feat(engine,producer,cli): drawElement release telemetry on render_complete

[vanceingalls]

What

Release telemetry for default-on drawElement (#1998): every render_complete event now reports the capture-mode outcome, which of the three decision layers disengaged drawElement (and why), the self-verification verdict with its margin, and the safety-net counters.

Why

#1998's safety case rests on "the failure mode is a slower render, never wrong pixels." That's only monitorable if we can see, in the wild: how often drawElement actually engages, what gates it out, how often the self-verify net fires, and — the early-warning signal — how much margin verification passes with (de_verify_min_db drifting toward the 32dB threshold predicts fallback spikes before they happen). Mirrors the static-dedup telemetry pattern (session fields → CapturePerfSummary → RenderPerfSummary.drawElement → snake_case event props) that powers the dedup dashboard.

New render_complete props

prop	meaning
de_capture_mode	final mode: drawelement | screenshot | beginframe
de_compile_gate	compile-time gate: 3d | mix_blend_mode | shader_transitions
de_clamp_reason	producer clamp: parallel | disk_path (unverified-path confinement)
de_gate_reason	engine init gate: swiftshader | css_effect:* | at_risk_timeline | 3d_init_failed | supersampling | render_mode_hint
de_worker_encode	verified drain active
de_verify_armed / de_verify_checked	ground-truth samples armed at init / compared at drain
de_verify_min_db	min PSNR across checked samples — margin above the 32dB threshold
de_verify_init_ms	ground-truth capture cost at init
de_self_verify_fallback / de_fallback_reason	net fired → whole-render screenshot re-run; psnr | blank
de_blank_suspects / de_blank_deterministic_accepts / de_blank_recaptures	blank-guard counters
de_boundary_frames	Lim 6 clip-cut frames captured via per-frame screenshot
de_ncpr_fallbacks	per-frame "No cached paint record" screenshot fallbacks

All reason fields are low-cardinality by construction; parallel workers |-join distinct values (same convention as static_dedup_skip_reason).

Test plan

• Live renders, three paths validated end-to-end (engine → producer summary → CLI props):
  • drawElement path: mode=drawelement, verify_armed=4, verify_checked=4, verify_min_db=57.8, verify_init_ms=271, boundary_frames=5
  • blur-gated comp: mode=screenshot, gate_reason=css_effect:filter
  • forced verification failure (HF_DE_VERIFY_MIN_DB=99): self_verify_fallback=true, fallback_reason=psnr
• tsc clean across engine/producer/cli; engine suite 899 pass, CLI suite 1261 pass (both pre-existing failures reproduce without this diff — upstream color-grading test, figma import tests).

🤖 Generated with Claude Code

[vanceingalls]

• feat(engine,producer,cli): drawElement release telemetry on render_complete #2002 👈 (View in Graphite)
• feat(engine,producer): drawElement fast-capture default-on with runtime self-verification safety net #1998
• perf(engine,producer): batch N drawElement frames per CDP round-trip (HF_DE_BATCH) #1928
• feat(lint,player): fast-capture lint rule + player media sync #1921
• feat(producer): fast-capture render stages + remote bg-image localizer #1920
• feat(engine): frame-capture core — fast-capture routing, worker-encode, dedup extension #1919
• feat(engine): 3D projection + compositor-effect risk gate #1918
• feat(engine): drawElementImage capture service #1917
• feat(engine,cli): drawElement fast-capture config + CLI flag #1916
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[miguel-heygen]

Current-head review at e73ff00b5f6abf24b078a20df32dc59f0b9b08dd.

Audited: packages/engine/src/services/frameCapture.ts, packages/engine/src/types.ts, packages/producer/src/services/render/stages/captureStreamingStage.ts, packages/producer/src/services/render/stages/compileStage.ts, packages/producer/src/services/render/perfSummary.ts, packages/producer/src/services/renderOrchestrator.ts, and the CLI telemetry mapping in packages/cli/src/commands/render.ts / packages/cli/src/telemetry/events.ts.

The telemetry plumbing matches the stated shape: engine session fields flow into CapturePerfSummary, producer compile/clamp/drain state is folded by aggregateDrawElement, and the CLI emits stable snake_case render_complete props. The compile-time gates (3d, mix_blend_mode, shader_transitions), producer clamp (parallel, disk_path), engine gates, NCPR counts, boundary-frame count, and worker-drain verification counters all have clear sources.

Important notes, not blockers:

• packages/producer/src/services/renderOrchestrator.ts:1884-1887 derives de_fallback_reason from the error message with /blank/i; a typed reason on DrawElementVerificationError would be more robust if the upstream message text changes.
• packages/producer/src/services/render/perfSummary.ts:45-87 is the highest-value unit-test seam. A small aggregator test covering divergent worker modes/gate reasons plus summed counters would lock down the low-cardinality | join behavior.
• The stack below this PR is still carrying separate blockers on #1918/#1919/#1998. This approval is for the #2002 telemetry delta only.

CI note: no failed check-runs are currently reported for this head; Graphite/regression shards are still pending.

Verdict: APPROVE
Reasoning: The PR is straight, coherent telemetry plumbing with bounded-cardinality fields and no #2002-specific blocker found. Remaining risk is pending CI and the already-recorded downstack issues, not this diff.

• Magi

[miguel-heygen]

Restack re-review at 52b282cd227f3b6ae127d33b05c6175415db5a6b.

Audited: the current #2002 telemetry diff across frameCapture.ts, types.ts, captureStreamingStage.ts, compileStage.ts, perfSummary.ts, renderOrchestrator.ts, and the CLI telemetry event mapping.

The telemetry delta still reads clean after the restack. The data flow remains session fields -> CapturePerfSummary -> RenderPerfSummary.drawElement -> snake_case render_complete props, with compile gates, producer clamp reasons, engine gate reasons, worker/drain counters, boundary/NCPR counters, and self-verify fallback state all surfaced. My earlier non-blocking notes still stand: typed DrawElementVerificationError fallback reasons and an aggregateDrawElement unit test would harden future refactors, but I do not see a blocker in this diff.

CI note: the current active PR check view still has Graphite, Preview parity, and regression shards pending. I am approving the code delta, not declaring the full stack merge-ready before those finish.

Verdict: APPROVE
Reasoning: The #2002 telemetry wiring remains coherent at the new head, and no new restack-specific blocker showed up.

• Magi

[terencecho]

LGTM per my per-PR review + re-review on the Slack thread. Non-blocking nits I raised (where present) are fair as post-merge follow-ups.

— Review by tai (pr-review)

The base branch was changed.