[00:00:02] Heyeverybody,goodtoseeactualfacesinsteadoflittleloadingcirclestoday.Jackson,thatplantbehindyouhasgottenenormoussinceourlastsync.Eitheryoufinallywatereditoryouborrowedabetterplantforthecamera.Beforewegetserious,doesanybodyhaveasmallpersonalvictoryfromthelasttwoweeks?Mineisthatmycoffeemakerstoppedmakingthenoiseofadepartinghelicopter.Theplantisreal,andthesecretisthatImoveditawayfromtheheater.Apparentlyplantsprefernottoliveinsideahairdryer.Mypersonalvictoryismakingitthroughanentiregrocerytripwithoutcheckinganerroralert.Ididcheckthedashboardintheparkinglotafterward,butIamchoosingtocountthatasprogressratherthaninventingastricterscoringsystem.Isuccessfullyexplainedmyinternshiptomygrandmotherwithoutsayingmachinelearning,whichfeltlikeanachievement.ItoldherIhelpacomputerunderstandpeoplewhentheytalk.Sheimmediatelyaskedwhyhertelevisionunderstandseveryoneexcepther.SoapparentlyIhaveacquiredasecondevaluationproject,andthestakeholderhasstrongopinionsabouttheremotecontrolandabsolutelynopatienceforadiscoveryinterview.Thatisaverygoodstakeholder,honestly.Myweekendachievementwasassemblingabookshelfwithouthavingamysterioushandfulofextrascrewsafterward.Caitlin,Ialsotriedthelittlebakeryyourecommended.Youwererightaboutthebread,butyouundersoldhowlongthelinegets.IhadenoughtimetomentallyredesigntheirorderingsystembeforeIreachedthecounter,whichprobablymeansIneedahobby.IdeliberatelyleftoutthelinebecauseIwantedyoutogo,thatisprobablynotaproductprincipleweshouldcarryintovoiceflow.Mywinislessexciting.Ifinallyclearedthestackofunopenedmailnexttomydoor.Halfofitwasofferstohelpmeorganizemylife,whichfeltalittleaggressivecomingfromtheactualsourceofthemess.Allright,excellentevidencethatwearefunctioningadultsinatleastafewsettings.Imean,thebarisnothigh,butweclearedit.Todayisourbi-weeklyvoiceflowsync,andIwantdecisionsratherthanaguidedtourthrougheveryticket.Wewillcoverlaunchreadiness,deploymentcosts,ashortinvestorcheck-in,thendesignandAFASevaluationworkbeforewelockownersanddates.Saoirseistravelingandlisteningfromherphone,sowewilltakeinfrastructurebeforethebreak.Kavehwilljoinpartwaythroughforaquickconversationaboutthebusinessmetrics.Afterthat,wewilltakefiveminutes,thencomebackforthehomepage,settings,andnamerecognition.Imean,thoselasttopicsareconnectedmorecloselythantheylook,sopleasekeepthecustomerexperienceinview.OnehousekeepingthingbeforeJacksonstarts.Thelaunchdocumentistheversiontitledcontrolledrollout,nottheoldergeneralavailabilitydraft.Iputthelinkinthemeetingnotes.Priyahasthesameversionatthepartnercompany,andsheknowswearedecidingthenextcohorttoday.Ifsomebodyopenstheolddocumentandseesapromiseofunlimitedusage,pleasecloseitbeforeyourbloodpressurechanges.Onlaunchstatus,thecoredictationpathishealthy.Intakeacceptstheaudio,createsthejob,andhandsittotheAIbuilderspacetranscriptionAPI.WethenapplyourformattingrulesandreturnthedrafttoVoiceFlow.Thefailurerateisdownsubstantiallyfromtheprevioussync.Whatremainsembarrassingispropernames.Ordinarysentenceslookclean,andthensomebodyintroducesthemselves,andthetranscriptinventsacompletelydifferentperson.Thatmatchestheintakeinterviews.Peoplearewillingtoforgiveamissingcomma,buttheynoticeimmediatelywhentheirownnameiswrong.JacksonbecameJacksoninthreesessions,andSochitelgotthreedifferentspellingsinoneconversation.Aphahadanevenstrangersubstitution.Thelargerissueisthatusersinterpretthosemistakesasevidencethatwedonotunderstandtheirworkplace,evenwhentherestofthedictationisuseful.Letusseparateavailabilityfromqualitybeforethosebecomeonegiantredstatusdot.Jackson,arewesayingthesystemisreliableenoughforthenextcohort,butnotaccurateenoughtoinviteeverybody?Imean,thatsoundsmanageableifwearehonestabouttheboundary.Idonotwantusholdingasmall,usefulreleasehostagetoastandardwehavenotactuallywrittendown.Yes,thatisthedistinction.Reliabilityisgoodenoughforacontrolledexpansion.Namehandlingisnotgoodenoughforanunqualifiedlaunchclaim.Wealsofixedtheduplicatejobproblem.Ifabrowserretriesduringanetworkinterruption,intakenowreusestheexistingjobidentifier.Thatmeanswenolongerpayfortwotranscriptionsjustbecausesomeonewalksbetweenroomsandtheirwirelessconnectionbrieflybecomesaphilosophicalconcept.CanIcheckwhatcountsasanamefailureinthelaunchdocument?IfsomeonesaysCaitlinandweoutputCaitlin,thatsoundsidenticalbutisstillwrongfortheperson.IfsomeonesaysCaveandweoutputCoffee,thatiswronginadifferentway.Myclipsdistinguishspellingerrorsfromsubstitutions,andIwouldrathernotcollapsethembeforeweknowwhichproblemthenextcohortactuallyexperiences.Keepthatdistinction,Aoife.Forthelaunchgate,eitheronecountsasafailedexactnamematch,butthebreakdownbelongsbesidetheheadlinenumber.Weshouldalsorecordwhetherausercancorrectitonceandhavethatcorrectionstick.Aspellingmistakethattakesonecorrectionisdifferentfromamistakethatreturnseverymorning,likeanextremelypersistentvisitorwhohaslearnednothingfrompreviousconversations.Thecorrectionflowcurrentlymakesthatsecondproblemworse.Someonefixesanameinthetranscript,butnothingtellsthemwhetherVoiceFlowhaslearnedanything.Theinterfaceimpliesconfidencewithoutexplainingscope.Icanchangetheconfirmationsoitsaysthecorrectionappliestothistranscript,unlessweactuallysaveapreferredspelling.Weshouldnotletasmallanimationquietlymakeapromisethebackendcannotkeep.Exactly.Rightnow,editingtextchangesthatdocumentonly.Wehaveaworkspacevocabularytablebehindafeatureflag,butwearenotconsistentlysendingthoseentriesthroughtotheAIbuilderspacetranscriptionAPI.Priyaconfirmedthesupportedhintformatyesterday.Ineedtowirethatintointakeandaddasizelimitbecausesendingsomebody'sentireemployeedirectorywitheveryshortrecordingwouldbeabaddefault.Thatsoundslikeapracticallever.Jackson,pleasedonotcallitlearningintheinterfaceifwearereallypassingavocabularyhint.Imean,thedistinctionmatterswhenauseraskswhyacorrectionworkedinoneworkspaceanddisappearedinanother.Caitlin,canwemakethenextcohortexplicitlytestthatboundaryratherthanjustcollectinganothergeneralsatisfactionscorethattellsuseverybodyhasfeelings?Yes,Iwouldliketenadditionalworkspaceswithatleasttworegulardictationusersineach.Wecanaskeachworkspaceforasmalloptionallistoffrequentlyspokennames,explainwherethatlistisstored,andcomparetheirfirstsessionwiththeirthird.Priyacanrecruitfourofthoseworkspacesfromherpartnernetwork.Theremainingsixshouldcomefromourexistingwaitlist,sothecohortisnotentirelyfriendlycolleagues.Pleaseincludeatleastoneworkspacewherepeopleusenamesfromseverallanguagebackgroundsinthesamerecording.AcleanEnglishsentencecontainingSochitel,Saoirse,andCaveisausefulcase,notanunusualedgecaseweshoulddiscard.Ialsowantafewspeakerswhosaytheirownnamenaturally,insteadofreadingitverycarefullybecausetheyknowamodelisbeingtestedandwanttohelpitsucceed.Thereisanotherintakeissuethatissmallerbutworthsayingoutloud.Wepreserveaudiountilthejobfinishes,thenfollowtheworkspaceretentionsetting.Wewerenotshowingthatsettingclearlyduringonboarding.Nothingissecretlykeepingrecordingsforever,buttheexplanationisburied.Iwantthelaunchchecklisttoincludeatestthatthedisplayedpolicymatcheswhatthedeletionworkeractuallydoesforthatworkspace.Icanmakethatexplanationvisiblewithoutturningonboardingintoaconsentmaze.Ashortsummarybesidetherecordingcontrolswouldhelp,withalinktothefullsetting.Jackson,Iwillneedplainlanguageforthedifferencebetweenrawaudioandthetranscript.Peopleassumedeletingonedeletestheother.Ifthatisnotalwaystrue,thescreenhastosaysobeforetheymakethechoice.Forlaunchmessaging,Iamproposingfastdictationwitheditabletranscripts,notperfecttranscription,anddefinitelynotautomaticunderstandingofyourentirecompany.Wecanmentioncustomvocabularyonlyoncethehintpathisworkingandmeasured.PriyaaskedwhetherAIbuilderspaceneedstoappearintheonboardingcopy.Myinstinctistoputtheproviderinformationintheprivacyexplanationratherthancrowdingthefirstscreenwithinfrastructurenames.Thatworksaslongasthedisclosureiseasytofindandspecificenoughtobemeaningful.Wearenothidingaprovider,andwearenotmakingpeoplereadanarchitecturediagrambeforetheydictateasentence.Jackson,whatwouldactuallystopthecohortexpansionthisweek?Giveustheshortlistofthingsthatwouldmakeyousayno,evenifeverybodyelsewereexcited.Arepeatablelossofaudiobeforeprocessingcompleteswouldstopit.Sowouldcross-workspaceaccesstoatranscript,obviously,orabillingpaththatsubmitsthesamerecordingrepeatedly.Thosechecksarepassingrightnow.Aslowerresponseunderaburstwouldmakemereducethecohortoraddaqueuelimit,notautomaticallycancel.Incorrectnamesareaqualitygate,butIneedIFA'smeasuredbaselinebeforeIproposearesponsiblethreshold.Thenthelaunchstatusislimitedexpansionpendingthesafetychecksandtheevaluationreadout,withnopublicannouncementyet.Iwillupdatethedocumentinexactlythoseterms.Ialsowantonesharedsupportlabelfornameproblems.Rightnowtheyarescatteredacrossaccuracy,onboarding,andsomethingcalledweirdoutput,whichishonest,butnotparticularlyusefulwhenwetrytocounttheimpactacrosscustomers.Good.Imean,weirdoutputisamood,notataxonomy.Letusgivesupportcategoriesthatmaptodecisions.Wrongspelling,wrongpersonorphrase,correctionnotretained,andvocabularyhintapparentlyignored.IFA,youcantelluslaterifthatsplitistechnicallysensible.Caitlin,makesurePriyagetsthesamelabels,sopartnerreportsdonotarriveasaseparatepileofanecdotes.Thatsplitworks,withonecaution:apparentlyignoredisanobservation,notproofthattheproviderignoredthehint.Theaudiomightbeambiguous,orthehintmighthavebeenmalformedbeforeitreachedtheprovider.Icanlabeltheuservisiblesymptomandattachtherequestmetadataseparately.ThatshouldletJacksoninvestigatewithoutforcingsupporttomakeaninferencetheycannotactuallyverifyfromascreenshot.Icanaddasafetracefieldthatrecordswhethervocabularywasattached,howmanyentriesitcontained,andwhichformatversionweused.Itwillnotcontaintheactualnames.Thatshouldhelpusdistinguishamissinghintfromanineffectivehintwhilekeepingthelogsboringinthebestpossibleway.IwillcoordinatethefieldnameswithPriyasoherteamcansenduscomparablereports.Sarsha'sonthephone.Sarsha,goahead.Beforeyoustart,wecanhearsomestationnoise,butyourvoiceiscomingthroughclearly.Giveusthebillchangefirst,thenwhatcausedit,thenthemoveyouareproposing.Afterthat,Jacksoncanaskaboutthenumbers,andwewilldecidewhetherThursdayisarealmigrationwindoworjustanoptimisticcalendardecorationthatmakeseverybodyfeelbetter.Thanks.Andyes,thatisastationannouncementbehindme,notaproductionalarm.TheKoyabillwentupby$480thismonth,from$620to$1,100.Mostofthatincreasecamefromthevocabularyre-rankerservice,whichneverwenttosleep.Itkeptaminimumincentivealive,evenwhenintakehadnojobs.Trafficincreasedalittle,butnotnearlyenoughtoexplainthatjump,sopleasedonotdescribethisascustomergrowthpayingforitself.Thanks,Aish.Wasthatminimuminstanceintheserviceconfiguration,ordidthehealthcheckkeepwakingit?IaskbecauseIrememberchangingthereadinessroutewhenweaddedthere-ranker.Iftheplatformisfollowingasettingwegaveit,Iwantustofixthesetting.Ifweareaccidentallygeneratingactivity,movingthesamebehaviortoanothermachinewilljustmakethewastelessvisible.Itwasaminimuminstancesetting,notthereadinessroute.Thevocabularyre-rankerinheritedthealways-ondeploymentprofilefromthetranscriptiongateway,andnobodychangeditwhenweflippedservices.Ihavecorrectedthatinstagingandverifiedthattheidleinstancestops.Productionisstillintheoldprofileuntilwechoosethemigrationpath,becausechangingtwothingsatoncewouldmuddythecomparison.Thetranscriptiongatewayshouldremainavailable.There-rankeristhepartThatisahelpfuldistinction.Imean,weshouldnotsolveanavailabilityrequirementbyaccidentallyapplyingittoeverythingwithanetworkport.Sarsa,telluswhichservicesmoves,whatstayswhereitis,andtheactualdayyouwanttodoit.IalsowantthefallbackstatedclearlyenoughthatCaitlincanexplainittoacustomerwithoutborrowingJackson'sentirevocabularyfortheafternoon.Myplanistomovethevocabularyre-rankeronThursday,June11thtoourownboxrunningDLLM.IntakeandthetranscriptiongatewaystayonKoyet,andtheAIbuildersaystranscriptionAPIremainsourtranscriptionprovider.Thisisnotareplacementspeechmodellaunch.Wearemovingthesmalltextmodelthat'sforVocabularycandidatesaftertranscription.Ifthenewpathmisbehaves,afeatureflagroutesrequestsbacktotheexistingCOIBservice,whichstaysavailableduringtheobservationwindow.Thatswitchdoesnotrequireanewclientrelease.Thatlastpointmatters.Icanexplainabackendmovewithanimmediatefallback,butIdonotwantthecohorthearingthatwechangedthewholedictationengine.Seriously,willtherebeanyvisiblemaintenanceperiod?PriyaisschedulingonboardingsessionsforThursdaymorning,andIwouldrathermovethosenowthandiscoverthatourcarefulinternalwordingfailedtomentionanactualinterruptionforhercustomers.Thereshouldbenoplannedinterruptions,becausetheflagchangesthererankfordestinationwhiletheoldserviceremainshealthy.Iwanttostartat10inthemorningPacific,afterPriya'searliestonboardingsession,andsendonlyinternalworkspacesfirst.Theboxhasenoughmemoryforthecurrentmodelandourmeasuredconcurrency,butthatisnotpermissiontosendeveryqueuejobatonce.Iwillcapin-flightre-rankerrequests,watchthequeue,andexpandgradually.Ifthequeuegrowscontinuouslyfortenminutes,werollbackratherthanimprovise.Canwegettheperformancenumbersbehindthatconfidence?Specifically,IwanttheKVcachehitrate,whatworkloadproducedit,andwhethertheprefilltimeincludeswaitinginthequeue.Wehavehadafewdashboardnumbersthatlookedwonderfulbecausethetimerstartedafterallthepainfulwaitingwasover.Iwouldpreferanugliernumberthatreflectswhatthepersondictatingactuallyexperiencesontheirscreen....themasend-to-endlatency.Mediumqueuewaitwas30milliseconds,andthe96thpercentilewas140millisecondsatourreplayload.Thereplayusedstandardizedpromptstructuresandsyntheticvocabularylists,notproductionrecordingsorcustomernames.Saoirse,aretherepeatedprefixesmostlyourinstructiontemplate,orarerepeatedvocabularylistsdoingthework?Iwanttounderstandwhetherthe68%resultsurvivesthenextcohort.Ifthereplayaccidentallysendsthesameshortlistoverandover,theKVcachecouldlookmuchbetterthanitwillwhenPriyabringsinseveralworkspaceswithdifferentnamesandpeoplestarteditingthoselistsbetweensessions.Cohortprobablywill.Withshuttleslists,thehitratedroppedto53%,andmedianprefillroseto210milliseconds.Thatisstillbetter,butitisthemoreconservativeplanningnumber.Iwantyoutohelpbuildalessfriendlyreplay,includinglistedits,longnames,andalternatingworkspaces,beforeThursday'sproductionswitchbeginsinearnest.Good.Letuscarrytheconservativenumberforward.Imean,acashhitrateisonlyusefulifwerememberwhatwashittingthecash.Sasha,namethepeopleyouneed,becausetravelshouldnotturnthisintoyoudebuggingfromatrainplatformalone.Jackson,Iamassumingyoucanowntherequestrouting,butIwantyoutoconfirmbeforeIvolunteeryourentireThursdaymorning.Accessfortheroutingflag,timeouthandling,andatestedrollbackcommand.IneedAphaforthatadversarialreplayandaquickaccuracycomparisonsothefasterpathdoesnotquietlychangerankings.IneedFreyatoconfirmhersessionscheduleandsendonerepresentativesanitizedrequestshape.Caitlincancoordinatethatintroduction.Iwillowndeployment,dashboards,andthecostcomparison.Thecheetahdoesnotneedtoredesignanythingforthismove.Iftheuserseesanewscreenbecauseofit,wehaveprobablytakenanunnecessarydetour.Confirmed.IcanpairwithyouontheflagWednesdayandbeonlineThursdayfortherollout.Iwouldalsoliketotestadeadconnectiontothebox,notjustacleanerrorresponse.Ahangingrequestishowasupposedlyoptionalre-rankerturnsintoatranscriptionoutage.Weshouldreturntheoriginaltranscriptwhenthere-rankertimesout,thenrecordthatdegradationwithoutretryingforeverinthebackground.Agreed.Thetimeoutwillbebounded,andare-rankerfailurewillreturnthebasetranscriptratherthanblockdelivery.Onelastsoftcaveat:theownedboxisnotfreejustbecauseitisalreadypaidfor.Iwillreportallocatedhardwarecosts,power,andoperationstimealongsidethereducedCOIofcharge.Myfirstgoalisremovingthat$480surprise,notclaimingmagicalzero-costinference.Iwillpostagoorno-goupdateWednesdayafternoon,andIwillnotrunthemigrationwhileI'mstilltraveling.IwillcontactPriyarightafterthisandgetthesessionscheduleintoSarah'srolloutnote.Forthecustomer-facingdescription,Iwillsay,transcriptsremainavailableiftheoptionalvocabularypassistemporarilyunavailable.Jackson,pleaseconfirmthatwordingagainsttheactualtimeoutbehavior.Iwouldratherpromiseaslightlylesspolishedtranscriptthanimplythewholeproductisdownwhenonlytheextranamerankingpasshasbeenskipped.Thatistheplan.Imean,theheadlineisnotthatwediscoveredacheapermachine.Itisthatwenowunderstandtheidleservice,thefallback,andtherealcostboundary.Sostayforanythingyoucan,butdonotmissyourtrainonourbehalf.Wewillrepeatyourownersanddatesattheend,andnobodyshouldtreatadroppedphoneconnectionasapprovalforariskychange.Kavehjustjoined,soletmebringhiminbriefly.Kaveh,welcome.Wehavecoveredlaunchreadinessandatargetedinfrastructuremove,andweareabouttogetintoproductquality.Youaskedforafewminutesonretentionandwhatevidencewouldmatterforthenextround.Wehaveaboutfourminutesnow,sogiveusthepracticalversion.Thanks.ThepracticalversionisthatIcarelessaboutabigregistrationspikethanaboutwhetherpeoplereturnaftertheirfirstusefultranscript.Retentionshouldtellusthatvoiceflowbecomespartofaworkflow.Pleaseseparateuserswhotriedonedemofrompeoplewhoactuallyhadanotherreasontodictateandchoseyouagainthatweek.ThatishowI'mframingthenextcohort.Wewilltrackafirstsuccessfuldictation,thenareturnonadifferentday,andthenrecurringuseacrossthefollowingweeks.Wearenotplanningtocountopeninganonboardingemailasproductactivity,evenifthatwouldmakethechartlookfriendlier.Good.Dailyactivesareusefuliftheactivitymeanssomethingconsistent.Iwoulddefineanactiveuseraroundacompleteddictation,withareasonableminimum,soanaccidentalbuttontapdoesnotcount.Thenshowdailyactivesalongsideweeklyactives.Somevaluableusersdictatetwiceaweek,andadailychartalonecanmakethathealthybehaviorlookdisappointing.Thatfitsthejobswearehearingabout.Imean,apersondictatingmeetingnotesonTuesdaysandThursdaysisnotnecessarilylesssuccessfulthansomebodysendingfivetinymessageseveryday.Wewillmaketheeventdefinitionexplicitandsegmentbyworkflowinsteadofquietlyrewardingwhicheverbehaviorproducesthelargestnumber.Forthenextfundinground,thequestionwillbewhetherusagerepeats,whetherthoseuserscanbecomepayingcustomers,andwhetherservingthemhassensibleeconomics.Alowinfrastructurebillisnice,butitisnotasubstituteforretention.Likewise,strongretentiondoesnotexcuseacostcurvethatgetsworsewitheveryadditionalminuteofaudio.Iwouldalsokeepthecohortsvisible.Ifthenewestcohortretainsbetterafterthenamefixes,thatisausefulstoryaboutlearningandexecution.Ifalltheimprovementcomesfrommixingnewuserswithafewveryloyaltesters,itishardertointerpret.Smallsamplesarefineatthisstage.Hiddendenominatorsarenot.Understood.Iwillincludethecohortsizes,theactivationdefinition,andtheusagewindowonthesamepageasthecharts.Forcosts,wewillseparatetranscriptionproviderchargesfromthevocabularypassandsharedinfrastructure.Sashaalreadywarnedusnottodescribeanownedmachineasfree,whichseemsespeciallyrelevanttothisdiscussion.Exactly,andwhenyoutalktouserswhostopped,askwhatinterruptedthehabit.Wasitaccuracy,speed,trust,orsimplynorecurringmeet?Ashort,honestexplanationfromfivepeoplecanbemoreactionablethanabroadsatisfactionaverage.Iwouldbeespeciallyinterestedinwhetherrepeatednamecorrectionsareenoughtomakesomeoneabandonanotherwiseusefulproduct.Ineedtodropnow,butthisistherightleveloffocus.Sendmethecohortreadoutonceithasactualobservations,notaprettierforecast.Iamhappytoreviewthemetricdefinitionsbeforehandifthatsavesrework.Thanks,everyone,andSaoirse,safetravels.Iwillleaveyoutotheproductdecisionsratherthantakingtherestofyourmeeting.Let'stakeafive-minutebreaknow,andwhenwereturn,wewillstartwithXochitl'sdesignrefreshbeforemovingintoIFA'sevaluationresultsandourfinalassignments.Alright,welcomebackeverybody.Satchel,pleasetakeusthroughthehomepagebuttonfirst,thenthecolormismatchwithAIBuilderSpace,andfinallythesettingspage.Thehomepagehasthreethingsthatlookliketheprimaryaction,andonlyoneofthemstartsdictation.Thelargebuttonsayscreate,themicrophonetilesayscapture,andthesmallbuttonintheheadersaysnewsession.Thatistoomuchvocabularyforthesamebasicintent.Iwantonemainbuttonlabeled"StartDictation"withthemicrophoneiconsupportingthewordsratherthanreplacingthementirely.DoesStartDictationrequestmicrophoneaccessimmediately,ordoesitopentherecordingscreenfirst?Rightnow,thebrowserpermissionpromptappearsafterthesecondclick.Iwouldloveto

