Publish first usable Windows release instructions and validation

README.md

@@ -79,12 +79,15 @@ autoplays when you select a result.
 
 ## Get searchi
 
-**searchi is in active development. Ready-to-install downloads have not been
-published yet.** Windows, macOS and Linux are the intended desktop platforms;
-packaged releases and platform testing are still being completed.
+**The first Windows x64 version is available.**
 
-Check the [releases page](https://tea.karlcom.de/karl/searchi/releases) for
-published downloads, or let your local AI agent handle setup for you.
+- [Download the Windows installer](https://tea.karlcom.de/karl/searchi/releases/download/v0.1.0/searchi_0.1.0_x64-setup.exe) — about 99 MB.
+- [Download the portable ZIP](https://tea.karlcom.de/karl/searchi/releases/download/v0.1.0/searchi-0.1.0-windows-x64.zip) — about 145 MB. Extract the entire folder and open `searchi-desktop.exe`.
+- [Checksums](https://tea.karlcom.de/karl/searchi/releases/download/v0.1.0/SHA256SUMS.txt) · [Release notes](https://tea.karlcom.de/karl/searchi/releases/tag/v0.1.0).
+
+The installer creates a launcher and sets up WebView2 if needed. This initial
+release is unsigned. macOS and Linux packages follow; source build instructions
+are available now. You can also let your local AI agent handle setup.
 
 ### Install with your agent — copy and paste
 
@@ -95,7 +98,8 @@ Install searchi on this computer and get it ready for my first search.
 Official repository: https://tea.karlcom.de/karl/searchi
 
 Detect my operating system and processor architecture. Use a suitable official
-release if one is available, verifying any published checksums. Otherwise,
+release from the README or releases page, including the initial Windows release,
+verifying its published checksums. Otherwise,
 set up the current source version using the repository's installation/build
 instructions. Handle the required tools and dependencies for me; don't just
 give me a list of commands to run. Keep the installation in a dedicated folder
@@ -119,10 +123,10 @@ search followed by a search in my own words.
 ```
 
 Your agent needs internet access and permission to install software locally.
-Since installers aren't published yet, setup currently involves building the
-app and may take a while. The local search model downloads approximately
-**865 MB**. Windows, macOS and Linux release validation is still in progress;
-your agent should report any platform limitation it encounters.
+Windows x64 setup can use the published installer; other platforms currently
+build from source. The local search model downloads approximately **865 MB**
+once, and cached assets are reused. Your agent should report any platform
+limitation it encounters.
 
 Prefer to set it up yourself? See the
 [build and launch instructions](docs/BUILDING.md).

docs/BUILDING.md

@@ -87,7 +87,8 @@ cargo build --manifest-path apps/desktop/src-tauri/Cargo.toml --release --featur
 ```
 
 The crate rejects Release compilation without this feature; the Tauri build
-CLI enables it automatically. A development-server URL is not a release gate.
+CLI also needs `--features custom-protocol` explicitly for this crate. A
+development-server URL is not a release gate.
 Record native window interaction separately from process startup, browser
 screenshots and worker inference; those are different release gates.
 

docs/DEPENDENCY-REVIEW.md

@@ -9,7 +9,8 @@ and native desktop validation remain pending; the broad security gate stays open
 
 Recorded 2026-10-13 against commit `4e60038c2f3b0a7b1d2dd0f4361309a735ee140f`.
 This is an advisory inventory and dependency-path review, not a complete security
-assessment or a cleared production release gate. No dependencies were changed.
+assessment or a cleared production release gate. The initial inventory changed
+no dependencies; the subsequent compatible backport is documented below.
 
 ## Method and reproducible inputs
 

docs/MEDIA-QUALITY.md

@@ -92,6 +92,41 @@ Measured hybrid p95 was **146.83 ms** on this run. [Recorded output](../fixtures
 
