_commit_list (98), t_nfc (54), t_pad_hid (19), and t_system_event. Their production/fixture sources, host headers, profile inputs, build flags and target recipes were verified identical to this head. This is shared fixture coverage, not a full host build or full graphics run of this PR.

Local native tools are Xcode 27.0/Metal 32023.921; supported Xcode 27.2/Metal 32023.922 verification belongs to the new CI run. No gameplay, physical-display timing or interactive HUD check, extraction rebuild, device installation, game shader rebuild, or fresh game shader/interface comparison was performed. Earlier full-suite/game evidence remains historical at its recorded revisions.

[seedds]

Landed as f4bfc01, with you as co-author. The link_latency bound, the t_present_rescue additions, the env_switches.tsv line and the sentence in the existing frame-rate.md entry are in as you wrote them; t_present_rescue --metal passes 80 checks with the debug layer here. I left out the new frame-rate.md section and its symptoms.md line: the symptom only appears when a debug switch is set outside its documented range, which now says 1 or 2. #603 stays open for its other findings. Thanks!