[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-\u002Fhe\u002Fdocs\u002Fsupervision\u002Fbenchmarks":3,"docs-nav-he":466},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"title":8,"description":9,"body":10,"_type":460,"_id":461,"_source":462,"_file":463,"_stem":464,"_extension":465},"\u002Fhe\u002Fdocs\u002Fsupervision\u002Fbenchmarks","supervision",false,"","Benchmarks","מדידת איכות pipeline עם תרחישים ניתנים לשחזור. זיהוי רגרסיות לפני הפרודקשן.",{"type":11,"children":12,"toc":451},"root",[13,34,40,45,65,70,76,81,104,110,266,271,277,302,331,341,347,366,371,403,408],{"type":14,"tag":15,"props":16,"children":17},"element","p",{},[18,25,27,32],{"type":14,"tag":19,"props":20,"children":21},"strong",{},[22],{"type":23,"value":24},"text","benchmark",{"type":23,"value":26}," הוא אוסף תרחישי הערכה שמופעלים אוטומטית מול pipeline. זהו הכלי למדידה ",{"type":14,"tag":19,"props":28,"children":29},{},[30],{"type":23,"value":31},"אובייקטיבית",{"type":23,"value":33}," של איכות agent ולגילוי רגרסיות לפני שהן מגיעות לפרודקשן.",{"type":14,"tag":35,"props":36,"children":38},"h2",{"id":37},"מדוע",[39],{"type":23,"value":37},{"type":14,"tag":15,"props":41,"children":42},{},[43],{"type":23,"value":44},"agent ש« נראה כאילו עובד » יכול לרגרס בשקט לאחר:",{"type":14,"tag":46,"props":47,"children":48},"ul",{},[49,55,60],{"type":14,"tag":50,"props":51,"children":52},"li",{},[53],{"type":23,"value":54},"שינוי משימה;",{"type":14,"tag":50,"props":56,"children":57},{},[58],{"type":23,"value":59},"שינוי מודל LLM;",{"type":14,"tag":50,"props":61,"children":62},{},[63],{"type":23,"value":64},"הוספת tool חדש.",{"type":14,"tag":15,"props":66,"children":67},{},[68],{"type":23,"value":69},"ללא benchmark, מגלים זאת בפרודקשן. עם benchmark, רואים זאת בהרצה הבאה — לפני המשתמשים שלכם.",{"type":14,"tag":35,"props":71,"children":73},{"id":72},"אנטומיה-של-תרחיש",[74],{"type":23,"value":75},"אנטומיה של תרחיש",{"type":14,"tag":15,"props":77,"children":78},{},[79],{"type":23,"value":80},"תרחיש מכיל:",{"type":14,"tag":46,"props":82,"children":83},{},[84,94],{"type":14,"tag":50,"props":85,"children":86},{},[87,92],{"type":14,"tag":19,"props":88,"children":89},{},[90],{"type":23,"value":91},"גירוי",{"type":23,"value":93}," — הקלט שה-pipeline מקבל (הודעה או רצף תורות, fixtures של נתונים).",{"type":14,"tag":50,"props":95,"children":96},{},[97,102],{"type":14,"tag":19,"props":98,"children":99},{},[100],{"type":23,"value":101},"checks",{"type":23,"value":103}," — אסרציות מפורשות המוערכות על כל תור: הפלט שיוצר, צורת התגובה, שרשרת קריאות ה-tools, עלות הפעולות.",{"type":14,"tag":35,"props":105,"children":107},{"id":106},"קטלוג-checks",[108],{"type":23,"value":109},"קטלוג checks",{"type":14,"tag":111,"props":112,"children":113},"table",{},[114,133],{"type":14,"tag":115,"props":116,"children":117},"thead",{},[118],{"type":14,"tag":119,"props":120,"children":121},"tr",{},[122,128],{"type":14,"tag":123,"props":124,"children":125},"th",{},[126],{"type":23,"value":127},"Check",{"type":14,"tag":123,"props":129,"children":130},{},[131],{"type":23,"value":132},"בודק",{"type":14,"tag":134,"props":135,"children":136},"tbody",{},[137,156,178,195,217,234],{"type":14,"tag":119,"props":138,"children":139},{},[140,151],{"type":14,"tag":141,"props":142,"children":143},"td",{},[144],{"type":14,"tag":145,"props":146,"children":148},"code",{"className":147},[],[149],{"type":23,"value":150},"pipeline_output_match",{"type":14,"tag":141,"props":152,"children":153},{},[154],{"type":23,"value":155},"מפתח בזרימת הפלט תואם לערך הצפוי (נתיב + אסרציה)",{"type":14,"tag":119,"props":157,"children":158},{},[159,168],{"type":14,"tag":141,"props":160,"children":161},{},[162],{"type":14,"tag":145,"props":163,"children":165},{"className":164},[],[166],{"type":23,"value":167},"response_shape",{"type":14,"tag":141,"props":169,"children":170},{},[171,176],{"type":14,"tag":19,"props":172,"children":173},{},[174],{"type":23,"value":175},"צורת",{"type":23,"value":177}," התגובה של ה-agent (אורך, מבנה), נמדדת ללא שופט",{"type":14,"tag":119,"props":179,"children":180},{},[181,190],{"type":14,"tag":141,"props":182,"children":183},{},[184],{"type":14,"tag":145,"props":185,"children":187},{"className":186},[],[188],{"type":23,"value":189},"llm_rubric",{"type":14,"tag":141,"props":191,"children":192},{},[193],{"type":23,"value":194},"תוכן התגובה, מדורג על-ידי שופט LLM לפי רובריקה של קריטריונים ניתנים לאימות",{"type":14,"tag":119,"props":196,"children":197},{},[198,207],{"type":14,"tag":141,"props":199,"children":200},{},[201],{"type":14,"tag":145,"props":202,"children":204},{"className":203},[],[205],{"type":23,"value":206},"tool_chain_analysis",{"type":14,"tag":141,"props":208,"children":209},{},[210,215],{"type":14,"tag":19,"props":211,"children":212},{},[213],{"type":23,"value":214},"רצף",{"type":23,"value":216}," קריאות ה-tools תואם את האסטרטגיה הצפויה",{"type":14,"tag":119,"props":218,"children":219},{},[220,229],{"type":14,"tag":141,"props":221,"children":222},{},[223],{"type":14,"tag":145,"props":224,"children":226},{"className":225},[],[227],{"type":23,"value":228},"tool_cost",{"type":14,"tag":141,"props":230,"children":231},{},[232],{"type":23,"value":233},"תקציב הפעולות: לכל קריאה יש עלות, הגעה ליעד מזכה בבונוס",{"type":14,"tag":119,"props":235,"children":236},{},[237,261],{"type":14,"tag":141,"props":238,"children":239},{},[240,246,248,254,255],{"type":14,"tag":145,"props":241,"children":243},{"className":242},[],[244],{"type":23,"value":245},"static",{"type":23,"value":247}," \u002F ",{"type":14,"tag":145,"props":249,"children":251},{"className":250},[],[252],{"type":23,"value":253},"tests",{"type":23,"value":247},{"type":14,"tag":145,"props":256,"children":258},{"className":257},[],[259],{"type":23,"value":260},"diff_quality",{"type":14,"tag":141,"props":262,"children":263},{},[264],{"type":23,"value":265},"checks ממוקדי קבצים וקוד (פלט של pipeline פיתוח)",{"type":14,"tag":15,"props":267,"children":268},{},[269],{"type":23,"value":270},"הקטלוג ניתן להרחבה בצד הפלטפורמה: כל check הוא plugin הרשום בצד השרת.",{"type":14,"tag":35,"props":272,"children":274},{"id":273},"יצירת-benchmark",[275],{"type":23,"value":276},"יצירת benchmark",{"type":14,"tag":278,"props":279,"children":280},"ol",{},[281,291],{"type":14,"tag":50,"props":282,"children":283},{},[284,289],{"type":14,"tag":19,"props":285,"children":286},{},[287],{"type":23,"value":288},"יישומים → הערכה → תרחישים",{"type":23,"value":290},".",{"type":14,"tag":50,"props":292,"children":293},{},[294,296,301],{"type":23,"value":295},"צרו את התרחישים (גירוי, checks, fixtures של נתונים); כל תרחיש מוקדש למודל אחד (ה-pipeline היעד). ה-fixtures מנוהלים בלשונית ",{"type":14,"tag":19,"props":297,"children":298},{},[299],{"type":23,"value":300},"Fixtures",{"type":23,"value":290},{"type":14,"tag":15,"props":303,"children":304},{},[305,307,314,316,322,323,329],{"type":23,"value":306},"ניתן גם ליצור או לשנות תרחיש באמצעות הצעה מתוך ",{"type":14,"tag":308,"props":309,"children":311},"a",{"href":310},"\u002Fhe\u002Fdocs\u002Fdeveloppeurs\u002Fmcp",[312],{"type":23,"value":313},"שרת ה-MCP",{"type":23,"value":315}," (",{"type":14,"tag":145,"props":317,"children":319},{"className":318},[],[320],{"type":23,"value":321},"benchmark_create",{"type":23,"value":247},{"type":14,"tag":145,"props":324,"children":326},{"className":325},[],[327],{"type":23,"value":328},"benchmark_update",{"type":23,"value":330},"), שנשארת