 ## Scoring and release gates
 
+### Local recognized-speech supplement: actual Google diagnostic
+
+A subsequent internal experiment transcribed the unchanged NASA, piano, violin,
+bark and roaring-effect sources locally, then ranked actual recognized passages
+with Google EmbeddingGemma 2 on CPU, one thread. The base candidate supplied all
+30 valid segments from all five sources; music descriptions and hallucinated
+words remained competitors. Exact cosine ranking over source-preserving Google
+tokenizer windows, with a maximum per original segment, found the right file for
+**8/8** frozen NASA queries, but achieved **4/8 strict timestamp start hits** and
+**6/8 segment-overlap hits**. The Moon statement was merged into a segment starting
+at zero; the orange-soil exchange was missing. Funding and Mars passage starts
+passed their independent intervals.
+
+This is a direct recognized-text retrieval diagnostic, **not the production
+ANN/hybrid raw-audio pipeline**. The older production 0/8 point-hit result and
+these 4/8 hits are not equivalent head-to-head measurements and cannot establish
+a deployed improvement. The tiny paired English/German query set addresses four
+topics in one English recording; German speech and general ASR quality remain
+unverified. No model default, calibrated confidence or improved rating follows.
+
+The small candidate's Google diagnostic completed but excluded two invalid ASR
+timestamp windows (piano tail and bark), marking
+`quality_score_eligible=false`. Its reduced competitor pool receives **no
+accepted numeric quality score**. The base score report SHA is
+`8f92c8eeed41db7e78be0e188500e441d53b4fa49fa8c0353a77df74643ce895`;
+actual Google ranking reports are SHA
+`2fbc860f992c3ec561f54137e30744b00265455792e360010ab1fc78ab25173a`
+(base) and
+`aca99d36b3bdf013b1a1074c33f7232de1c1b79af62d00236d53267ce510c9c1`
+(small). Raw recognized text remains ignored. See
+[speech alignment evidence](SPEECH-ALIGNMENT.md) for worker/model provenance,
+corrected decoder validation, measured transcription cost and explicit failures.
+Production integration, speech gating/alignment and an equivalent pipeline rerun
+remain open; the experimental media self-rating stays **4/10**.
+
 Report recall@1/@5, MRR, language/type breakdowns, segment-overlap hits, precise timestamp point hits, and failure categories. No-answer return rate is distinct from positive recall. Do not invent an arbitrary cosine rejection threshold from these results; calibrate on a separate development set, then validate on held-out negatives.
 
 Self-rating rubric is declared before scores: **1–2/10** broken core retrieval; **3–4/10** experimental retrieval with missing essential capabilities or unvalidated abstention/metadata; **5–6/10** useful limited beta after independent basic retrieval/filter/metadata checks; **7–8/10** strong pilot with high recall and timestamp quality on representative corpora; **9–10/10** requires production release evidence across all operating systems, large representative libraries, calibrated no-answer behavior, and complete lifecycle/security/resource verification. A small diagnostic alone cannot justify a production score.

docs/ROADMAP.md

@@ -2,6 +2,11 @@
 
 Status: implementation started; no production release yet. Windows VM tests use CPU only.
 
+Karl's release priority, 2026-10-13: publish a usable first Windows version and
+install it on DEV-WIN11 now. Extensive platform/GPU, clean-machine, signing and
+large-library checks are deferred follow-up, not blockers for this initial
+release. Keep their evidence and unchecked production gates intact.
+
 ## Required product scope (user clarification)
 
 - Desktop UI is primary; semantic setup/indexing/search must require no terminal.

docs/SPEECH-ALIGNMENT.md

@@ -174,9 +174,10 @@ transcripts, worker errors and inferred content never become executable instruct
 - [ ] Validate normal package dependency closure and missing-driver CPU fallback
   on each native OS; execute physical GPU tests outside this Windows VM.
 
-Next implementation slice: isolated worker + manifest verification + genuine NASA
-proof. Keep the initial runtime experiment separate from schema/ranking changes
-until it establishes that recognized segments actually fix the measured problem.
+The runtime and genuine retrieval diagnostics below inform future work; they do
+not add a production speech feature. Further ASR/VAD research is deferred while
+the usable Windows desktop takes priority. Revisit gating/alignment before schema
+or ranking changes that would expose unreliable recognized speech to users.
 