[00:00:00] Thesettingspage.Thehomepagehasthreethingsthatlookliketheprimaryaction,andonlyoneofthemstartsdictation.Thelargebuttonsayscreate,themicrophonetilesayscapture,andthesmallbuttonintheheadersaysnewsession.Thatistoomuchvocabularyforthesamebasicintent.Iwantonemainbuttonlabeledstartdictation,withthemicrophoneiconsupportingthewords,ratherthanreplacingthementirely.Doesstartdictationrequestmicrophoneaccessimmediately,ordoesitopentherecordingscreenfirst?Rightnow,thebrowserpermissionpromptappearsafterthesecondclick.Iwouldlovetoremoveastep,butIdonotwantthepageaskingformicrophoneaccessbeforesomebodyunderstandswhattheyarestarting.ThatisespeciallyawkwardiftheyaresharingtheirscreenduringapartneronboardingcallwithPriya.Itopenstherecordingscreenfirst,andthatscreenexplainswhathappenswhenyoupressrecord.Iamnotproposingsurpriserecordingorsurprisepermissions.Forreturninguserswhoalreadygrantedaccess,wecanexploreafasterpathlater.Forthisrefresh,themainimprovementisclarity.Oneentrance,onelabel,andavisibleexplanationofwhetherthemicrophoneisready,unavailable,orwaitingforpermissionfromthebrowser.Thathelpsonboarding.Intwointakecalls,peoplethoughtcreatemeantcreatingaproject,sotheywentlookingforaseparaterecordingbutton.Onepersonopenedsettingsbecausethemicrophoneicontherelookedmorerecognizablethanthemainaction.Idonotthinkthoseusersfailedadiscoverabilitytest.Ithinkourinterfaceaskedthemtoguesswhichofthreenearsynonymsmatchedaninvisibledistinctionwehadneverexplained.Agreed.Imean,iftheproductisfordictation,wecanprobablysurviveputtingtheworddictationonthebutton.So,Jill,canthesamelabelsurviveonanarrowmobilescreenwithoutbecomingthreelinestall?Iamalsocuriouswhetherweneedasecondaryactionforuploadingaclip,orwhetherthatshouldstayavailablebutclearlylessprominentthanspeakingintothemicrophone.Themobileversionfitscomfortablywithashorterhelpersentencebeneathit.Uploadaudiobecomesatextbuttonbelowthemainaction,notanotherbrightblockcompetingforattention.Theiconhasalargertouchtargetthanitappearsto,andthevisiblelabelremainspartoftheaccessiblename.Ialsocheckedkeyboardorder,becausethecurrentheaderactiongrabsfocusbeforethemainpageheading,whichmakesthescreenfeeloddlybackward.Couldtheuploadactiontellpeoplewhichformatsworkbeforetheychooseafile?Formyevaluationclips,Ikeptopeningthepickerandthenrememberingthatoneexportformatwasunsupported.Itisnotahugeproblemwhenyouknowtheproduct,butafirst-timeusercouldinterpretafaileduploadasabrokentranscriptionsystem.Ashort,supportedformatsnotewouldalsosavemefrommaintainingastickynotebesidemylaptop.Yes,andthatnotecansharespacewiththesizelimit,ratherthanbecominganothermodal.Thecolorissueisseparatebutrelatedtotrust.OurmainpurplecurrentlylooksclosetoAIbuilderspacepurplewithoutactuallymatchingit.Onthepartnerlandingpage,thetwositnexttoeachotherandlooklikeanaccidentalmismatch.Weneedeitheradeliberatevoiceflowcolororanapprovedexactpartnertreatment,notanalmostcopy.Priyanoticedthattoo.ShewasnotaskingustobecomeanAIbuilderspacesub-brand,shejustwantedthepartnerbadgetousetheapprovedasset.IwouldpreferVoiceFlowtokeepitsownidentity.Thecurrentpurplecamefromanearlyscreenshot,notabranddecision.Soweshouldnotspendaweekdefendingitasthoughourusershaveformedadeepemotionalattachmenttothatexactshade.Thenletuschoosedeliberateseparation.Imean,anearmatchistheworstofbothworlds.Itlooksborrowedandstilllookswrong.SousetheapprovedAIbuilderspacebadgewhereweidentifytheintegrationandpickaVoiceFlowcolorbasedonreadabilityandtherestoftheinterface.Pleasedonotmakethedecisiondependonhowitlooksononlyyourverynicemonitor.Ihaveadarkerblue-greencandidatethatkeepstheinterfacecalmerandgivesusstrongcontrastwithwhitetext.Icheckedtheprimarybutton,disabledstates,focusoutlines,anderrorbannersseparately.Theerrorcolorstaysdistinct.Wearenotmakingeverystateavariationofthebrandcolor.Iwillsendscreenshotsandaclickableprototype,butIalsowanttwoquicksessionsonactuallaptopsbeforewecallthevisualdecisionfinished.Pleaseincludealoadingstateforatranscriptthatisreadybutstillwaitingonthevocabularypass.AfterSears'timeoutchange,thatstateshouldbeshort,butitstillexists.Weshouldavoidabuttonthatsaysdonewhilethetextisvisiblychangingunderneathit.Maybethebasetranscriptappearsimmediatelywithaquietrefiningnamesstatus,providedwecanexplainthatwithoutoverstatingwhatthere-rankeractuallyguarantees.Iwouldrathersaycheckingvocabularythanrefiningnames,becausethemodelcanchangeotherdomaintermstoo.ButIneedtoseethetimingbeforeaddinganewvisiblestate.Ifthepassusuallyfinishesbeforeapersoncanreadthefirstline,alittlestatusbadgemaycreatemoredistractionthanclarity.Jackson,sendmearecordingfromthestagingreplaysoIcandesignfortheactualtransition.Thatisagoodexampleofwhythedeploymentnumbersmatteroutsideinfrastructure.Weshouldusetheconservativereplay,notthefriendliestcacheresultforthatrecording.Searsysaidtheshuffledlistprefillwas210milliseconds,plusqueuingandtherestoftherequest.Theuserseesthefullsequence,notaprefillchart.Iwantourdesignreviewtolookatthatfullexperienceunderalittlepressure.Movingtosettings,Iamgroupingthepageintorecording,vocabulary,privacy,andworkspaceaccess.Rightnow,personalmicrophonepreferencessitbesidesharedvocabulary,andpeoplecannottellwhichchangesaffectcolleagues.Everysharedcontrolwillsaythatitappliestotheworkspace.Personalcontrolswillsayonlyyou.Iwantthescopeinplaintextnearthecontrol,nothiddeninsideatooltipthatmobileusersmayneverdiscover.Forvocabulary,canweshowanexampleentrywithbothspellingandashortcontext?Somethinglike,aperson'spreferrednameandthefactthatthey'reateammate,withoutencouraginguserstopasteabiography.Weshouldalsosaythataddinganentryimprovesthehintwesend,butdoesnotguaranteerecognition.Thatwouldmatchtheevaluationuncertaintybetterthanabigsavebuttonfollowedbyacheckmarkthatfeelslikecertification.Yes,althoughcontextsupportneedsJackson'sconfirmationbeforeitbecomesarealinput.Forthefirstversion,Icanshowpreferredspellingwithamodestexplanationofwhathintsdo.Thesaveconfirmationcansay,vocabularylistupdated.Separately,deletinganentryneedsaclearsharedscopewarning,butnotascaryfull-screendialogforeverychange.Anundooptionwouldbemoreforgivingifthebackendcansupportitsafely.Preferredspellingissupported.Freeformcontextisnotready.Pre'sAPInotesincludeaconstrainedhintstructure,andIwanttostayinsidethatforthefirstrollout.Anundoispossibleifwekeepthepreviouslistversionbrieflyandenforceworkspacepermissionsonrestore.IwillgiveyoutheexactbehaviorbyWednesday.Pleasedesignthepermissiondeniedcasetoo,becauseviewersshouldnotdiscovertheiraccesslevelonlyaftereditingawholelist.Iwillrecruittwocurrentusersforthesettingscheck,includingsomebodywhoisnotaworkspaceadministrator.Weshouldaskthemtopredictwhathappensbeforetheysaveachange.Iftheythinkamicrophonepreferencechangeseverybody'sdevice,oravocabularycorrectionstaysprivatewhenitisshared,wehaveacomprehensionissue,eveniftheycantechnicallyfinishthetask.Thatismorevaluablethanaskingwhetherthescreenlookscleaner.Great.Thedesigndecisionisonecleardictationaction,adistinctvoiceflowcolorwiththeapprovedpartnerbadge,andsettingsthatmakescopevisible.Schochulownstheprototype,Jacksonownsthebehaviorcontract,andCaitlinownstheusersessions.Wewillputdatesonthoseattheend.Aoife,thatvocabularydiscussionisausefulbridgeintoyourclips,sotakeusthroughwhatyouactuallymeasuredratherthanstartingwithamodelproposal.Ibuiltanevaluationsetof100hand-labeledclips.Eachcliphasareferencetranscript,markednamespans,speakerconditions,andnotesonbackgroundnoise.TheclipsareshortenoughthatIcouldrevieweverylabeltwice,buttheyincludecompletesentencesratherthanisolatednames.Ihavenotusedtheseclipstotrainanything.Theyaremeanttotelluswhetherachangehelps,includingchangesthatsoundsensiblebutmaynotwork.Beforewegettheresults,howdidyouchoosethenames?Iwanttomakesurewedidnotaccidentallycreateatestconsistingmostlyofourownteamrepeatedinslightlydifferentsentences.Thatwouldbeusefulfordebugging,butlessusefulasalaunchgate.Also,didthepeoplerecordingtheclipsknowwhichnamethesystemwasexpectedtorecognize,orweretheyreadingnaturalworkplacematerialwithoutspecialcoaching?Therearefamiliarnames,includingJackson,Caitlin,Sochitel,Saoirse,Kaveh,andPriya,buttheyareonlypartoftheset.Iincludedothernameswithvariedspellingpatternsandlanguageorigins,plusorganizationandprojectnames.Speakersreadshortworkplacepromptsconversationally.Theysawthetext,soitisstillreadspeech,andIamlabelingthatlimitationexplicitly.Weshouldnotpretendthissetcapturesallspontaneousdictationoreveryspeakercommunity.Thatisasensiblestartingpoint.Didyoubalancecleanandnoisyclips,orisnoisejustwhateverhappenedduringrecording?Iamaskingbecauseintakecallshappeninactualkitchens,offices,andcarsparkednearroads.Ifanameimprovementonlyworksinaquietroomwithacarefulspeaker,weshouldunderstandthatbeforeweputthenewcohort'sexperiencebehindaveryconfidentlaunchstatementaboutbetterrecognition.Sixtyclipsarerelativelyclean,twentyhavesteadybackgroundnoise,andtwentyhavemildinterruptionsorconversationalhesitations.Ididnotincludespeechthatisgenuinelyunintelligibletoahumanreviewer.Thatwouldtestadifferentboundary.Ialsokeptanotewhenanamehadmultipleplausiblespellings,becauseexactmatchcanbeharshwithoutasuppliedvocabularylist.Thosecasesremainintheresults,butIcanshowthemseparatelyratherthanquietlyforgivingthem.Good.Imean,ausefulevaluationisallowedtobeimperfect,aslongasitslimitsarevisible.Telluswhatnamerecallmeansinyourreport,becauseIcanimaginethreepeopleinthismeetingusingthatphrasedifferently.Then,giveusthebaseline,thehintedversion,andtheplaceswheretheaveragehidesafailureweshouldcareaboutbeforeweinviteanothergroupofusers.Namerecallhereistheshareoflabelednamementionsthatthesystemreturnswiththeexactexpectedspelling.Thereare160namementionsacrossthe100clips.Withoutvocabularyhints,wewerecovered96,whichis60%.Withthesupportedhintformat,wewerecovered128,whichis80%.Thatisameaningfulimprovementontheset,notapromisethateveryuser'snameaccuracywillbe80%.Thosenumbersareencouraging.Didthehintedrequestsalsointroducenamesthatwerenotspoken?Asystemcouldimproverecallbyeagerlyinsertingvocabularyentrieswhenevertheaudioisuncertain,andthatwouldbeaterribletradeformeetingnotes.Iwantthefalseadditionsbesiderecall.Ifpreviouslistcontains10colleagues,wedonotwantall10turningupasimaginaryattendeesjustbecausethemodellikeshavingsomethingspecifictochoose.Icheckedthat.Thebaselineintroducedtwoincorrectnamementions,andthehintedversionintroducedfive.Thatisasmallcount,butthedirectionmatters.Threeofthefivecamefromclipswhereanordinarywordsoundedclosetoasuppliedname.Ialsomeasuredoverallworderrorrate,sowecanseewhetherimprovingnamematchesdamagesthesurroundingsentence.Theaggregatebarelychanged,whichiswhyasingleoverallaccuracynumberwouldhavehiddentheusefuldifference.Thatmakestheinterfacewordingevenmoreimportant.Ifavocabularylistcanoccasionallyencourageafalsematch,weshouldcallitahintandkeepeditingeasy.Ialsowonderwhethershowingtoomanysuggestedentriesduringonboardingwouldmaketheproblemworse.Wecouldencourageashortlistoffrequentlyspokennamesinsteadofimplyingthatimportinganentiredirectoryisthemostprofessionalorcompletewaytosetupvoiceflow.Iagree,andIwanttotestlistlengthdirectly.Rightnow,everyhintedrequestusesonlytherelevantsmalllistforthatfictionalworkspace.Ihavenottestedadirectorywithhundredsofentries.Beforewerecommendbulkimport,weneeddistractornamesandnearmatches.Thatalsooverlapswithsearch'slessfriendlyreplay—alternatinglistscanstresstheKVcache,whilesimilarnamescanstresstherankingqualityinthesamerun.Wecanbuildthatastwoseparatemeasurementsusingthesamerequestfixtures.OnerunmeasurestheAPIoutputagainstyourreferences,anothermeasureslatencyandresourcebehavior.Idonotwantperformanceloggingtoaccidentallyleakthereferencenamesormergeintoproductiontelemetry.Ifa,ifyougivemefixtureidentifiersandexpectedresultsinaseparatefile,Icankeeptherequesttraceusefulwithoutmakingthelogsasecondcopyofthedataset.Thatworks.OntheLoRAquestion,Idonotrecommendtrainingoneyet.Ahundredclipsareanevaluationstartingpoint,notaconvincingtrainingcorpus.Andtrainingontheevaluationsetwoulddestroyitsusefulnessasacleantest.Wealsoneedtospecifywhichmodelwewouldadapt.ALoRAonthetextre-rankerisnotthesameinterventionasadaptingaspeechmodel,andneitherautomaticallyfixeseverynamethatthetranscriptionAPImisses.Thankyouformakingthatdistinction.Imean,weshouldnotletaconcretetechniquebecometheplanbeforewehavedescribedthefailureitismeanttosolve.Jackson,isthereanyreasontorushintoadaptationnow,otherthanthefactthatitsoundsmoreimpressivethanpassingawell-formedvocabularyhint?Iwouldratherspendthisweekprovingthesimplepathworksandunderstandingwhereitstops.Nourgentreason.Thestrongestnear-termimprovementsarereliablehints,boundedre-ranking,andacorrectionflowthatstorestheuser'sintendedspellingattherightscope.Eloracouldbecomeusefulifwefindastablepatternofdomainerrorsandhaveenoughconsentedtrainingexamples,butwewouldstillneedaheld-outtest,adeploymentstory,andevidencethatthegainssurviveoutsidethetrainingdistribution.Thatisaproject,notaThursdayafternooncheckbox.Thenletusdocumentthedecisionasdefertraining,notrejectitforever.Otherwise,inamonth,somebodywillrememberthisasthemeetingwherewedecidedadaptationwaspointless.Ifa,whatwouldmakeyoureopenit?Iwouldlikeaconcretetriggerinvolvingerrorpatternsandavailabledata,notavaguestatementthatweshouldtrainoncewefeelmorematureasacompanyorgetboredwiththecurrenttools.Iwouldreopenitifthehintpathplateaus,theremainingerrorsclusterinarepeatableway,andwecancollectaseparateconsentedcorpuslargeenoughforameaningfulexperiment.Wewouldreserveuntouchedtestmaterialbeforetraining.Iwouldalsowantabaselinethatincludessimplerfixes,sotheLoRAhastobeatsomethingreal.Thetriggerisevidencethatadaptationaddressesapersistentfailure,notmerelythatourcalendarhasreachedanewmonth.WheredoesMLXfitintothat?Isawthenoteinyourexperimentdocumentandwasnotsurewhetheritwaspartoftheproductionrecommendationorjusthowyouwererunningthingslocally.Ifitislocal,thatisfine,butIwantthedesignandproductnotestoavoidturninganexperimentonsomebody'slaptopintoaclaimthattheproductnowprocesseseveryone'saudioontheirowndevice.MLXisonlyforalocalexperimentonmylaptoprightnow.IusedittocheckthatIcouldrunasmalltextmodelandreproduceabasicscoringtaskwithoutbookingsharedhardware.ItisnotpartofVoiceFlow'sproductionpath,anditdoesnotmakeourdictationon-device.IfwelatertestElarawithMLX,thatwouldbeanexperimentwithseparatedataandexplicitgoals,notanunannouncedchangetoourprivacymodel.Thatexplanationbelongsinthenotes,almostwordforword.Ialsowanttheevaluationreporttodistinguishinternaltestclipsfromrealcustomerrecordings,becausepeopleoftenheardatasetandassumewearequietlycollectingeverything.Priyawillaskaboutprovenancebeforesheencouragespartneruserstoparticipate.Weshouldhaveasimpleansweraboutconsent,storage,access,anddeletion,ratherthanassemblingthatanswerfromseveraldifferentchatthreadsaftersheasks.Icanaddadatasetcardwiththosedetailsandthelimitationswediscussed.Thecurrentclipswererecordedforthisevaluationwithpermission,andthelabelsareaccessibleonlytotheprojectgroup.IwillaskJacksontoreviewthestoragesettingsratherthanjusttrustingthefoldername.Forthenextbatch,Iwantspontaneousdictationwithexplicitconsent,butthatwillbeaseparatecollectioneffortandnotaconditionforthisweek'sinfrastructuremove.Excellent.Theimmediatecallistokeepthe100clipsasevaluationmaterial,shipthesupportedhintpathbehindacontrolledrollout,andtesttheharderlistsbeforeexpandingconfidence.Wearenottrainingalowerthisweek,andMLXstaysaninternalexperiment.Aoife,yourreportshouldincludethefalseadditionsasprominentlyastherecallgain.Letusturnallofthatintoassignmentswithdatessotheusefuldetailsurvivesthiscall.Jackson,byTuesday,June9th,youowntheintakevocabularywiring,includingtheentrylimit,safetracefields,andthetimeoutbehaviorthatreturnsthebasetranscript.PleasesendPriyatheformatversionwesupportandaskhertoconfirmtherequestshape.Imean,thisisonedeliverablewithafewchecks,notpermissiontobuildageneralvocabularyplatformbeforeweletanotherpersonusetheexistingproduct.IhavethatforTuesday,June9th.Iwillputthebehaviorcontractinthelaunchdocumentandlinkthetestsratherthanpastingawalloflogs.Searchit,Iwillalsosendthestagingrecordingofthebasetranscriptfollowedbythevocabularypass.Fortherollback,IwillpairwithCirceonWednesday,June10th,andtestbothacleansurfaceerrorandaconnectionthatsimplyhangsuntilourtimeoutfires.Saoirse,youownthewrittengoorno-gobyWednesday,June10th,andthemigrationwindowonThursday,June11that10inthemorningPacific.Yournoteneedstheroutingflag,queuethreshold,rollbackcommand,andbothcostviews.Wearecarrying53%KVcachehitrateand210millisecondsmedianprefillastheconservativereplayfigures,notpresentingthefriendlier68%resultasaguarantee.Aoife,byWednesday,June10th,youowntheharderreplayfixturesandtheevaluationreport,includingthe100hand-labeledclips,exactnamerecall,falsenameadditions,anddatasetprovenance.PleasecoordinatethefixtureformatwithJacksonandsendasanitizedrequestshapestoSaoirse.ThereportshouldsayexplicitlythattheclipshavenotbeenusedfortrainingandthattheMLXexperimentisseparatefromtheproductionpath.Thatisclear.IwilldeliverthereportandfixturesWednesday,June10thwiththereadspeechlimitationupfrontratherthanburiedatthebottom.Saoirsewillgetalternatingworkspaces,editedlists,andsimilarnamedistractors.Caitlin,IwillalsogiveyouashorterexplanationforPriyasoshecandiscusstheevaluationwithpartnerswithouthavingtotranslateatechnicalappendixduringanonboardingcalloraccidentallyimplytheresultsapplyuniversally.Caitlin,byTuesday,June9th,youownPriya'sscheduleconfirmation,therevisedcontrolledrolloutdocument,andthetenworkspacerecruitmentplan.ByFriday,June12th,youalsoownthemetricdefinitionsforactivation,completeddictationdailyactives,weeklyactives,andreturnusebycohort.Kaveaskedforhonestdenominators,soincludecohortsizeandobservationwindowdirectlyonthepage.Donotwaituntilwehavebeautifulretentionresultstodecidewhatretentionmeans.Confirmedonbothdates.Iwillseparaterecruitmentfromactualactivation,sowedonotcountaninvitedworkspaceasausersuccess.Priyagetstheupdatedlaunchboundariesandsupportlabelstomorrow.IwillalsoschedulethetwosettingsessionsforFriday,June12th,withoneadministratorandonenon-administrator.Xochitl,IwillsendthesessiontimesbeforeIsendinvitations,soyouarenotaccidentallydouble-bookedwiththereviewyouneedtolead.Xochitl,byWednesday,June10th,youowntheprototypewiththestartdictationbutton,distinctvoiceflowcolor,approvedAIbuilderspacebadge,andsettingscopelabels.ByMonday,June15th,youowntherevisionsfromFriday'susersessions.Pleaseincludeuploadformats,microphonepermissionstates,andthevocabularysavelanguage.Imean,thepointisfewerguessesforusers,notmerelyacleanerscreenshotthatstillleavesthesamequestionsunanswered.IhaveWednesdayfortheprototypeandMonday,June15thfortherevisions.Iwillkeepunsupportedcontextfieldsoutofthefirstversionandmakepermissionsvisiblebeforeediting.Jackson,ifundoturnsouttobemorecomplicatedthanexpected,tellmeWednesdayandIwilluseasimplerconfirmationflow.Wecanmakeatrustworthyfirstversionwithoutpretendingeveryconveniencefeatureisnecessaryforlaunchorworthdelayingtheclearermainaction.IownthefinalcohortexpansiondecisiononMonday,June15th.AfterJackson'ssafetychecks,officereadout,Saoirse'smigrationobservation,andtheinitialdesignsessionsareavailable.IwillalsoreviewtheproviderdisclosurewithCaitlinbyWednesday,June10th.Theexpansionisstillcontrolled,notapubliclaunch.IfThursdayrollsback,thatisinformationforthedecision,notanautomaticfailure,andnotsomethingwehidebehindagreenstatuslabel.Onefinaldependencycheck.Thenamehintworkcanshipindependentlyofmovingthere-ranker,correct?Iwantthatexplicitbecauseitmeanswecanstilllearnfromthenextcohortiftheboxneedsmoretuning.ThesupportedhintformatgoestotheAIbuilderspacetranscriptionAPIeitherway.Weshouldnotmaketheentirequalityexperimentdependonprovingthenewinfrastructurepathonthesameday,eventhoughtheysharesomefixtures.Correct.Thoseareindependentswitches,andIwillputthatinthedecisionnote.Thanks,everybody.Wehaveaboundedrollout,arealexplanationfortheKoyakincrease,andanevaluationplanthatdoesnotconfuseapromisingresultwithauniversalclaim.Sierra,safetravels,andpleasestayoffthemigrationcontrolsuntilyouarehome.Ournextbi-weeklysyncisMonday,June22nd.Untilthen,pulseblockerswiththeowneranddecisionneeded.