What changed, when it was released, and what is still being prepared.
This website package
v0.1.0 · Build 541
This page belongs to this website package. The version button inside your app identifies the app you have loaded.
Technical version details
Source: 382c63867ce7d2d63bd4601e2fd16bea73101fe8
Not released
Being prepared for production
These changes are not listed as a completed production release.
October 8 public repair library prepared for DEV: nine original AI-assisted guides include faucet aerator, cabinet hinge and toilet flapper checks, symptom decisions, six bundled schematic illustrations, source-check dates, related reading and category navigation. Corrected towel fold advice for excessive height. Guides stay free, ad-free and readable without login or JavaScript; no professional review, field testing or AdSense approval is claimed. DEV deployment/QA pending; production526, native27, free beta and model/payment settings unchanged.
October 6 DEV539 live: step pictures restored beside two AI answers, with publisher links as optional extras. One fresh synthetic mug-washing guide returned both answers and all three saved pictures; Next/Previous passed desktop and390x844 checks. Restricted cache inspection verified three ready frames, one bridge submission each and matching saved-file digests. This fixes the link-only regression, not all image reliability or object consistency. Exact package/rollback and QA in the technical release log. Production526, native27, existing providers and free beta unchanged.
October 6 DEV picture restoration prepared: restore step-picture requests and one carousel alongside two AI answers, retaining publisher links as optional extras. Normal results and eligible sandbox plans also request pictures again. Exact ordered, language-specific beta cache identity is independent of the experimental image model; unconfirmed old purchases block new-key retries. Completed cache stays available when the job journal is full. Providers, review, free beta, production526 and native27 unchanged. Deployment and actual-image browser checks pending.
October 6 DEV537 live: two real AI answers and the correct Epson ET-2850 illustrated manufacturer guide passed desktop and phone-sized checks without any picture-generation request or console error. Wrong-brand questions did not inherit that guide. Seven fixed English source links are available; unknown products retain their answer with a no-match notice. This is a link-only trial, not dynamic search or inline pictures. Home/plans describe it accurately. Durable picture jobs stop automatic repurchase after an uncertain outcome. Production526, native27 and free beta unchanged; exact receipt in the technical release log.
October 6 DEV536 source trial live check returned both AI answers and opened the correct Epson ET-2850 illustrated PDF without a picture-generation call; HP and mixed-brand questions did not inherit it. Follow-up offer copy now describes two answers and external source links instead of promising generated pictures. Free-beta and entitlements unchanged. Final follow-up activation/QA pending; production526 and native27 unchanged.
October 6 source-guide DEV trial prepared: a seven-source publisher-link index replaces automatic picture requests in the candidate website. It checks submitted task/product identity, labels selectors and unknown matches honestly, and keeps both AI answers first. These are external English source links, not inline pictures or dynamic search. A durable legacy media journal prevents automatic repurchase after an uncertain result; Sunburst customer activation remains off. DEV activation/live QA pending. Production526, native27 and free beta unchanged.
October 6 trial clarification: the isolated client recorded a provider-network/response failure while OpenRouter recorded a completed image. A timeout specifically is not established; the earlier wording was too specific. Reconciled trial cost remains $0.385565. No live site changed and picture acceptance remains incomplete.
October 6 Sunburst synthetic DEV trial concluded, not released: only the authorized image model was added to the shared allowlist; privacy switches and spending limits stayed unchanged. Total reconciled trial cost was $0.385565 under the $1 cap. Step pictures improved, but manual checks found misplaced hardware, the reviewer missed an ambiguous window detail, and one locally timed-out image completed upstream. These are unresolved acceptance failures, not a fixed picture service. No customer content or live cache was used. Prepared website533 is not deployed; DEV website529/local AI531, production526 and native27 remain unchanged.
October 6 structured step pictures prepared, not deployed: Ollama-first scene planning, one action per frame, preceding checked-frame references, exact ordered guide/language cache identity, and separate accuracy-versus-service errors. Only Sunburst was added to the shared model allowlist for the owner's capped synthetic DEV trial; privacy and spending settings are unchanged. All 2,154 gateway tests, build and lint pass. Multi-step and browser acceptance remain pending. DEV website529/local AI531, production526 and native27 remain active.
October 6 DEV529/local AI531 live: English Sarah playback, pause/resume/stop and mobile question/submit layout passed. The drawer guide recovered its missing picture and all four frames were navigable. Picture recovery and single-scene replacement are live, but a fresh screen-tape guide still had a rejected frame after browser retry; complete picture reliability is not claimed. All 2,144 gateway and 1,910 Flutter tests pass. Production526 and native27 are unchanged. This receipt appears in the next immutable website package.
October 6 DEV picture composition prepared: after DEV530 recovered the drawer picture, a fresh screen-tape guide still had one rejected frame. The existing replacement now uses one positive, clear instructional scene, retaining the original action and independent safety review. An isolated replacement passed review and visual inspection. No new model or extra image attempts; live browser recovery is pending. Website DEV529, production526 and native27 are unchanged.
October 6 DEV picture follow-up prepared: rejected candidates supply bounded visual corrections to the existing replacement attempt; transient review outages retry the same bytes once without buying another image or bypassing review. All 2,143 gateway tests pass. A synthetic drawer-tightening replacement passed all six checks and visual inspection. Live DEV activation and browser retry remain pending. Website DEV529, production526 and native27 are unchanged.
October 6 DEV preparation: English voice switches from unavailable Sienna to the existing free Sarah voice, with no paid fallback. Step-picture prompts now avoid unrelated tools and show the relevant object boundaries; visual and safety review stays enabled. Local Proxmox and cached-health routing survive subsequent DEV releases. Real English audio and a fresh reviewed towel frame passed isolated tests; live DEV browser QA remains pending. Production526 and native27 are unchanged.
October 5 production deployment: Website Build526 is live from the exact approved DEV artifact, with two AI answers and local Proxmox inference routing. Premium remains free for a limited time. Live production returned both answers and the fresh mobile question/submit layout passed. A towel guide's pictures failed quality review, and English Sienna remains unavailable upstream; neither is reported fixed. Independent monitoring was switched to cached real Proxmox AI checks. Native iOS/Android remain Build27. This operational receipt will appear with the next website package; existing artifacts remain immutable.
October 5 Proxmox AI and two-answer comparison prepared: the local DEV candidate returns real text and reviewed pictures without Cloudflare inference; Chinese voice passed, while English Sienna remains unavailable on both old and new routes. Two-answer generation, review, display counts and plan copy are updated, with free Premium preserved and persistent local abuse counters. Website traffic has not switched and production is unchanged; exact DEV acceptance and production promotion remain pending. Native apps remain Build27.
October 1 DEV picture retry improvement activated: a short image-review rate limit retries the same picture once within the existing deadline, without buying another image or skipping review. Long provider waits remain respected. Tests pass; a fresh DEV window answer displayed all four cached step pictures. The owner's exact failed guide is not yet verified. Website packages unchanged: DEV517/production506.
October 1 comparison service restored: the preceding picture deployment omitted existing comparison bindings. The default deployment now preserves both bindings and checks them before deployment. A fresh DEV window-installation question returned all three answers and displayed its first picture. Shared infrastructure restored; website packages remain DEV517/production506 and native apps27. This receipt will appear with the next website package.
