Skip to content

Latest implementation and acceptance results

Latest full application round: 2026-10-07, version 0.2.1; portable draft publication: 2026-10-08, version 0.2.2. This is a result index; active scope/removed conditions are in Plan, criteria in V01–V31. Detailed records retain complete commands/counts/environments/first failures/recovery; this page does not repeat every stage.

Local environment: Windows 11 Pro x64 10.0.26300, Ryzen 9 9950X3D, about 125 GiB RAM; Node 24.21.0, Rust 1.98, Tauri CLI 2.12.0, WebView2 154.0.4258.53, locked dependencies. Linux uses isolated rootless Podman on WSL/Debian. CI runners/commits are recorded separately.

Current results

Check Verified result Record and scope
0.3.2 compatibility repairs Four Codex and 27 Kilo versions checked individually; creation fingerprints no longer reset large-file reads repeatedly; 493 Codex files/one Kilo database recover in an independent old-DB copy without changing prior usage; ten-locale OTel wording, local checks, Windows build and 11 headless checks pass; all 12 tag CI jobs pass, 13 assets are verified and Release-triggered documentation refresh passes 0.3.2 checks; unknown versions keep compatibility notices; public Release ID 409806776; no additional local native GUI, installer lifecycle or installed-client upgrade acceptance
0.3.1 dependency upgrades Latest compatible direct dependencies, Rust 1.99, synchronized npm/pnpm/Cargo locks; local product/native/documentation checks pass; all 12 tag jobs pass, 13 downloaded assets match digests/source reports, downloaded Windows package passes 19 native GUI checks 0.3.1 checks; TypeScript 6 compiler API exception and upstream indirect pins documented; draft ID 407495650; public metadata endpoint verified with v0.3.0
Software update checks Development checks no longer require package identity; GitHub request failures can use bounded documentation snapshots with the same package validation Check repairs; independent native development IPC and local fallback tests; deployment observation remains separate
0.3.0 local repairs Cost cards stay within bounds, settings use styled controls, eight native Codex versions are registered and null compaction carries are handled; all 332 Codex files in the old DB copy recover without losing original usage 0.3.0 checks; Windows local build, synthetic update replacement and native data-copy checks are separate from remote release/platform acceptance
Software updates Local implementation adds launch/daily/weekly/manual checks, optional automatic downloads, shared progress and explicit installation; native Windows portable replacement verified with a synthetic update Update validation; final checks recorded there. Real newer public release, NSIS updater execution and Linux/macOS update GUI acceptance remain separate
Local unified/browser checks verify exit 0: Rust 1,032, frontend 22, scripts 5; 8 platform tests ignored; clean types. Six Claude tests also pass, overlapping unified counts; earlier Edge alert-display/deduplication checks retained VS discovery, Codex, round, Claude; browser not rerun in this repair; existing simulated IPC/three-source/M8 results retained
Windows executable 0.2.1 application release/11 headless cases, local four-call VS check/rescan, Codex old-DB backfill/rescan, Claude new/old-DB correction pass; earlier 19 native WebView2/IPC cases including detail Merge/alerts, 20 startup checks and 8 receiver cases retained VS, Codex, Claude, round, three sources; GUI/IPC not rerun in that repair; synthetic DB/source isolation distinct from native samples
Windows installation 12 NSIS upgrade/rollback/uninstall/reinstall/failure-abort cases pass Lifecycle; owned tasks/startup entries only, other values retained
Linux executable/install Debian workspace 1,004 tests; deb lifecycle and actual AppImage FUSE/GTK/Orca: 9 groups/47 cases pass; earlier 40 checks each at GTK scales 1/2 retained separately Three sources, lifecycle, ten-language Orca; actual read-only mounts/release on exit checked
Native credentials Two Windows sets of 100 parallel rounds after handling confirmation; six isolated Linux D-Bus/keyring cases and two macOS CI Keychain/HTTP cases pass Credentials, CI; earlier Windows revoke anomaly remains unexplained, first failure/residual checks retained
Batch remote CI/downloads Two runs, eight successful jobs each; three archives match API digests/CRC, four packages match SHA-256/size/report revision CI 37486581699, source 61223e591815a4369a85a00fa23ff1ab2819d6d5; details in CI record
Portable release 0.2.2 Six native Windows/Linux/macOS x64/arm64 .tar.zst archives; all 12 tag jobs pass after artifact-upload recovery. Six extracted headless checks, both Linux GUI checks and local Windows x64 19 native GUI checks pass; 13 uploaded assets match sizes/digests and overwrite succeeds Tag CI 37761919575, source 39e74d09fa2b4e300dca3f7d0ba2edf4b6e95ba2; CI record. No new installer lifecycle, Windows arm64/macOS GUI or signing/notarization acceptance
Scale/resources Uncached million/ten-million query P95 47.05/133.86 ms; million-event GUI import peak 371.09 MiB; ten-minute idle mean/peak 299.90/363.30 MiB meet revised limits Scale, incremental/native; development machine, synthetic sources, whole-process-tree private bytes; original measurements/limit changes retained