 ## Actual internal CPU diagnostic, 2026-10-13
 
@@ -221,7 +222,150 @@ timestamp output. They do not establish speech recognition quality.
 Historical raw/checkpoint hashes and invalid classification remain under ignored
 `.searchi/speech/invalid-truncated-audio.json`; model bytes and actual raw outputs
 are excluded from Git. The owned sequential session 20331 is terminal with exit 1
-following intentional invalid-experiment termination. Corrected base/small
-inference is paused pending the coordinator's CPU slot, followed by independent
-caption review and genuine Google transcript ranking. Neither candidate is an
-accepted product default; the media quality rating remains 4/10.
+following intentional invalid-experiment termination. A subsequent corrected run
+also detected a source-timeline issue: NASA's container ends at 85.333333 seconds,
+but its primary audio ends at 85.262 seconds. Session 6691 failed closed on the
+72 ms tail discrepancy, preserving its completed checkpoint. The harness now
+uses the actual primary audio stream's duration/start/index, absolute seeking
+with `-seek_timestamp 1`, and explicit stream mapping. Missing stream duration
+is an actionable error; container duration is never silently substituted.
+
+Twelve tests passed in 1.937 seconds after that correction, including actual
+FFmpeg mux/decode with a longer video, audio offset 0.125 seconds and container
+offset five seconds. The audio timeline was 5.125–36.125 seconds; its one-second
+tail decoded 15997 samples, within the unchanged one-millisecond rounding bound.
+The zero-start 44.1/48 kHz tests still decoded exactly 480000/16000 samples.
+
+## Corrected base candidate evidence
+
+The corrected base evaluation completed all five assets with zero timestamp or
+sample-coverage validation failures. Report SHA:
+`9e33984afac9480452f506070dee0156fcc8d7f0d6882c8f5feeeaf5cae57de2`.
+Executed frozen harness SHA:
+`16e29bb70d3e1eeaad1f5b8ec223bff1a6f359215bcf8dabe52512a451855346`.
+
+| Source | Windows | Recognized segments | ASR process seconds | Decode seconds |
+| --- | ---: | ---: | ---: | ---: |
+| NASA | 3 | 19 | 23.322 | 0.542 |
+| Piano | 9 | 9 | 210.720 | 2.167 |
+| Violin | 1 | 1 | 4.471 | 0.212 |
+| Bark | 1 | 0 | 4.373 | 0.113 |
+| Roaring effect | 1 | 1 | 4.732 | 0.111 |
+
+These are observed serial CPU process times including model startup, not warmed
+throughput or a controlled hardware comparison. NASA's decoded windows contained
+480000, 480000 and 404192 samples, covering the actual primary audio range.
+Manual comparison with publisher captions finds the Artemis funding and Mars
+statements recognized, but the initial Moon-choice statement is merged into a
+segment starting at zero and the orange-soil exchange is omitted. Recognition
+does not prove Google ranking or a precise timestamp success rate.
+
+The instrumental piano emitted an unrelated water phrase and repetitive numbers
+or punctuation in several windows. Music descriptions also appear. These must
+remain competing retrieved candidates in the next diagnostic; filtering them
+by knowledge of labels would hide failure. The source's published instrumental
+description supports a hallucination concern, but independent audio listening
+remains open. The current report field `quality_failed=false` means only that
+structural timestamp/sample validation passed; **it does not certify speech or
+search quality**. No model default is accepted.
+
+## Corrected small candidate evidence
+
+Owned sequential session 89200 finished with exit zero after evaluating all five
+assets, base then small. Both used the same frozen corrected harness and verified
+model bytes. The complete small report SHA is
+`1660c95c71c9b8b1cb3b77f8f1ec422fb0954f2aacd2c16ea773901c0cfbe099`.
+
+| Source | Windows | Valid recognized segments | ASR process seconds | Decode seconds |
+| --- | ---: | ---: | ---: | ---: |
+| NASA | 3 | 19 | 86.250 | 0.601 |
+| Piano | 9 | 8 | 1176.426 | 2.399 |
+| Violin | 1 | 1 | 16.045 | 0.162 |
+| Bark | 1 | 0 | 16.701 | 0.112 |
+| Roaring effect | 1 | 1 | 25.082 | 0.113 |
+
+Small emitted repetitive inferred words in non-English languages on the
+instrumental piano, with substantial CPU cost. Two actual timestamp outputs
+failed validation: the 14.454-second piano tail returned a 0–30-second segment,
+and the 2.039-second bark window returned a 0–2.5-second segment. Decoded sample
+counts were correct (231254 and 32618 respectively); these are ASR timestamp
+failures, not the historical resampling bug. Raw JSON is retained; no timestamp
+was clamped. The report's `complete=true` establishes complete evaluation,
