AVE 1.0.15

AVE 1.0.15 Release Notes

Status: Released July 10, 2026. Signed packaging and installed-app smoke passed.

Base: changes after 5319fb3 (Mark AVE 1.0.14 released) through fd00864 (Harden AVE runtime and source intelligence).

Highlights

• Added strict project-envelope validation, malformed-primary quarantine, newest-valid-backup recovery, serialized project writes, and a close/save handshake.
• Bound analysis completion to project, asset, path, and media revision so switching projects or relinking footage cannot apply stale results to the wrong state.
• Added owned lifecycle and cancellation contracts for local sidecars, render jobs, generated media jobs, model downloads, runtime installs, and other long-running work.
• Added the Plan prepare/apply/acknowledge protocol so stale or altered work is rejected and Free execution usage is recorded only after a fully successful renderer acknowledgement.
• Added one shared tool-contract registry, scoped pending chat intents, bounded planner context, and parity checks across routing, approval, workflow metadata, and execution support.
• Replaced unused project-index work with production Source Intelligence that combines transcript, visual, frame, and asset evidence into traceable source ranges.
• Added content-based analysis identity, component-level compatibility checks, resumable partial reuse, scene-aware sampling, and evidence-aware content cutdowns.
• Made local model downloads revision-isolated and checksum-aware, made managed Python runtime activation recoverable, and rejected generated files that do not decode as their declared media type.
• Made render output commit atomically, improved cancellation cleanup, and ensured the render worker terminates its FFmpeg children if Electron exits unexpectedly.
• Added a 1.5-second minimum animated AVE launch screen that owns startup state without flashing Project Home, plus a transparent rounded-corner packaged icon.

Verification Completed

• node --check electron/main.cjs
• npm run build
• npm --prefix render-worker run build
• npm run test:run: 1,331 passed, 1 skipped
• npm run ai:prompt-regression
• npm run ai:source-eval: precision 1.0, recall 1.0, 100-asset search p95 12.70 ms against a 500 ms budget
• npm audit --json: 0 vulnerabilities
• npm --prefix render-worker audit --json: 0 vulnerabilities
• Production Polar checkout: AVE Pro Lifetime, $79 one-time subtotal, active 50% launch promotion, and matching organization/product/benefit identifiers
• git --no-pager diff --check
• Operator smoke: project open, Plan rough cut, persistence, launch screen, and native export passed; export was noticeably faster in the tested project.
• Signed package verification: app and DMG signatures, Gatekeeper assessment, stapled notarization tickets, and DMG integrity passed.
• Packaged bundle version: AVE 1.0.15 (16).
• npm run desktop:audit-package: 7,895 app.asar entries passed; app-update.yml is present in both the app bundle and updater ZIP.
• Signed DMG installation smoke passed before publication.
• Settings > Updates successfully upgraded the installed previous public version to AVE 1.0.15.

The Source Intelligence evaluation uses deterministic synthetic fixtures. It proves contract correctness and bounded performance, not broad human editorial quality on real footage.

Known Caveats

• Project schema v2 remains in dual-write mode: analysis stays embedded in project.json and is also written to sidecars for downgrade safety.
• A database-backed project index remains intentionally absent until measured real projects show that the bounded in-memory Source Intelligence path is a bottleneck.
• Human review on broader real-footage fixtures remains ongoing even though the current operator rough-cut smoke and synthetic retrieval gate passed.
• The packaged rounded icon was not separately re-smoked on the older macOS machine before publication.