October 1 DEV image backend correction activated: the same window-measurement guide now displays all four step pictures after retry, retaining three AI answers. Width, height and depth frames were visually checked. The general DEV brief now permits native measuring-tool graduations and one visible subaction per multi-action caption, while keeping object, geometry and safety review. One candidate needed the existing fallback. Gateway tests/build/lint/dry run passed; website packages remain DEV517/production506 and native apps27. This receipt will appear with the next website package.
October 1 DEV image backend correction prepared: live window-measurement picture reviews repeatedly rejected visible writing. The general DEV image brief now permits native measuring-tool scales and illustrates one visible action from a multi-action step, with matching reviewer instructions. Added captions, foreign writing, logos and gibberish remain disallowed; action, geometry and safety checks remain. Focused gateway tests, build and lint pass; deployment and live QA are pending. Website packages remain DEV517 and production506; native apps27. This backend note will appear with the next website package.
October 1 Website Build517 prepared for DEV: a fresh screen-mesh guide returned all three AI answers and three of four reviewed Recraft step pictures. The missing frame stayed unavailable after a DEV-only Recraft-then-Gemini fallback; neither candidate passed the unchanged independent visual review. The website backend also repeated the whole failed gateway call, adding cost and delay. This candidate skips that duplicate retry only for completed DEV gateway 503s while retaining a bounded retry for local bridge transport failures. The release-history validator now retains an existing longer DEV note without editing it. Production's picture model and retry behavior remain unchanged. Gateway, media, website-visual, management and preview-deployment tests pass; exact Build517 staging, activation and live picture QA are pending. Production remains Website Build506; iOS and Android remain Build27.
October 1 Website Build511 DEV activation: exact source 0eec0ba5985b76096149bd5021cc7e450f455f90 is live on DEV with public provenance SHA-256 ea20f147b00b5184cab812e4cdb993c0c833baad71b3a702194a73dd52e868e3 and web tree SHA-256 6a0f97a103f0bfaf580918e601d5a1222c66942f1fdf6abe386dd64f5369d18f. Pinned CT159 staging passed 149 intake, 99 management and 91 visual tests; full gateway and Flutter suites passed 2,126 and 1,910 tests. New signed-in bifold-door and jacket-zipper questions each returned three independently reviewed, model-generated repair answers. New pictures remain blocked: OpenRouter returned HTTP 402 and its refreshed balance was $0.91, below the provider's $1 image pre-authorization threshold. The DEV picture path now stops a rejected frame without a second attempt, though other parallel frames may already be in flight, and displays only independently reviewed, hash-checked ready frames. Successful new-picture output is not verified. No credit, billing, production or native release change; production apex/www remain Website Build506 and native apps Build27.
October 1 Website Build511 prepared for DEV: a live image request returned OpenRouter HTTP 402 while the signed-in account showed $0.91 available, below the provider's $1 image pre-authorization threshold. This is not a question-specific rule. The DEV picture path now stops a rejected frame without its second attempt, though other parallel frames may already be in flight, and can display only independently reviewed, hash-checked frames already completed when another frame is pending or failed. Unreviewed images stay hidden; the production inference entrypoint retains its prior behavior. Gateway and focused Python/Flutter tests pass; full Flutter and exact DEV deployment checks remain pending. No credits or billing settings changed. DEV remains Build509, production Build506, and native apps Build27.
October 1 Website Build509 DEV activation: exact source e445599713b3f81525ebc57139651e84840026f9 is live on DEV with public provenance SHA-256 53f3846011b143923b3626aff8235670baf9b0c4fef96a794c2a03388091ea4b and web tree SHA-256 c532bb3cbe91b9348ebe4d25f7ee5f28d43f8c1db4beba4dd7d9cd0e9ca9ba81. Pinned CT159 staging passed 149 intake, 99 management and 88 visual tests. Signed-in screen-door and dresser-drawer questions each returned three dynamically generated, independently reviewed AI answers. The light-switch question returned two reviewed answers and one safety-withheld answer, not a model failure. A live spare-model activation was not observed, though bounded tests passed. Separately, picture calls failed with provider HTTP 402 while OpenRouter showed $0.91 available, below its $1 image pre-authorization threshold. Pictures are not certified working; production remains Website Build506 and native apps Build27.
September 30 Website Build509 prepared for DEV: one slow or failed three-AI comparison slot can hedge to a fourth distinct Ollama model within the same bounded deadline; no question-specific answer is added. The first valid draft still passes both independent safety reviewers before display, and model names remain hidden. Promptly completed requests make no spare call; partial availability remains honest when no reviewed draft arrives. The production-origin path is unchanged. All 2,122 gateway tests, lint, build and independent diff review pass. DEV activation and live fallback QA are pending. Production remains Website Build506; iOS and Android remain Build27.
September 30 Website Build508 DEV activation: exact source 15694dcc3f3b0ff78c298a7008a5a7fd42dad61c is live on DEV with public provenance SHA-256 d43b22a5238606ca64ad46d83fd754e220807695e797c9b29bf914bf90332b98 and web tree SHA-256 b2394013bf5870554300c1da1bbd1c6ac818fbfeffeb94f6d00444c67d1634d4. Immutable staging passed 149 intake, 99 operations and 88 visual tests. Signed-in fitted-sheet, cabinet-door and Wi-Fi questions returned three dynamically generated and independently reviewed answers with loaded step pictures. A light-switch question returned two reviewed answers and fixed safety diagrams; the third Ollama model failed during generation, not review. A general fallback is being prepared. Production remains Website Build506; iOS and Android remain Build27.
September 30 Website Build508 prepared for DEV: all three independently generated AI answers and eligible conceptual picture plans receive a separate two-model Ollama safety review, not a one-off rule for each question. Each slot is approved, rejected or marked review-unavailable separately; an unreviewed draft is never displayed. The website explains reviewer outages and offers retry. Local gateway and Flutter suites pass; live reviewer latency, DEV package activation and signed-in browser QA are pending. DEV remains Website Build507, production Website Build506, and native apps Build27.
September 30 Website Build507 DEV activation: source 704fab6ecc248e6c287bc2e6452fd598a6fab4be is live on DEV with public provenance SHA-256 96c2f6f8e44bf23accf01d79ad24ff7caafbd4bbf3eaac198cc8789acf10bc70. Signed-in fitted-sheet and cabinet-door checks returned three AI answers and step pictures. A light-switch check showed two fixed safety diagrams but conservatively withheld three AI drafts; Build508 addresses that over-withholding dynamically. Production remains Website Build506; iOS and Android remain Build27.
September 30 Website Build506 prepared for DEV: exact Build505 DEV light-switch and bare-dimmer checks each returned two fixed safety diagrams and withheld three unsafe AI replies. Phone-width review found that the breaker-OFF instruction and panel warning can scroll out of view on step two. This candidate places a visible, screen-reader-labeled breaker-OFF reminder inside both diagram cards. No generated electrical photos, account/payment change, or native update. DEV remains Build505 pending exact-candidate visual QA; production remains Build496/native27.
September 30 Website Build505 prepared for DEV: incomplete bare-dimmer answers or follow-up questions can now retain only the fixed breaker-OFF and closed-plate verification diagrams for eligible English switch/dimmer results. The diagrams do not authorize wiring work or prove a circuit is de-energized. Emergency, non-English and explicit breaker-negating results receive no diagrams; unsupported electrical answers never fall through to generated photos or cached wiring images. Exact-package DEV staging, signed-in switch/dimmer checks and mobile visual QA are pending. Production remains Build496 and native apps remain Build27.
September 30 Website Build503 DEV activation receipt: exact source 039b9964f066aa110e713526eac3d8fe559b5679 is live on DEV with public provenance SHA-256 400a52445fbd13614f369f0095f027319c838da4e84762782f6f7c04bc810a87 and web tree SHA-256 fd9d1ad6379b8a079391bf55d4f7cdd6567a247c1bac2de26aa6890af5944aeb. Immutable staging passed 149 management/intake, 99 operations and 68 visual tests. This is not production approval; production remains Build496 and native apps remain Build27.
September 30 Website Build503 prepared for DEV and production review: live DEV Build500 showed breaker OFF, then closed-plate power verification, and withheld three unsafe light-switch comparison replies. An ordinary fitted-sheet question still returned three answers and pictures. This candidate also protects bare dimmer requests with the fixed electrical safety diagrams instead of generated wiring pictures. Exact Build503 DEV release checks and a separately reviewed production comparison-gateway update remain pending; production is still Build496 and native apps are still Build27.
September 30 Website Build500 prepared for DEV: a deterministic safety gate withholds a full light-switch comparison answer when it treats a silent non-contact tester as proof wires are safe, shows a labeled server-authored fallback, and prevents withheld text or generated photos from being spoken or displayed. The exact two app-drawn Build499 safety diagrams remain available. A newly named DEV-only Worker and preview route isolate first testing from the existing production adapter and shared inference service. Activation, live QA and production approval remain pending. Production stays Website Build496; native apps stay Build27.
September 30 Website Build499 DEV visual check: exact source c5ee9e169bfa3b386c40a788f6e85cb887348147 activated after 149 management/intake, 99 operations and 66 visual staging tests. A fresh signed-in comparison showed breaker OFF then closed-plate verification as app-drawn diagrams, not AI photos, on desktop and a 450-pixel-wide viewport. Its AI answer path still needs the separate Build500 safety gate. Production stays Website Build496; native apps stay Build27.
September 30 Website Build499 prepared for DEV: live Build498 visual review found its breaker photo did not unmistakably show OFF, and its tester photo showed exposed switch wiring. Build499 uses fixed safety diagrams and stops generating or replaying AI light-switch photos, including cached ones. Old signed image links are rejected server-side, though browser-stored bytes cannot be erased. A live AI answer also treated a silent pen tester as proof of safety: the diagram warns against that, but a server-side answer gate remains necessary before production approval. Fresh DEV rendering is pending. Production remains Website Build496; native apps remain Build27.
September 30 Website Build498 DEV visual QA no-go: the exact package activated, a fresh signed-in question returned three AI answers and the UI kept breaker-off then tester captions with a persistent warning. The generated photos themselves were ambiguous and are not approved for production. Build499 prepares deterministic diagrams instead. Production remains Website Build496; native apps remain Build27.
September 30 Website Build498 prepared for DEV: a light-switch answer could say to turn off the breaker while structured picture actions omitted it. This candidate places breaker-off and power-verification steps first, shows a prominent reminder, stops before switch or wire contact imagery, and avoids reusing older electrical pictures. Electrical pictures are unavailable if the written answer lacks the breaker-off prerequisite or uses an unsupported language. DEV deployment and fresh real answer, picture, cache and mobile checks are pending. Production remains Website Build496; iOS and Android remain Build27.
September 30 DEV Build489 live: source bc1cb022b3e4b050b5fabb8c3db3f03e8f59a903 is active as billing release bc1cb022b3e4b050b5fabb8c3db3f03e8f59a903-214a20f3c9b96437e77ff4ca94e9619f6ce9942c55e8236bfbed3f624874c4a7 after 148 staged tests. UTF-8 prompt transport, one global provider slot and one bounded transient retry produced four ready JPEGs (194,453, 198,303, 183,482 and 170,314 bytes) for a fresh signed-in screen-door question; Next, Previous and the final disabled state passed. Nginx and firewall hashes are unchanged. Production stays Build435/native27; the website package, answer-model order, iOS and Android are unchanged.
September 30 DEV Build489 prepared: Build488 passed all 147 staged tests and activated. Its fresh browser run stored reviewed PNGs for steps two through four (446,123, 392,626 and 397,311 bytes), while step one had one transient miss; the exact unchanged step-one replay then succeeded as a reviewed 344,686-byte PNG in 18.15 seconds. DEV now retries exactly once only for bounded rate-limit or unavailable responses, keeps one global provider slot and refreshes the failed cache. Staging, activation and rendered four-picture QA are pending. Production stays Build435/native27; website, answer models, iOS and Android are unchanged.
September 30 DEV Build488 prepared: Build487 passed all staged tests and activated. Its fresh signed-in browser run proved UTF-8 transport works by storing two reviewed PNGs (359,032 and 463,556 bytes), while each simultaneous pair still lost one picture. DEV now combines UTF-8 JSON with one process-wide provider slot and refreshes the partial cache. Four concurrent callers must all finish ready while observed provider concurrency remains one. Staging, activation and rendered four-picture QA are pending. Production stays Build435/native27; website, answer models, iOS and Android are unchanged.
September 30 DEV Build487 prepared: Build486 passed all staged tests and activated, but a same-user `_image` call failed in 0.01 seconds while the identical raw provider request returned a valid reviewed image. The traceback found the real defect: the Unicode picture prompt was passed to Python's HTTP client as text, causing local Latin-1 encoding failure before the provider or file store. DEV now sends explicit UTF-8 JSON bytes, regression-tests the real em dash, retains one shared two-request provider limit across all threads and refreshes failed cache keys. The 12 media tests pass; staging, activation and rendered QA are pending. Production stays Build435/native27; website, answer models, iOS and Android are unchanged.
September 30 DEV Build486 prepared: Build485 passed all 147 staged tests and activated, but a fresh browser run still failed because its one-request serialization did not cover overlapping threaded media requests. The exact current first-step prompt succeeded alone as a reviewed 180,611-byte JPEG in 7.64 seconds. DEV now enforces one shared picture-provider slot across all request threads, keeps each four-step loop serial and refreshes failed cache keys. Staging, activation and rendered QA are pending. Production stays Build435/native27; website, answer models, iOS and Android are unchanged.
September 30 DEV Build485 prepared: Build484 activated cleanly, but a fresh signed-in screen-door question proved two reviewed image requests still fail together. The exact first-step request succeeded alone through the same private bridge and service identity as a 412,275-byte PNG in 9.39 seconds. DEV now serializes the at-most-four step pictures and refreshes failed cache keys. Deploy-stage tests, activation and a rendered four-picture check are pending. Production stays Build435/native27; website, answer models, iOS and Android are unchanged.
September 30 DEV Build484 prepared: the real screen-door result had no structured actions, so its introduction became picture step one, while four simultaneous image jobs all failed. One live bridge request and a two-request batch succeeded; DEV now extracts the four numbered actions, limits generation to two parallel jobs and refreshes failed caches. All 12 media and 52 deployment tests pass. Billing stage/activation and fresh rendered QA remain pending. Production stays Build435/native27; website, gateway and answer models are unchanged.