+while `quality_failed=true` records failed structural validation.
+
+NASA's small recognition retained funding and Mars passages but misrecognized
+the orange-soil words and still produced coarse segment boundaries. No numeric
+timestamp retrieval improvement follows from caption inspection. The small
+ranker may retain diagnostic results, but its reduced valid-window pool must be
+`quality_score_eligible=false`; the scorer refuses an accepted numeric quality
+score for incomplete or reduced competitor pools. Thirteen harness regressions,
+including real FFmpeg coverage/timeline checks and this scoring gate, passed in
+the coordinator's verified-FFmpeg run (2.100 seconds).
+
+Final process inspection found no owned ASR/runner processes; the CPU slot was
+released for genuine Google transcript ranking. The current run removed its
+temporary decoded WAV/JSON. An older scratch directory from intentionally stopped
+invalid session 20331 remains ignored: automatic approval review rejected native
+PowerShell recursive cleanup with only “blocked by policy” as its reason. No
+alternate deletion method was attempted.
+
+Neither ASR model is an accepted product default. Production Google transcript integration,
+speech activity gating, tighter alignment, hallucination control, German speech,
+resource scheduling and native distribution still require evidence. Publish raw
+Google media search before completing derived transcription so slow ASR cannot
+block initial semantic readiness. Transcript publication needs its own bounded
+completed-source work and atomic eligibility update. The production media rating
+remains 4/10.
+
+## Actual Google transcript retrieval diagnostic
+
+The coordinator's owned session 2984 completed base and small ranking with exit
+zero. The shared real Google adapter ran on CPU, one thread: document encoding
+used the normal indexing role (2048-token context), then the document process was
+dropped before opening the query role (512 tokens). Both reported
+`google/embeddinggemma-2`, 768 dimensions and `b11461-no-logits-v1`. The diagnostic
+runner was a Debug build with SHA
+`be096c6dfe158b991520c060ca83478bdd0cd5c34d64fab4a2413d8b01de8f15`;
+the actual native Google worker SHA was
+`c01cb3ec13cfca722ee768e8350e1399a8b9a4d14a24a35adc3473a5cfc82a73`.
+
+The base experiment ranked all **30 valid recognized segments from all five
+sources**, preserving music/sound descriptions and hallucinated text as
+competitors. Actual Google tokenizer windows preserved original UTF-8 source
+ranges. Segment ranking used exact cosine and maximum over its token windows;
+the frozen labels did not select or rewrite candidates. The eight unchanged
+English/German NASA queries produced:
+
+| Metric | Result |
+| --- | ---: |
+| Correct file at rank one | 8/8 |
+| Strict timestamp start hit at rank one | 4/8 |
+| Segment overlap at rank one | 6/8 |
+
+Both Moon-choice queries selected the merged segment starting at zero, before
+the labelled 1.551–3.603-second utterance. Both orange-soil queries selected an
+unrelated passage starting at 24.320 seconds because the recognized exchange
+was missing. Funding queries selected 43.920 seconds and Mars queries selected
+60 seconds, within the independently frozen intervals. Correct-file success
+does not establish precise spoken-passage localization.
+
+The complete base Google report SHA is
+`2fbc860f992c3ec561f54137e30744b00265455792e360010ab1fc78ab25173a`;
+the accepted diagnostic scoring artifact SHA is
+`8f92c8eeed41db7e78be0e188500e441d53b4fa49fa8c0353a77df74643ce895`.
+Raw artifacts remain ignored under `.searchi/speech`. This is **direct exact-cosine
+retrieval of recognized text**, not production ANN/hybrid retrieval of raw audio.
+The historical production 0/8 audio-window result is useful motivation but is
+not an equivalent head-to-head experiment; no production improvement or score
+change is claimed. Eight correlated queries about four topics in one English
+video do not validate German speech, held-out relevance or abstention.
+
+The small diagnostic also completed, with report SHA
+`aca99d36b3bdf013b1a1074c33f7232de1c1b79af62d00236d53267ce510c9c1`.
+Its two invalid ASR windows (piano tail and bark) were explicitly excluded:
+`complete=true`, `quality_score_eligible=false`, `excluded_window_count=2`.
+Diagnostic ranks are retained, but **no accepted numeric quality score** is
+published for that reduced competitor pool. Neither model is adopted. Next steps
+are speech gating/alignment proof, representative negative controls, actual
+production integration and an equivalent shared-pipeline rerun; the media rating
+remains 4/10.

docs/SPEECH-VAD.md

@@ -0,0 +1,147 @@
+# Speech gating and timestamp research
+
+Research recorded 2026-10-13 against whisper.cpp **v1.9.5**, exact commit
+`d1be6fde11ac6e0407606b4e42fe72d34add8037`. No VAD inference, alignment
+experiment, product integration or default change has been performed. The
+corrected ungated ASR observations remain in [speech alignment](SPEECH-ALIGNMENT.md).
+The instrumental piano's large CPU cost and unrelated text motivate testing;
+they do not establish that any particular VAD model fixes the problem.
+
+## Frozen baseline parameters
+
+Use the pinned implementation's defaults first, before looking at new results.
+Record these values explicitly in an experiment manifest rather than trusting
+future upstream defaults:
+
+| VAD option | Pinned default |
+| --- | ---: |
+| Enabled | false |
+| Threshold | 0.5 |
+| Minimum speech | 250 ms |
+| Minimum silence for splitting | 100 ms |
+| Maximum speech | `FLT_MAX` (no practical limit) |
+| Speech padding | 30 ms |
+| Samples overlap | 0.1 seconds |
+
+Pinned CLI options include `--vad`, `--vad-model`, `--vad-threshold`,
+`--vad-min-speech-duration-ms`, `--vad-min-silence-duration-ms`,
+`--vad-max-speech-duration-s`, `--vad-speech-pad-ms`, and
+`--vad-samples-overlap`. Detection is a speech probability decision, not a
+transcription confidence threshold. [Pinned CLI source](https://github.com/ggml-org/whisper.cpp/blob/d1be6fde11ac6e0407606b4e42fe72d34add8037/examples/cli/cli.cpp#L107)
+
+Relevant decoding defaults are temperature 0, temperature increment 0.2,
+entropy threshold 2.4, log-probability threshold -1, no-speech threshold 0.6,
+word timestamp threshold 0.01, unlimited segment character length, and flash
+attention enabled. The greedy/beam default uses five candidates. Preserve the
+ungated experiment's explicit language `auto`, one processor and CPU policy.
+Changing decoding parameters together with VAD would confound the comparison.
+These are inspected source defaults, not a recommended tuned policy.
+
+## Two implementation traps
+
+The integrated VAD path creates `whisper_vad_context_params` using the library
+default: **four threads, GPU disabled**. It does not copy the transcription
+`--threads` value into `n_threads`. Therefore `whisper-cli --threads 1 --vad`
+does **not** prove a one-thread VAD budget. Source evidence is
+`src/whisper.cpp:4588` and `:6826`. The standalone
+`examples/vad-speech-segments/speech.cpp:110` explicitly copies its own CLI
+thread count. Prefer that VAD-only path or a small, separately hashed diagnostic
+wrapper that sets `n_threads=1` and `use_gpu=false` before loading VAD.
+[Pinned library implementation](https://github.com/ggml-org/whisper.cpp/blob/d1be6fde11ac6e0407606b4e42fe72d34add8037/src/whisper.cpp#L4588)
+
+The pinned full-JSON serializer reads raw token-data `t0/t1`; the dedicated
+`whisper_full_get_token_t0/t1` accessors apply the VAD-to-original timeline map.
+Their implementations differ. Segment timestamps use mapped accessors, so valid
+segment offsets do not validate token offsets. Inspect CLI lines 765–789 and
+library lines 8249–8318. Treat integrated-VAD token JSON as unsafe for source
+seeking until a regression proves correct mapping or the diagnostic serializer
+uses mapped accessors while preserving merged UTF-8 tokens.
+
+## Candidate assets, verified without inference
+
+The upstream downloader lists both Silero v5.1.2 and v6.2.0. The pinned README
+uses v6.2.0; choose it as the **first preregistered candidate**, not because of
+NASA results. Both exports were downloaded into ignored
+`.searchi/speech/vad-research/` and locally verified against Hugging Face LFS
+metadata. Repository revision: `9ffd54a1e1ee413ddf265af9913beaf518d1639b`.
+[Pinned export files](https://huggingface.co/ggml-org/whisper-vad/tree/9ffd54a1e1ee413ddf265af9913beaf518d1639b)
+
+| Export | Bytes | SHA-256 |
+| --- | ---: | --- |
+| `ggml-silero-v5.1.2.bin` | 885098 | `29940d98d42b91fbd05ce489f3ecf7c72f0a42f027e4875919a28fb4c04ea2cf` |
+| `ggml-silero-v6.2.0.bin` | 885098 | `2aa269b785eeb53a82983a20501ddf7c1d9c48e33ab63a41391ac6c9f7fb6987` |
+
+Download via `https://huggingface.co/ggml-org/whisper-vad/resolve/`
+followed by the exact revision and filename, never floating `main`.
+The export model card declares MIT. Silero's upstream MIT license was preserved
+from commit `1e261b036686cd0017d500ee96acd1c4ba572a9d`: 1075 bytes,
+SHA-256 `2e63e9a38b6e8fc0c7bc37ce174caca1862870856c6daf5697cfb785e925520b`.
+The card does not provide a reproducible original-weight-to-export commit chain;
+do not claim one. Distribution needs the actual license notice and a reviewed
