docs(changelog): weekly digest 2026-07-27–2026-08-03

[jrusso1020]

What

Adds the weekly digest entry for July 27 to August 3, 2026 to docs/weekly-updates.mdx.

Generated with bun run changelog:weekly --from 2026-07-27 --to 2026-08-03 --write, then rewritten to publish quality: entries grouped under Features, Fixes, Under the hood, and Docs, prose rewritten to describe user impact, and the <!-- TODO: review --> marker removed.

Why

Keeps the public weekly-updates page current. The digest is the RSS-backed summary readers get between versioned releases.

This was a 219-commit week across 14 releases (v0.7.77 to v0.7.90), and the generator's own draft surfaced 10 items, so the rewrite is substantially wider than usual.

How

• Ran the generator over the July 27 to August 3 range.
• Rewrote the new <Update> block: 7 Features, 15 Fixes, 2 Under the hood, 6 Docs bullets.
• Headline items: per-property keyframe lanes in the Studio timeline with virtualization on by default; the keyframe-targeting correctness sweep; three themed registry families adding 29 catalog items from an outside contributor; transparent GIF output; the end-to-end media-probing hardening; a zero-advisory install.
• No SHA or PR number was hand-typed. Every link was expanded by script from a short SHA, and the script refuses to emit unless each one resolves to a 40-character commit that is an ancestor of origin/main and lands inside the July 27 to August 3 window.

Two corrections worth flagging, since both would have shipped as over-promises

• The short-comp drawElement band changes no routing yet. The generator, and my first draft, read this as a default-on speedup. feat(producer): short-comp DE inversion band — baseline release (telemetry only, routing off) #2875 is explicitly "PR A of two" and ships the band decision as telemetry only; at origin/main the route is gated on HF_DE_SHORT_BAND_ROUTE === "true", off by default, with the flip deferred to a follow-up. It is written up under Under the hood as groundwork, not as a render that got faster. Separately, feat(producer): lower parallel-DE router floor to 700 frames + power-state telemetry #2838's floor drop sits behind HF_DE_PARALLEL_ROUTER, also off by default, and says so.
• Registry item count. feat(registry): mk / yt / hw theme families + beat-freeze-cut — 30 items #2939's own commit body gives a per-family breakdown that sums to 30 while its subject says 29. Counting registry-item.json files the merge actually added to main gives 29 family items (hw- 11, mk- 9, yt- 9) plus beat-freeze-cut, so 30 in total. The digest publishes the counted numbers and drops the inconsistent block-versus-component split.

d74afc7b7 is described as observational because its own commit body states it deliberately does not feed the routing gate.

Test plan

Docs-only change. No runtime code touched.

• Unit tests added/updated
• Manual testing performed
• Documentation updated (if applicable)

Verification performed:

• git diff origin/main --numstat shows only docs/weekly-updates.mdx, 55 insertions and 0 deletions. Purely additive; both prior entries are untouched and the file still holds 3 balanced <Update> blocks.
• All 121 commit links and 72 PR links checked: every commit resolves, is an ancestor of origin/main, and is inside the window; every referenced number was confirmed via the API to be a real pull request whose title matches the claim it is cited for.
• The #2791 and #2695 attributions were confirmed by second-parent ancestry rather than inferred, because merge-ancestry mapping is wrong for squash-merged commits.
• Confirmed the TODO marker is gone, the new block contains no em-dashes, no stray braces, no unintended JSX tags, and no unbalanced inline-code backticks.
• Commit is signed, and lefthook largefiles, tracked-artifacts, and commitlint all pass without --no-verify.

The video version follows in this PR once it renders and clears its gates.

— Rames Jusso

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated (UTC)
hyperframes	🟢 Ready	View Preview	Aug 3, 2026, 4:42 PM

💡 Tip: Enable Workflows to automatically generate PRs for you.

[jrusso1020]

Weekly video is embedded in the July 27 block as of 05b8e68f.

https://static.heygen.ai/hyperframes/changelog-videos/weekly-changelog-2026-07-27-2026-08-03.mp4

51.7s, square 1080, built with the repo's changelog-video skill. Five scenes, each an animated mock of the surface rather than a bullet list:

1. Timeline — a track expands into one lane per animated property, a keyframe drags with a snap line, then the row count shows only the visible rows rendered.
2. Targeting — the edit lands in the clicked element's lane, and the inferred target that would have been used fades out.
3. Catalog — the three families fill in to 29 items, credited to the contributor.
4. Output — --format gif is typed as the visible cause, then the alpha shows through against a checkerboard.
5. Install — a clean install reporting zero advisories.

Gates, all green at this head:

• hyperframes check: 0 errors, contrast 60/60 WCAG AA, runtime and motion clean.
• Seam gate: 0 fail, 0 warn across 6 seams.
• Rendered-frame check at every scene midpoint: caption rail reads 236 to 240 on each frame inside a spoken window against ~40 for a bare rail, background video is live across neighbouring frames, and the mix sits at -19.9 dB with the music bed about 9 dB under the voice.