Local Windows/Linux packages and CI downloads are checked separately; CI archive integrity does not establish local installation. The 2026-10-07 application round retained user-reported main merge/signing/notarization/Release completion without independently reverifying those external results. The 2026-10-08 portable draft publication is verified separately above; it includes no new signing/notarization. Later record-only commits are not another source CI run.

Source and feature records

Scope Records
Core storage/source identity/aggregate exchange M1, M1a
Complete normalized detail Merge, alerts, Zed external provider Round; transactional conflict/replay, single currency/unknown/deduplication and Zed 1.22.0 two nonempty model samples separately; endpoint is Coding Plan
Initial Codex/Pi/OMP/Claude/Gemini/Qwen reads/version selection Codex, M2 recovery, version directories, CLI install scope; initial sample state retained
Missing current-day Codex collection Rotating reads/old-DB recovery; large history no longer postpones unvisited files; local per-record checks/rescan; later exact-version and creation-fingerprint corrections in 0.3.2 checks
Claude Code 2.1.197 Domestic download/native samples; Zhipu two-model main loop, per-row version, block deduplication, initialized-zero unknown and old DB; Anthropic/other cases require separate checks
Qwen 0.25.0 native/SDK and OpenCode 1.18.34 Container sources; choose per-call/native/sealed partitions; title/background coverage separate
Cline SDK/Hermes/OpenClaw schema 24/MiMo/Zoo/DSH Cline, Hermes, OpenClaw, three sources; actual nonempty records and current executable upgrades/rescans checked
Kilo/Kimi/ZCode/CodeBuddy/WorkBuddy M3/M4, Buddy, Kilo health/range; CLI/extension/comparison/trace scope distinct
M8 second batch 18 adapters, historical ten sources, Zed/current container limits; unavailable account/protocol/distribution cases removed from active work; Qoder detection only
Copilot four interfaces/quota/telemetry Review, VS Code, VS samples, VS cross-version discovery, JetBrains inspection, configuration
Dashboard/scheduling/cancellation/concurrency/old-rule recovery Native/scale, queries/background, complete old summaries, source rules
Costs/online refresh/current API references/archive corrections Engine, refresh, references, archives; native occurrence-time estimates still require known channels

Acceptance scope

  • Native source results apply only to tested versions/formats/nonempty scenarios. Preserve unknown fields, unsaved calls and cumulative intervals without zero filling or expanded product claims. See adapter matrix.
  • DPI 144/UI Automation names, GTK scales 1/2 and Orca ten-language five-page navigation are independently valid. ALSA null does not verify physical audio/all controls; control names do not establish complete Narrator/NVDA use.
  • Linux containers use normal application users/default seccomp, no network/host mounts; FUSE adds SYS_ADMIN explicitly only to owned containers. Results do not verify host login/logout/full desktop/other distributions.
  • macOS native credential/build CI passes; desktop/specific hardware not required this round.
  • Synthetic scale/GUI import does not verify other native sources. App scheduling is not OS wake; warm empty pages do not establish fresh-WebView minimum overhead; cooperative interruption cannot force-cancel OS-blocked reads.
  • Observed WebView requests without external HTTP are not whole-process outbound auditing; receiver/credential checks do not verify other real exporters.
  • User removed more DPI/full screen-reader/host login/logout/OS-wake work. Environment-blocked source extensions/F1/system investigation conditions remain in round records; removal is not passing.

Temporary logs/check DBs/raw records stay in ignored task directories under root build/. Public records contain minimal redacted results; owned receiver credentials, disposable keyrings and daemons cleaned per acceptance run. User-requested Zed provider keys remain in Windows credential storage, absent from normal settings.json.