September 30 DEV Build482 live: source aee7994b is active as gateway 487ad2ce at 100%. Two real private bridge checks returned reviewed JPEGs after the owner added only Gemini 3.1 Flash Image to OpenRouter's allowlist. All 1,982 Worker tests and deployment checks pass. Signed-in DEV displayed four sink-removal pictures and working step navigation. The failed Build481 fallback never reached users. Production stays Build435/native27; no website package or customer question changed.
September 30 DEV Build482 prepared: Build481's FLUX fallback timed out twice and was rolled back before user traffic. Build482 removes it. The owner approved adding only Gemini 3.1 Flash Image to OpenRouter's allowlist; one model became eligible with budgets, privacy controls and existing models unchanged. The private bridge then returned a reviewed 228,685-byte JPEG in 8.8 seconds. Clean gateway activation and repeated verification are pending. Production stays Build435/native27.
September 30 DEV Build481 prepared: the picture bridge returns MEDIA_UNAVAILABLE because its preferred OpenRouter image model is outside the workspace key's allowed route; Workers AI is out of daily free allocation. DEV now falls back to the proven OpenRouter FLUX.2 Klein 4B route with model-compatible parameters, while retaining exact-step, configuration, plausibility and no-writing review. Deployment and a real bridge check are pending. Production stays Build435/native27.
September 30 DEV Build479 live: website answers use Ollama DeepSeek V4.1 Flash first; OpenRouter Gemini 3.6 Flash runs only after a retryable failure, never in a race. Exact DEV-origin and Worker gates isolate it; staging and production are disabled. All 1,982 Worker tests and deployment checks pass. Gateway fe84a23f and adapter 5b2f4f79 are active. A fresh DEV answer used Ollama with no fallback event; forced-failure order is regression-tested. Production stays Build435/native27.
September 29 DEV Build478 prepared: the live structured reviewer correctly rejected both fast-model candidates for wrong action, configuration and geometry. DEV now uses Gemini 3.1 Flash Image at supported 512 and 16:9 settings, without unsupported format metadata. Exact-action, plausible-geometry and no-writing review still fails closed with one retry. All 1,980 Worker tests pass. DEV activation and signed-in mobile sink-picture QA are pending. Production stays Build435/native27.
September 29 DEV Build477 prepared: the OpenRouter vision review now uses Repair Robin's proven strict structured-output contract: an exact five-boolean schema, minimal reasoning, privacy-preserving routing, fallbacks, latency ordering and price ceilings. The fail-closed picture quality decision is unchanged. All 1,980 Worker tests pass. DEV gateway, website and billing activation plus signed-in mobile sink-picture QA are pending. Production stays Build435/native27.
September 29 DEV Build476 prepared: the funded OpenRouter picture reviewer returns bounded multimodal text parts rather than one plain string. The parser now accepts exact text parts within the existing size limit, then applies the unchanged five-key quality decision. Empty, oversized, mixed or unexpected parts still fail closed. All 1,980 Worker tests pass. DEV activation, billing update and rendered sink QA are pending. Production stays Build435/native27.
September 29 DEV Build475 prepared: Build474's bounded trace found OpenRouter created each picture but the separate Cloudflare vision call was unavailable. Picture review now uses DEV's funded OpenRouter vision route and still rejects wrong actions/configurations, mixed steps, implausible geometry or any writing. Prompts and pictures are not logged; deadlines and one retry remain. All 1,979 Worker tests pass. DEV activation and rendered sink QA are pending. Production stays Build435/native27.
September 29 DEV Build474 prepared: Build473 restored fast image creation but its live bridge check still failed inside visual review. The reviewer now emits only a fixed content-free reason code for JSON shape, invalid envelope, timeout or model failure, never prompts, pictures, responses, accounts or credentials. The quality gate is unchanged. All 1,979 Worker tests pass. DEV gateway deployment and a fresh bridge trace are pending. Production stays Build435/native27.
September 29 DEV Build473 prepared: Build472's live bridge check showed that removing OpenRouter latency routing made both image attempts exhaust their windows. The known-fast routing is restored while native Cloudflare verifier-envelope support remains. Content-free phases distinguish provider, decode and visual-review failures without logging customer or provider content. All 1,978 Worker tests pass. DEV activation and rendered sink QA are pending. Production stays Build435/native27.
September 29 DEV Build472 prepared: live Build471 tracing found pictures discarded because Gemma returned Cloudflare's native response envelope. The reviewer now accepts either bounded envelope while retaining fail-closed visual checks. Undocumented image routing metadata is removed, warning/helper prose is excluded from fallback steps, and failed caches refresh. All 1,978 Worker and 145 billing tests pass. DEV activation and rendered sink QA are pending. Production stays Build435/native27.
September 29 DEV Build471 prepared: picture series skip warning-only prose, lead each image request with the exact current action and prevent the final outcome from overriding that step. Failed picture caches refresh. The funded OpenRouter route and strict separate vision review remain. All 1,977 Worker, 1,840 Flutter, 144 billing and 64 release tests pass with static analysis/build/lint. DEV packaging, activation and rendered sink QA are pending. Production stays Build435/native27.
September 29 DEV Build470 prepared: the picture reviewer no longer asks Gemma 4 vision for unsupported JSON mode. It uses Cloudflare's documented image-message shape and still accepts only an exact five-boolean JSON object, with one whole JSON fence allowed. All 1,977 Worker tests pass. The first Build469 gateway made pictures unavailable; its web/billing packages stayed inactive. Corrected DEV deployment and rendered sink QA are pending. Production stays Build435/native27.
September 29 DEV Build469 prepared: a separate vision model reviews each step picture for the exact action, confirmed configuration, mechanical plausibility and zero writing or gibberish in any language. One rejected picture is regenerated; two failures show unavailable instead of caching a bad image. All 1,976 Worker tests pass. Inactive Build468 is superseded; DEV deployment and rendered sink QA are pending. Production stays Build435/native27.
September 29 DEV Build468 prepared: step pictures receive only their matching action plus confirmed follow-up details, preventing mixed repair stages and sink-type changes. Pictures forbid all visible writing in any language, while captions remain outside the image. A new cache generation replaces earlier bad images. Focused regressions pass; deployment and rendered provider QA are pending. Production stays Build435/native27.
September 29 DEV Build466 live: ordinary DIY guidance can receive numbered pictures and how-to references even when legacy metadata recommends a professional; only an explicit active emergency suppresses pictures. A signed-in Max toilet-installation test returned four navigable pictures at 390x844, how-to search, zero video assets and a clean console. All staging checks passed. Production stays Build435/native27; receipt467 is source-only.
September 29 DEV Build464 live: generated solution video is removed; Plus and Max retain four-step picture sequences and reviewed external tutorial links. A signed-in Max test returned four pictures with working navigation at 390x844, no generated-video element/request and a clean console. Legacy video rows stayed unchanged and unreachable. All automated, staging and activation checks passed. Production stays Build435/native27; receipt465 is source-only.
September 29 DEV Build462 live: Plus and Max can generate up to four numbered pictures, one for each written action, with mobile-safe Previous/Next navigation and separate protected caching; Max retains video. A signed-in Max fitted-sheet test returned four matching pictures and video, both navigation directions passed at 390x844, and the console was clean. Safety/professional suppression and plan checks remain. Production stays Build435/native27; receipt463 does not deploy a website.