+manifest, beyond this research cache.
+[Export model card](https://huggingface.co/ggml-org/whisper-vad),
+[upstream license](https://github.com/snakers4/silero-vad/blob/1e261b036686cd0017d500ee96acd1c4ba572a9d/LICENSE)
+
+## Word and token timestamps
+
+`--output-json-full` enables token timestamps and emits token text, probability,
+offsets and `t_dtw`. Tokens are subword units, potentially merged for valid UTF-8;
+they are not guaranteed to be words. `--max-len` and `--split-on-word` change
+segment boundaries but do not prove alignment accuracy. The upstream describes
+word timestamps as experimental.
+[Pinned timestamp documentation](https://github.com/ggml-org/whisper.cpp/blob/d1be6fde11ac6e0407606b4e42fe72d34add8037/README.md#word-level-timestamp-experimental)
+
+`--dtw base` and `--dtw small` select model-specific alignment presets. They
+require `--no-flash-attn`: otherwise this pinned library warns and disables DTW.
+The default DTW memory budget is 128 MiB. Neither token time nor DTW repairs an
+omitted or hallucinated word. Measure ordinary full JSON first, then DTW as a
+separate preregistered experiment; retain process flags and actual enabled-state
+logs. Do not call either forced alignment or a precise word-seeking guarantee.
+
+## Proposed experiment and integration sequence
+
+1. Build a CPU-only **diagnostic** VAD-only worker from the exact pinned source,
+   with one explicit thread. Test timestamp-unit conversion: VAD accessor/example
+   output is centiseconds, convert to milliseconds exactly once. Verify source
+   offsets, padding, overlapping intervals and 30-second boundary behavior.
+2. Freeze the v6.2.0 asset and the default parameter table above before inference.
+   Run detection over every existing speech/music/sound competitor, preserving
+   all decisions and probabilities. Record no-speech completion separately from
+   error; zero detected speech means no ASR process, not a successful transcript.
+3. Feed detected **original contiguous intervals** to unchanged ASR, retaining
+   their absolute source offset. Avoid concatenating disjoint speech intervals
+   for the first experiment; that needs a piecewise timestamp map. Deduplicate
+   padded overlap by source identity/times, never relevance labels. No manually
+   filtered piano phrases, repetitive numbers or other convenient exclusions.
+4. Compare complete ungated and gated pools using the same eight unchanged NASA
+   queries and genuine Google text retrieval. Report VAD/decode/model-load/ASR/
+   tokenization/Google/ANN/publication times separately. Any processing failure
+   invalidates the comparison rather than shrinking the competitor pool.
+5. Freeze an independent media corpus before selecting any production policy:
+   quiet/distant speech, accents, languages, very short utterances, speech over
+   instrumental music, singing, silence, synthetic effects and longer recordings.
+   Use independent listening/transcripts and publisher captions where possible.
+   Measure missed speech, false speech, hallucinated text, recognition accuracy,
+   timestamp error, retrieval, CPU cost and memory. Tune only on calibration;
+   preserve unseen held-out sources and queries.
+
+VAD may classify instruments or singing as speech and may reject quiet words or
+short phrases under the 250-ms default. That can reduce relevant retrieval recall
+even while removing bad candidates. A music speech gate must never disable
+Google's original audio embeddings: music/sound/image/video retrieval stays
+available independently of transcription. Expected efficiency is a hypothesis;
+no local speedup or false-positive rate has been measured.
+
+If later evidence warrants production ASR, make it a resumable, bounded,
+low-priority enrichment stage. Publish verified filename/text/Google media search
+before transcription; retain prepared readers/query caches during enrichment.
+Reuse the shared inference admission budget and yield between bounded source
+intervals. Track parser/source SHA and revision through decoding, recognition
+and publication; reject stale output. Persist provenance and explicit completion
+or no-speech status. Foreground search must remain usable while ASR is pending;
+an already executing encode cannot be claimed preemptible on a one-thread budget.
+
+Open gates: native VAD build, actual one-thread audit, detection/recognition
+comparison, ordinary and DTW timestamp regression, independent held-out quality,
+background starvation/cancellation/crash recovery, packaging/license review,
+native OS and physical GPU fallback. None is closed by this research document.