Every cut and internal beat is derived from the voiceover's word timings rather than hand-typed, so the motion lands on the narration.

One note on wording: the video does not describe the short-composition drawElement band or the parallel drawElement router as making renders faster, since both remain behind default-off flags this week.

— Rames Jusso

[miga-heygen]

R1 — Weekly digest 2026-07-27–2026-08-03 (head 05b8e68)

Verdict: Approve

Purely additive docs change — 59 lines to docs/weekly-updates.mdx, no runtime code. Well-structured (Features, Fixes, Under the hood, Docs) and accurately scoped.

What was verified

PR references: Spot-checked 7 of the 72 referenced PRs (#2791, #2926, #2955, #2327, #2855, #2939, #2875) — all MERGED, titles match the digest's descriptions.

Two corrections the author flagged are correctly applied:

• Short-comp drawElement band (#2875) is described as telemetry-only groundwork under "Under the hood," not as a shipped speedup. Confirmed the PR title says "telemetry only, routing off."
• Registry item count: digest says "29 items" for the three families + "beat-freeze-cut joins the transition set as a thirtieth." Consistent with the gh pr view title showing "30 items" total. Correctly split.

MDX structure: <Update> tag balanced with correct props, <Frame> wraps the video embed, section headers use ##, no stray JSX or broken backticks.

Date range: Consistent across PR title, commit messages, <Update> label/description, and video filename.

No findings. Ship it.

---

Review by Miga

[vanceingalls]

Verdict: APPROVE

Docs-only, docs/weekly-updates.mdx +55/-0. CI clean where it runs (Mintlify, CodeQL, Format, Analyze all pass; the rest skip on docs-path filter). The BLOCKED merge state is reviewer-gate.

High-stakes claims verified

Both "off by default" assertions in "Under the hood" are correct in current main:

• HF_DE_SHORT_BAND_ROUTE — renderOrchestrator.ts:2682 reads process.env.HF_DE_SHORT_BAND_ROUTE === "true"; code comment at :2673-2674 says "EVALUATES the band on every render and emits the decision, but only routes on it when HF_DE_SHORT_BAND_ROUTE=true (flipped by default in a follow-up release)." Digest phrasing ("stays behind HF_DE_SHORT_BAND_ROUTE, so the follow-up flip is measurable against a real baseline") matches the code's own comment exactly.
• HF_DE_PARALLEL_ROUTER — renderOrchestrator.ts:2750 reads === "true"; JSDoc at :1596 states "the router's own kill switch, default off." Digest phrasing ("The router itself stays off by default behind HF_DE_PARALLEL_ROUTER") matches.

Both preventable over-promises were self-caught in the PR body — good discipline.

Registry count reconciliation (29 family + 1 beat-freeze-cut transition = 30 total, per the counted registry-item.json files) is also correct and stated cleanly.

One observation, non-blocker

#2907 "fix: align local WebGPU capture behavior" (merged 2026-07-31) is not cited anywhere in the block. It's inside the window and, from its own body, is more than an alignment nit:

• Introduces data-requires-webgpu as the composition capability marker with actionable diagnostic when auto-GPU resolves to software (previously silent SwiftShader fallback).
• Adds an hf-seek event.detail.waitUntil(...) completion contract awaited before frame capture.
• Re-presents paused WebGPU canvases via same-time heartbeat without advancing simulation.
• Promotes WebGPU validation warnings to deduped runtime errors in hyperframes check.
• 142 tests across core/engine/cli.

Not a blocker — the generator's 10-item draft naturally trims, and WebGPU is niche. But if you want another Fix or Feature bullet, "WebGPU compositions capture the correct GPU path" would slot in cleanly (probably alongside the data-requires-webgpu marker introducing a new authoring surface).

Nits (none blocking)

None. TODO marker gone, no em-dashes, <Update> tags balanced, MDX renders (Mintlify preview passed), commit signed, lefthook clean per PR body.

Review by Via

[miguel-heygen]

Exact-head review at 05b8e68.

No blocking findings.

• The authoritative diff is additive and docs-only: +59/-0 in docs/weekly-updates.mdx; git diff --check is clean.
• All 72 referenced PRs are merged, and every merge falls within July 27–August 3, 2026. All 121 referenced commit SHAs resolve, are reachable from current main, and fall within the stated window.
• The MDX structure is balanced, the Mintlify preview renders the new date block, and the embedded video returns HTTP 200 with video/mp4.
• I independently rechecked the two default-off orchestration claims and the 29 family items + 1 transition registry count; the wording matches the code and source PRs.
• Exact-head CI is terminal-green: 17 successful checks, 27 intentional skips, no pending or failing checks.

Non-blocking editorial notes only: the PR description still reports the pre-video +55 diff and 7/15/2/6 section counts; the exact head is +59 with 7/15/3/7. Also, #2907 is an optional in-window omission, as Via noted; this digest is curated rather than exhaustive.

Verdict: APPROVE
Reasoning: The published update is structurally valid, source-backed, in-window, and accurately describes the shipped behavior. The remaining discrepancies are PR-description bookkeeping, not defects in the changelog.

— Magi