September 29 DEV Build454 live: Ask and photo help start dictation with one Speak tap, place recognized words in the editable field and let Stop & keep text end capture. The dialog, separate Start action and Use this text step are gone; speech never submits automatically. A 390x844 live check confirmed the one-tap state change and clean console. Automated QA supplied no spoken audio. Production remains Build435/native27; the pending image-Worker timeout change is not claimed live.
September 29 DEV Build454 one-tap voice input prepared: Ask and photo help begin dictation with one Speak tap and put recognized words directly in the existing editable field. Pausing ends capture automatically, or Stop & keep text ends it manually. The old dialog, separate Start action and Use this text step are gone. Speech never submits automatically. Static analysis and all 1,833 Flutter tests pass. DEV deployment and live mobile-browser QA are pending; production remains Build435.
September 29 DEV Build453 picture reliability prepared: two normal Max tests on live Build452 returned all three answers and a playable generated video, but the picture request consistently exhausted its 22-second provider deadline. The correction requests compact JPEG output and uses bounded 80-second Worker, 90-second private proxy and 100-second billing-bridge limits beneath the existing 125-second public ceiling. DEV deployment and fresh media QA are pending; production remains Build435.
September 29 DEV Build452 Max media correction prepared: the normal Max three-answer screen now prepares and displays its server-verified picture and video. A real DEV request returned HTTP404 for the earlier image model; the corrected gateway returned both assets ready and confirmed protected cache reuse. Idempotent media lookups now recover from temporary billing contention. Production remains Build435; immutable DEV activation and normal-Max browser QA are pending.
September 29 DEV Build452 live: the normal Max three-answer screen now requests its server-verified picture and video. All three answers and both media surfaces appeared; the generated video completed and played from protected same-origin storage. Two fresh picture requests failed after about 25 seconds, exposing a fixed 22-second provider deadline rather than a missing UI. Production remained Build435; Build453 prepares the timeout correction.
September 29 DEV Build451 generated solution media prepared: Plus creates and reuses a safe same-origin picture; Max adds a reusable four-second silent video. Both Ask paths use the server-verified Stripe tier and show explicit progress or unavailable states. Failed Build450 activation exposed and rolled back an Nginx route-quoting issue before customer or billing code changed. Provider keys stay server-side; production remains Build435; activation and real provider/browser QA are pending.
September 28 DEV Build448 media access prepared: Free and Basic are text-only, Plus gets approved pictures when available, and Max gets approved pictures and videos when available. DEV billing returns these capabilities with verified Stripe plans; the client fails closed on inconsistent access and saved interest cannot unlock benefits. Shared generation, storage and delivery remain pending, so media is not claimed live. Production website435, backend437, native27 and payments-off are unchanged.
September 28 external alerts verified at 18:26:55 America/New_York: a scheduled cloud run delivered one labeled test to the existing private Telegram chat. Test mode is removed; only new or changed active problems notify. Healthy runs, repeats and recoveries stay silent. The Codex heartbeat is paused and muted. All 137 offline tests pass. Delivery no longer requires this Mac, but Cloudflare or notifier outages can still interrupt alerts. Website435, backend437 and native27 are unchanged.
September 28 problem-only alert delivery in preparation: routine Codex monitoring chatter has been paused while an independent cloud sender is prepared for the existing private Telegram chat. It will send only new or changed active problems, with healthy checks and recoveries silent. One labeled setup test is planned; delivery is not yet verified. Existing cloud checks, website435, customer backend437, native27 and free beta are unchanged.
September 28 monitoring verified at 17:20:16 America/New_York after a runtime fix removed false failures. A scheduled check passed both website shells, operations health, a real primary-AI answer and three model catalog entries; catalog presence does not certify model inference. All 90 offline tests pass. Recovery notified once and unchanged success stayed quiet. Codex alerts still require the Mac awake and running. Website435, backend437, native27, models and payments are unchanged.
September 28 monitor correction prepared: scheduled checks exposed an unsupported Cloudflare HTTP option causing false failure reports while the real primary-AI check passed. A narrow monitor-only correction and runtime regression tests are being verified; deployment and a healthy scheduled result remain pending. Website435, customer backend437, native27, models, free beta and payments-off are unchanged. This source note does not publish a new website package.
September 28 monitoring activation pending: an independent five-minute website, operations and primary-AI monitor plus Codex outage/recovery follow-up are configured; 68 offline tests pass. The cloud schedule had not saved its first real result by 20:46 UTC, so readiness was not claimed. Codex alerts require the Mac awake and running; email/SMS is not connected. Website435, backend437, native27 and payments are unchanged. No website deployment is implied.
September 28 backend437 deployed: website AI answers recovered after replacing an unavailable cloud model. DEV comparison returned all three answers; production repair and how-to questions returned practical steps and completion checks. Guest signup limits remain. Website Build435, iOS/Android27, the free/unlimited beta and payments-off are unchanged. This is targeted backend recovery, not full-site QA or a new website build; the next package publishes these notes.
September 28 AI service recovery prepared: the configured DeepSeek V4 Flash no longer appears in Ollama's current cloud catalog. A website-only DeepSeek V4.1 Flash replacement and private, content-free error diagnostics are being tested. DEV and production activation are not yet claimed. Website Build435, native27, free/unlimited beta and production payments-off are unchanged.
September 22 Build435 prepared: six original, AI-assisted, source-checked public guides address Google's publisher-content rejection. The ad-free library works without sign-in or JavaScript, with practical steps, sources and editorial limits. Home links, sitemap, narrow production indexing and real missing-guide404s are included. DEV QA, production promotion and a new Google review request remain pending. Free/unlimited beta, payments-off and native27 are unchanged; no Google approval claimed.
September 21 Build434 correction in preparation: production browser testing found a JavaScript-only error that prevented the optional advertising choice from appearing. The correction preserves membership and consent protections. Google review remains pending; free/unlimited beta and payments-off are unchanged. Fresh deployment and live ad delivery are not yet claimed.
September 21 Build433 prepared, not deployed: Google EU and US privacy messages are published for Repair Robin only. The candidate adds narrow canonical membership checks and an explicit Allow choice before Google's paused tag setup. This corrects our earlier circular consent-loader restriction; publisher configuration is not Google approval. Review is still Getting ready. Fresh build/live QA remain pending; DEV431, production429, native27, the free beta and payments-off are unchanged.
September 21 DEV431: the towel guide, diagrams and separate ad privacy settings are live. Desktop/mobile navigation, the existing Max test account's ad-free state and saved opt-out passed; no Google ad requests were captured. 382 Flutter, 65 Node and 22 public checks passed. Free/unknown states have widget coverage only. Google ads remain off pending approval, CMP and membership verification. Production429/native27 and the free beta are unchanged; receipt432 is documentation only.
September 21 advertising still pending: website ownership is verified and Google review has been requested, but readiness approval and real ad serving are not enabled. Advertising consent, production paid-membership verification, native SDK integration, public-store linking and paired mobile releases remain required. DEV's membership-gated layout sample is not a Google ad or a production paid entitlement. The limited-time free beta continues and production payments remain off.
September 18 DEV422 live: all 11 languages returned non-silent MP3s and finished playback; English keeps Sienna. The user approved adding only the free multilingual model to the existing allowlist. A Chinese answer read all four passages; repeat/Pause/Resume/Stop and phone-sized/desktop checks passed; console clean. 320 Flutter and 1,883 Worker tests passed. Physical-phone sound/pronunciation unconfirmed. Production349/native27 unchanged; receipt423 is documentation only.