docs/WINDOWS-FIRST-RELEASE.md

@@ -0,0 +1,125 @@
+# First usable Windows build
+
+Status: first Windows installer and portable package built and locally installed;
+unsigned initial release, not production certification.
+Other-platform and million-record benchmarks are deferred at Karl's request.
+
+The normal Windows x64 distribution includes the same hardware-detecting native
+Google worker, optimized/baseline CPU modules, Vulkan provider and loader, and
+app-local Microsoft C++ runtime. It is not a special CPU distribution. VM
+inference and WebView rendering checks force CPU/software operation. Real GPU
+execution and clean-machine/older-CPU compatibility remain unexecuted.
+
+## Build procedure
+
+The checked-in Tauri bundle setting remains unchanged. An ignored temporary
+override enables NSIS and the silent WebView2 download bootstrapper:
+
+```json
+{"bundle":{"active":true,"windows":{"webviewInstallMode":{"type":"downloadBootstrapper","silent":true}}}}
+```
+
+From `apps/desktop`, with one Cargo job:
+
+```powershell
+$env:CARGO_BUILD_JOBS = '1'
+$env:SEARCHI_FORCE_CPU = '1'
+$env:WEBVIEW2_ADDITIONAL_BROWSER_ARGUMENTS = '--disable-gpu --disable-accelerated-video-decode'
+npm run tauri -- build --features custom-protocol --bundles nsis --config ../../.searchi/qa/windows-first-release-config.json
+```
+
+The explicit `custom-protocol` feature is essential: this CLI invocation did
+not add it automatically, and the first attempt correctly failed the release
+guard before packaging. The successful artifact must contain the embedded
+frontend rather than depend on a development server.
+
+The portable ZIP is prepared from that same executable using
+`python tools/package_windows.py`, which verifies normal CPU/GPU dependency
+inventory, architecture, pinned decoder provenance and file hashes. Models,
+personal settings and indexes are excluded. Managed Google assets download on
+first use; cached, hash-verified assets are reused.
+
+## Evidence to record before publication
+
+- [x] Completed embedded-frontend Release build and source/frontend/executable hashes.
+- [x] Portable ZIP member integrity and pinned decoder startup.
+- [x] Extracted normal worker auto-to-CPU query and baseline-only fallback.
+- [x] Extracted app starts outside the source checkout with isolated data and
+  model cache, and genuine local Google readiness/index generation.
+- [x] Installer bytes/hash and generated resource mapping inspected.
+- [ ] Execute installation/upgrade/uninstallation on a clean Windows machine.
+
+## Executed Windows VM evidence, 2026-10-13
+
+Source commit: `5eeba29`. The embedded frontend bundle is
+`index-CtE-o2qW.js`, SHA-256
+`deb3cdbeb4c4d34f17c0249674b37f7fd5cbdf7a01fef50c76d2eb50870817c5`.
+Release compilation completed in 1m58s after dependencies were built.
+
+| Artifact | Bytes | SHA-256 |
+| --- | ---: | --- |
+| Portable ZIP | 145,419,558 | `dbab40f77902a5183d986222fadd8417aa46f053ee6066d73a06e7f893692060` |
+| Embedded-frontend executable | 23,950,336 | `9f9243e15dc7fc907c44af37e45226a24eed642a32e5dc82959fab92516ff564` |
+| NSIS x64 installer | 98,894,969 | `1494372cfd9b87dd4de38fd347fe0d18f886a51542e2f49218699aa84979bb3d` |
+
+The ZIP contains 39 individually hash-verified application/resource files. No
+models, user settings or indexes are included. The normal worker SHA remains
+`c01cb3ec13cfca722ee768e8350e1399a8b9a4d14a24a35adc3473a5cfc82a73`.
+
+The real extracted worker ran its query role with 512 tokens, `device=auto`
+and a child-only disabled Vulkan provider. It selected CPU, loaded no Vulkan
+module and returned a finite normalized 768-dimensional Google code-query
+vector. Optimized `ggml-cpu-cascadelake.dll` and baseline-only `ggml-cpu-x64.dll`
+both passed, including a Unicode relocation directory. Startup/query/exit took
+1.722s and 1.844s respectively; these are not warm-search or first-UI-input
+benchmarks. The packaged pinned FFprobe helper also started successfully.
+
+The extracted desktop launched outside the source tree with isolated settings,
+data and hardlinked hash-verified Google assets. Two owned text/code fixtures
+were indexed through the shared engine. The app automatically published actual
+Google generation 1 with two chunks and a 6,544-byte ANN blob; publication was
+observed 10.106s after launch. It remained alive for another five seconds. The
+owned application process tree was then deliberately terminated. This proves
+native process/setup/indexing operation, not visual interaction or paint timing.
+
+Ignored raw evidence: `.searchi/qa/windows-first-preview-worker.json`,
+`.searchi/qa/windows-first-preview-baseline.json`, and
+`.searchi/packages/first-preview-qa-Gemäß_日本/result.json`.
+
+The initial NSIS attempt failed after its official ZIP was downloaded and hash
+verified, during tool extraction (`access denied`). A bundle-only retry uses
+the cached compiler and preserves the verified executable bytes with
+`--no-binary-patching`; it does not disable OS security or clear caches.
+That retry completed successfully with official NSIS 3.11. Its generated
+script installs app-local worker/CPU/GPU/decoder resources and downloads the
+Microsoft WebView2 bootstrapper when needed. The produced installer is
+`apps/desktop/src-tauri/target/release/bundle/nsis/searchi_0.1.0_x64-setup.exe`.
+No installer was executed during this evidence run: even a custom destination
+registers uninstall information and creates user shortcuts. User installation
+is handled separately. There is no signing, clean-machine WebView2 download,
+upgrade or uninstall claim.
+
+Native visual interaction, fresh-machine WebView2 acquisition, installer
+upgrade/uninstall, SmartScreen/signing and actual GPU performance remain
+separate checks. A working preview does not close those production release gates.
+
+## Local DEV-WIN11 installation
+
+The installer was executed on the existing Windows 11 development machine.
+Installation with an explicit `C:/Users/karl/Applications/searchi` destination
+returned exit code zero; uninstall registration reports version 0.1.0 at that
+location. The installed executable SHA matches the packaged executable above.
+
+An explicit destination avoids this Codex MSIX host's AppData filesystem
+virtualization. Existing local searchi data was preserved in an independent
+`C:/Users/karl/.searchi/desktop` profile; both genuine Google assets were
+hash-verified and hardlinked into the independent managed cache. No user folder
+was selected automatically. The desktop launcher sets only process-local CPU
+and software-WebView flags for this VM; the distributed package remains the
+normal hardware-detecting build.
+
+The installed app was started with a native `searchi` window (PID 7716, window
+handle 8324222) and its packaged query worker (PID 19864), using one CPU thread
+and the independent cache. A desktop shortcut was created. This demonstrates
+local installation/startup and package selection, not a clean-machine test or
+native visual interaction audit. The app is left running for Karl to use.