September 18 DEV422 prepared: after continued silent device voices, all ten non-English languages use authenticated generated MP3 audio through the fixed free Fish Audio model, not installed phone voices. English retains Sienna. Settings tests the real cloud path after sign-in; provider processing/retention is disclosed. No paid/English fallback; access and quotas unchanged. Deployment, real audio and physical-phone confirmation remain pending. Production349/native27 unchanged.
September 18 DEV420 live: Chinese voice lists warm before playback; Settings adds Test voice. Premature completion now shows an error, and another tap tries a different matching voice. Live recovery, repeat playback, Pause/Resume and a real Chinese answer passed browser checks; 206 Flutter tests passed. Physical phone/speaker audio unconfirmed. English Sienna, Workers, production349/native27 unchanged. Receipt421 is documentation only.
September 18 DEV420 prepared: DEV419 testing exposed a Chinese voice reporting a full sample finished in under half a second. The correction detects incomplete CJK audio and lets a new tap try another matching voice. No autoplay or English/paid fallback; Sienna is unchanged.419 passed deployment/readiness checks but this repeat-playback defect remains pending420 deployment and live QA. Physical audio unconfirmed; production349/native27 and Workers unchanged.
September 18 DEV419 prepared: Chinese/device playback preloads voices, handles a late voice list with an explicit Play tap, prefers ordinary Mandarin voices and uses shorter passages. Silent end-before-start is a failure, not completion. Settings adds a short Test voice sample for each non-English language and sound-output guidance. Device voice availability still varies; deployment/live QA pending. English Sienna, production349/native27 and Workers unchanged.
September 18 DEV417 live: Settings saves 11 answer/voice languages. Real Spanish answers, inline conceptual pictures and device-voice controls passed; existing answers keep their language. A fitted-sheet video played inside the answer. Desktop and 390/320px checks passed; console clean. Media is topic-dependent, not reviewed equipment photos. Physical-phone audio/photos untested. Production349/native27 unchanged; receipt418 is documentation only.
September 18 DEV417 prepared: retains the language/visual features from416 and corrects a pre-existing streaming-test timing race found on the dev server. Streaming behavior and its before-EOF check are unchanged. Website activation remains pending until staging and browser QA pass; the DEV inference update preserves production/native/speech modules. Production349/native27 unchanged.
September 18 DEV416 prepared: 11 answer/voice languages in Settings. New answers use that language; English keeps Sienna, others use an available matching device voice. Existing answers retain their language; menus and fixed safety notices remain English. Tutorials play inline after a tap; suitable low-risk tasks can include labeled conceptual diagrams. No arbitrary model media URLs or paid voice fallback. Deployment/live QA pending; production349/native27 unchanged.
September 18 DEV414: at 390x600 and 320x600 the Ask field and voice/photo/primary actions fit without scrolling. Desktop keeps its full introduction. Typed text survived resizing, Settings, voice cancel, photo, Feedback and Home/Ask navigation. 170 Flutter and 13 release checks passed; browser console clean. No AI request or physical-phone keyboard tested. Production349/native27, Workers, AI and access/payment behavior unchanged; receipt415 is documentation only.
September 18 DEV414 prepared: the mobile-width Ask page puts the text field first, removes the large hero and repeated beta messages, and keeps compact account, feedback/version and navigation controls. Optional explanations move below actions; privacy guidance and draft preservation remain. Deployment and browser QA pending. Desktop behavior, production349/native27, AI, accounts and payment/backend behavior unchanged; no native rollout.
September 18 DEV412: real guest-Wi-Fi and fitted-sheet questions returned three answers with expanded, matching visual guides. Desktop/phone-sized layouts and task switching passed. Guide taps emitted the expected external-open events; publisher playback remains unverified. Optional search review/cancel and cleanup passed. 129 Flutter and 13 release checks passed. Production349/native27, Workers, AI and access/payment behavior unchanged; receipt413 is documentation only.
September 18 DEV visual guides prepared: common how-to tasks gain direct source-checked video and illustrated-guide links, and guest Wi-Fi gains a network diagram. Available guides start expanded. Other topics offer clearly labeled searches instead of an empty reviewed-media promise. Source applicability, privacy, explicit external taps and hazard controls remain. Deployment and live QA pending; production349/native27 and AI models unchanged.
September 18 DEV410: Ask now supports everyday how-to and repairs. Real fitted-sheet and three-answer guest Wi-Fi setup requests returned useful steps and completion checks without an invented fault or forced pro referral. Phone-sized/desktop Home and Ask checks passed; optional visual search uses a reviewed public phrase. Live photos and physical phones remain unverified. Production349/native27, models and access rules unchanged; receipt411 does not deploy website411.
September 18 DEV how-to guidance prepared: ask about safe everyday tasks, setup, use and maintenance as well as repairs. Clear how-to requests should receive practical steps and a completion check, without an invented fault or forced professional referral. Photo, three-answer and optional visual-search guidance also support how-to. Deployment and real-model browser QA are pending; production349/native27, providers and safety/access controls unchanged.
September 18 DEV408: phone voice review keeps the text box and actions visible, dismisses the previous keyboard and adds Hide keyboard. Review/apply/cancel passed at phone-sized and reduced heights; drafts stay intact. Starting/resuming playback also dismisses edit focus in automated tests. Physical phones, mobile keyboards, Safari and landscape remain unverified. Production349/native27 and AI models unchanged; receipt409 does not deploy website409.
September 18 DEV phone voice fix prepared: voice input dismisses the keyboard before opening, recording and closing. Editing shows a compact text-first dialog, visible actions and Hide keyboard. Starting/resuming answer playback also dismisses the keyboard without losing the draft. Keyboard-inset regression tests pass. Deployment and browser QA pending; physical phones unverified. Production349/native27 and AI providers unchanged.
September 18 DEV406: Show me how is live. A real repair displayed the knob illustration, opened picture search and a matching playable tutorial. Desktop and phone-sized layout checks passed. Other repairs use clearly labeled searches; custom image/video generation is not included. Physical phones and every answer route were not live-tested. Production349/native27 unchanged; receipt407 does not deploy website407.
September 18 DEV visual help prepared: Show me how adds a conceptual knob diagram and checked external guides/videos for supported hardware, plus clearly labeled picture/video searches for other repairs. Searches are opt-in and editable; private repair details are not copied. Hazard results suppress DIY visuals. No automatic media, paid provider or voice/model change. Deployment and browser QA pending. Production349/native27 unchanged.
September 17 DEV404: free Sienna streams before the clip finishes. Two tests reached playing within about 0.9s and 0.8s of the speech request. Multiple passages, Pause/Resume/Stop, collapse/Settings cancellation and phone-sized controls passed. Timings are not guarantees; unsupported browsers keep complete-clip playback. Full-answer completion, heard quality and physical phones remain unverified. Production349/native27 unchanged; receipt405 does not deploy website405.
September 17 DEV Sienna streaming prepared: supported browsers can begin playing before the whole MP3 clip arrives. The speech service and private website proxy now pass bounded audio incrementally; unsupported browsers retain complete-clip playback. Free Sienna, full guidance, opt-in controls and existing access limits stay unchanged. Deployment and real startup measurements are pending. Production349/native27 unchanged; no paid fallback.