tools/speech-eval/README.md

@@ -25,6 +25,13 @@ model load, not retained-session throughput. This deliberately simple experiment
 does not implement overlap/dedup or production budgets. Segment JSON offsets are
 already milliseconds and get the window offset added exactly once.
 
+Windows follow the probed primary audio stream's start/duration/index, with
+explicit absolute `-seek_timestamp 1` seeking and stream mapping. Missing audio
+stream duration is rejected with an actionable error. Container duration cannot
+silently extend a shorter audio stream. Resampling occurs before sample trimming;
+decoded frames must match the planned duration within at most one millisecond.
+Every actual frame count and full probe JSON is retained in the report.
+
 Each completed window writes an incomplete checkpoint; the final report marks
 complete evaluation, with `quality_failed` and explicit `validation_failures` for
 invalid ASR timestamps. Bounded raw ASR JSON remains in each checkpoint before
@@ -42,6 +49,8 @@ the existing eight frozen NASA intervals unchanged. Input has `fixture_sha256`,
 `asr_report_sha256`, `google_worker_sha256`, and `rankings` mapping query ids to
 ordered `{asset,start_ms,end_ms}` records. It reports top1 overlap and strict start
 point hits; it never chooses a segment by looking at the answer intervals.
+The ranking report must have `complete=true`, `quality_score_eligible=true` and
+no excluded windows; a reduced competitor pool cannot receive an accepted score.
 Hash fields must be checked against retained artifacts by the evaluator; they are
 provenance declarations, not a replacement for inspecting those artifacts.