September 17 DEV402: Sienna starts with a short clip and prepares only the next during playback. One first request took 6.2s versus an earlier 15.0s sample; different answers, not a guaranteed speedup. Sequential playback, Pause/Resume/Stop and collapse/navigation cancellation passed. Cloud delays and gaps may remain; no paid fallback. Full-answer completion, heard quality and physical phones were not tested. Production349/native27 unchanged; receipt403 does not deploy website403.
September 17 DEV Sienna speed improvement prepared: start with a shorter segment and prepare one next segment while the current audio plays. Free Sienna and the complete written answer stay unchanged, with no paid fallback. Deployment and timed browser checks are pending. Production Build349 and native build27 are unchanged.
September 17 DEV Build400: free Sienna played two consecutive segments; Pause/Resume/Stop, answer switching, collapse cancellation and desktop/mobile-sized controls passed. Settings return retained three idle answers. Short segments may have gaps. The full answer, physical phones and heard voice quality were not verified. No paid fallback is enabled. Production Build349 and native build27 are unchanged. This documentation-only receipt does not deploy website Build401.
September 17 Sienna correction prepared: the free voice was missing from the provider model allowlist. That exact free model is now allowed, and a real short audio request passed. Shorter sequential audio segments are being prepared after a full-answer segment timed out; no guidance is truncated and no paid fallback is enabled. Full-answer browser checks remain pending. Production Build349 and native build27 are unchanged.
September 17 DEV Build398 checkpoint: Sienna controls are deployed, but live speech returned an error while repair answers and guest limits worked. Runtime diagnostics are being prepared to identify the failure without logging answer text or audio. Real playback is not yet verified. Production Build349 and native build27 remain unchanged; no paid voice fallback is enabled.
September 17 DEV Sienna voice prepared: replace the basic browser voice with free neural speech. Listen, Pause, Resume and Stop keep the answer visible; speech uses a separate provider without another repair question or automatic paid fallback. Free-service limits may affect availability. Deployment and real playback checks are pending. Production Build349, native build27 and Ollama repair models are unchanged.
September 17 DEV Build396 deployed: Listen to solution adds opt-in playback with Pause, Resume and Stop. Real browser controls, single-answer playback, collapse/navigation cleanup and desktop/mobile layouts passed with three AI answers. Physical phone audio remains untested. Production Build349, native build27 and AI models are unchanged; this receipt does not deploy Build397.
September 17 DEV voice playback prepared: Listen to solution adds Play/Pause/Resume/Stop for repair, photo, premium and sandbox plan-test answers. Safety guidance stays first and text remains visible; opening a result never plays audio or requests a microphone. Browser voices may process text online; unavailable audio has a reading fallback. Deployment and real-browser audio testing are pending. Production Build349, native build27, AI models and free-beta access are unchanged.
September 17 DEV Build394 deployed: Use voice now opens an explicit-start, editable review flow in repair and photo descriptions. Desktop and phone-sized review/apply/cancel checks passed; actual microphone transcription still needs device testing. DEV permits same-origin microphone requests, subject to browser permission. Production Build349, native build27 and AI models remain unchanged. This is a documentation-only receipt, not a Build395 deployment.
September 17 DEV voice deployment correction prepared: enable same-origin microphone permission requests on DEV, fixing a blocking security header found before Build393 was deployed. Browser permission is still required. Production and native apps remain unchanged; real browser verification is pending.
September 17 DEV voice input prepared: use a microphone to fill the repair or photo description, review/edit the transcript, then choose Use this text. Opening voice does not listen or submit a repair automatically. Browser support and microphone permission are required; typing remains available. Deployment/browser verification is pending. Production Build349, native build27 and AI models are unchanged.
September 16 DEV photo runtime391 verified with website Build390: real photo upload/remove/reselect and image-grounded repair steps passed on desktop and phone-sized layouts. Guidance now requires identifying the actual adjuster and making small reversible changes instead of guessing screw directions. Physical camera capture and repair execution remain device/manual checks. Production Build349 and native build27 are unchanged.
September 16 DEV Build390 photo upload verified: real JPEG upload and image-grounded repair steps work through the browser, with mobile layout and latest-text Find a pro handoff checked. A photo-only guidance improvement is being prepared to avoid guessing adjustment screws or turn directions. Physical phone camera capture remains a device check. Production Build349 and native build27 are unchanged.
September 16 DEV photo compatibility corrections prepared after Build389 live testing: fix browser-only photo decoding and narrow-screen title layout, and adapt the Ollama Cloud request. Real image analysis and repair-result verification are still pending. Production Build349, native build27 and existing text/premium inference are unchanged.
September 16 DEV photo diagnosis prepared: take or upload a photo from Fix, add an optional symptom, and request image-grounded repair steps using Ollama. Photos are resized and metadata stripped before sending, without adding a server photo library. Guest/account controls remain in place. Deployment and real image testing are pending. Production Build349 and native build27 are unchanged.
September 16 DEV Build387 deployed: matching customer, management and sandbox billing packages passed fresh status/refresh/Max-access checks. One mobile comparison returned three actionable answers labeled AI 1/2/3; Google operator sign-in also passed. No new payment-state changes or complete-site approval are claimed. Outage retries have automated coverage, not a forced live outage. Production Build349, native build27 and free beta are unchanged; this receipt does not deploy Build388.
September 16 DEV Build387 prepared, not deployed: bounded read-only retries and clearer feedback address the account-verification failure observed in Build386. Manual refresh recovered; the exact upstream cause is unproven. The focused billing suite passed 180 tests and combined Flutter checks passed 748. Live validation remains pending; payments and AI requests are not automatically retried.
September 16 DEV Build386 targeted QA: billing contention auto-recovered; mobile draft retention and fresh category repair, Network save/email/review/archive, Get help save/full email/admin notes/archive/consent restrictions, and real CAPTCHA feedback submission passed. The operator feedback record remains pending. Production Build349, native build27 and free beta are unchanged. This is not a complete application sign-off.
September 16 DEV readiness corrections prepared: billing status and plan-access checks recover from temporary contention without retrying payments or AI requests. Plans separates saved interest from sandbox subscriptions and links to benefit tests. Home service shortcuts start a fresh repair with the selected category. Deployment and live regression testing are pending; production Build349, native build27 and the limited-time free beta remain unchanged.
September 16 DEV Build384 verified: paid Plus access, declined-then-recovered Max access, three AI answers, external Maps, the scheduled-cancellation notice and mobile Payments layout/refresh passed targeted checks. One post-Portal status load needed manual retry; its cause is unconfirmed. Production Build349, native build27 and the limited-time free/unlimited beta are unchanged. This is a documentation-only receipt, not a Build385 deployment or approval of every app flow.
September 16 DEV Build382 is live: a real Basic AI answer passed and its daily test allowance changed from 5 to 4. Billing status during Portal-flow testing then found a configuration-read error before the Portal opened; Build384 is being prepared to fix it and is not deployed. Full subscription lifecycle, mobile and Max testing remain pending. Production Build349, native build27 and the limited-time free/unlimited beta are unchanged; no live payments are enabled.
September 15 DEV plan-benefits testing prepared: Payments links to a separate real-AI sandbox test for Free, Basic, Plus and Max, including three AI answers and external Google Maps results for verified Max. Normal beta remains free/unlimited. Exact-plan Portal switching, safe daily test usage and unpaid-upgrade checks are being integrated; deployment and real end-to-end verification are pending. Production and native apps are unchanged.
September 15 DEV Build380 deployed: Payments & subscription is visible from Account and Plans, real Stripe sandbox management returns to the payment page, and the cached missing payment icon is fixed. Mobile and desktop checks passed without clearing browser caches. Production Build349 and native apps are unchanged. The beta remains free/unlimited; paid-plan switching, limits and paid AI/Maps benefit integration remain unfinished.
September 15 DEV font-cache correction prepared: DEV379 now has a dedicated Payments page and a working Stripe Portal return. Browser testing found a missing payment icon from a cached font; the next package versions font URLs so clearing browser caches is unnecessary. Correction deployment and browser verification are pending. Production Build349, native apps and free-beta access are unchanged.
September 15 DEV payment navigation prepared: open Payments & subscription from Plans or Account to test monthly Stripe subscriptions, invoices, payment details and cancellation. Loading/retry messages remain visible on DEV. Deployment testing is pending; production and free-beta access are unchanged, and sandbox purchases do not grant paid plan benefits.
September 15 DEV billing runtime377: the scheduled-cancellation notice is fixed and verified after real sandbox Checkout, Portal cancellation and a full Account reload on mobile and desktop. Undo cancellation clears the notice; re-cancellation restores saved pending status. Website Build375 and production Build349 are unchanged; this backend-only update is not a complete production release. The free unlimited beta is unchanged.
Historical September 15 preparation checkpoint: the DEV billing correction was prepared, with deployment and real Portal retesting pending then. Runtime377 above now records the deployed cancellation-notice fix and verified undo cancellation.
Historical September 15 DEV Build375 checkpoint: sandbox Checkout, monthly renewal and final cancellation passed, but Account lacked the scheduled-cancellation notice. Runtime377 above resolves that failure. Production promotion remains pending; the free unlimited beta is unchanged and no live payments are enabled.
September 13 DEV update: show the signed-in account and Settings, save display names and appearance, and preserve unfinished repair, Get help and Network drafts. Production promotion is pending.
Test Stripe website subscriptions on DEV while the current limited-time free and unlimited offer remains unchanged; production payments and paid-access enforcement are not enabled.
DEV premium AI answers now include numbered how-to repair steps, preparation, and a clear test of whether the fix worked; production promotion is still pending.
The invited DEV premium test account gets three AI answers directly from the main repair button, with the question carried over automatically.
Test three separate repair answers labeled AI 1, AI 2 and AI 3 on DEV for an invited premium test account, without billing.
Use email, Google, Apple or Facebook with one shared Repair Robin sign-in account.
Promote the exact tested dev website package to production, with automated checks for matching features, settings and versions.
Recorded production releases
Dates are in America/New_York. This is a release record, not a live service-status check.
Released ·
v0.1.0 · Build 506
Visible breaker-off reminder in electrical safety diagrams
The approved DEV Build506 website and backend package was promoted unchanged to production after its fourteen-flow DEV check and exact-artifact staging.
The two fixed light-switch safety diagrams now retain a visible breaker-OFF reminder on each step; no AI-generated wiring photos are shown for that protected sequence.
A signed-in production fitted-sheet question returned three AI answers and a step picture. A light-switch question showed two fixed diagrams and withheld three unsafe model answers; this limitation remains visible rather than presenting them as successful answers.
The limited-time free Premium offer, account and payment settings, and native iOS/Android Build27 are unchanged. Physical-device audio/camera and every third-party flow were not re-certified.
Free Premium for everyone during the limited-time beta
DEV and production now serve the same tested website and management package. Every registered account receives temporary Max features without a tester invitation, payment or subscription.
Get three answers labeled AI 1, AI 2 and AI 3, plus step-by-step pictures. The first guest question remains free; further questions require a free account and remain free during the beta.
Pictures can be reused from matching solutions. Try pictures again rechecks an unavailable guide without resubmitting the question, while preserving ready pictures and the selected step. Generated video remains removed.
Real production testing returned three answers and four pictures with working Next and Previous. The complete picture set took about 83 seconds in that test; speed and image accuracy are not guaranteed.
Guest and signup checks, help and Network submissions, verified feedback, management readback and email transport passed. Payments remain disabled; no charge or automatic renewal is created.
Website Build496 does not update the unchanged iOS and Android Build27. Physical-device voice, camera and all other flows were not re-certified in this release.
Practical guide and optional advertising integration
DEV and production received the same tested customer and management package, with a towel-folding guide and three accessible diagrams.
All four shared sign-in providers and the limited-time free, unlimited beta remain available. Production payments remain off.
Post-release Chrome testing found an advertising session-marker error before the Allow button appeared. A corrective release is being prepared; actual ad delivery is not working yet.
Google has authorized ads.txt but its website review is still pending. This is not approval to serve ads or a claim of advertising revenue.
Website Build433 does not update the unchanged iOS and Android build27.
Ad-free paid-plan wording and publisher verification
Plans explain that active Basic, Plus and Max memberships will be ad-free; Free will be ad-supported when ads are enabled.
The limited-time free and unlimited beta continues. Production payments remain off and this update creates no charges.
Publisher ownership metadata and seller files are live. Google ownership verification passed and website review was requested; ads are not serving and Google approval is still pending.
DEV and production share the exact approved website package and all four sign-in providers. DEV alone offers the server-verified sandbox ad-layout preview, without an ad network or revenue.
Desktop and phone-sized production checks verified paid-plan wording, accessible question controls and sign-in icons. This update does not claim to fix intermittent AI availability.
Website Build429 is separate from unchanged iOS and Android build27; no native ads SDK or store release is included.
Optional website analytics with a clear privacy choice
Allow or Decline optional Google Analytics on the production website. Change your choice later in Account & settings or the privacy notice.
Analytics is off before consent and on DEV. It excludes private pages, questions, answers, uploads, form details, account identity and URL parameters; advertising features stay disabled.
Production and DEV retain the same approved website package and shared sign-in system. The limited-time free beta continues; production payments remain off.
Actual production checks confirmed consent controls, scrubbed page views, no duplicate navigation events and received traffic in the dedicated Analytics property.
AI responses can still be intermittently unavailable: two concurrent authenticated checks returned an unavailable error, and a later sequential request recovered. The cause is undetermined; this analytics release does not change or claim to fix AI availability.
Website Build426 is separate from unchanged iOS and Android build27. This update does not re-certify physical-phone audio, microphone or camera behavior.
The approved website update is live and still free
Production and DEV now share the same approved website features and account system, including email, Google, Apple and Facebook sign-in.
Ask repair and everyday how-to questions, with practical steps and optional photo or voice input.
Choose from 11 answer and playback languages. New cloud-generated playback supports the selected language; some fixed menus and headings remain English.
Use topic-dependent illustrated guidance and inline tutorials, alongside the existing account-authorized three-answer comparison.
Keep the question field and main controls accessible in the compact mobile layout.
Try one question without an account. Further questions require a free account and remain free and unlimited during the limited-time beta. Production payments remain disabled.
Website Build424 is separate from iOS and Android build27. Device audio, microphone and photo behavior still depend on browser/device support.