בהמתנה עד לקבלתה.",{"type":14,"tag":332,"props":333,"children":335},"alert",{"type":334},"info",[336],{"type":14,"tag":15,"props":337,"children":338},{},[339],{"type":23,"value":340},"benchmark טוב מכיל תמהיל: מקרה סטנדרטי, מקרי קצה, מקרים עוינים (ניסיונות מניפולציה, עמימות, נתונים חסרים).",{"type":14,"tag":35,"props":342,"children":344},{"id":343},"הפעלת-benchmark",[345],{"type":23,"value":346},"הפעלת benchmark",{"type":14,"tag":15,"props":348,"children":349},{},[350,352,357,359,364],{"type":23,"value":351},"מממשק הניהול, הפעילו ",{"type":14,"tag":19,"props":353,"children":354},{},[355],{"type":23,"value":356},"הרצה",{"type":23,"value":358},": היא נכנסת לתור עמיד ואז מבוצעת על-ידי worker ייעודי, מחוץ לשרת ה-web. אם ה-worker הזה נעצר באמצע הרצה, ההרצה מסומנת ",{"type":14,"tag":19,"props":360,"children":361},{},[362],{"type":23,"value":363},"כנכשלה",{"type":23,"value":365}," בהפעלה מחדש (היא אינה מתחדשת אוטומטית): הפעילו אותה שוב. drawer מציג את הפירוט לכל תרחיש.",{"type":14,"tag":15,"props":367,"children":368},{},[369],{"type":23,"value":370},"הדוח מציין:",{"type":14,"tag":46,"props":372,"children":373},{},[374,379,398],{"type":14,"tag":50,"props":375,"children":376},{},[377],{"type":23,"value":378},"שיעור הצלחה לפי check",{"type":14,"tag":50,"props":380,"children":381},{},[382,384,389,391,396],{"type":23,"value":383},"התפתחות הציונים בין ההרצות (לשוניות ",{"type":14,"tag":19,"props":385,"children":386},{},[387],{"type":23,"value":388},"היסטוריה",{"type":23,"value":390}," ו-",{"type":14,"tag":19,"props":392,"children":393},{},[394],{"type":23,"value":395},"התפתחות",{"type":23,"value":397},")",{"type":14,"tag":50,"props":399,"children":400},{},[401],{"type":23,"value":402},"רשימת התרחישים שנכשלו, עם פירוט כל check",{"type":14,"tag":35,"props":404,"children":406},{"id":405},"המלצות",[407],{"type":23,"value":405},{"type":14,"tag":46,"props":409,"children":410},{},[411,421,431,441],{"type":14,"tag":50,"props":412,"children":413},{},[414,419],{"type":14,"tag":19,"props":415,"children":416},{},[417],{"type":23,"value":418},"התחילו קטן",{"type":23,"value":420}," — 5 תרחישים ממוקדים שווים יותר מ-50 תרחישים גסים.",{"type":14,"tag":50,"props":422,"children":423},{},[424,429],{"type":14,"tag":19,"props":425,"children":426},{},[427],{"type":23,"value":428},"הריצו benchmark לפני כל פרסום",{"type":23,"value":430},", והשוו להרצה הקודמת לפני הדחיפה.",{"type":14,"tag":50,"props":432,"children":433},{},[434,439],{"type":14,"tag":19,"props":435,"children":436},{},[437],{"type":23,"value":438},"גרסו את המשימות.",{"type":23,"value":440}," לא רק pipelines: גם המשימות מתפתחות.",{"type":14,"tag":50,"props":442,"children":443},{},[444,449],{"type":14,"tag":19,"props":445,"children":446},{},[447],{"type":23,"value":448},"שמרו על קריאות ה-checks.",{"type":23,"value":450}," check שדורש 20 שורות הסבר כדי להבין אותו הוא check שצריך לפרק.",{"title":7,"searchDepth":452,"depth":452,"links":453},2,[454,455,456,457,458,459],{"id":37,"depth":452,"text":37},{"id":72,"depth":452,"text":75},{"id":106,"depth":452,"text":109},{"id":273,"depth":452,"text":276},{"id":343,"depth":452,"text":346},{"id":405,"depth":452,"text":405},"markdown","content:he:docs:4.supervision:3.benchmarks.md","content","he\u002Fdocs\u002F4.supervision\u002F3.benchmarks.md","he\u002Fdocs\u002F4.supervision\u002F3.benchmarks","md",[467],{"title":468,"_path":469,"children":470},"He","\u002Fhe",[471],{"title":472,"_path":473,"children":474},"Docs","\u002Fhe\u002Fdocs",[475,488,516,544,555,571,583,644],{"title":476,"_path":477,"children":478},"תחילת העבודה","\u002Fhe\u002Fdocs\u002Fdemarrer",[479,482,485],{"title":480,"_path":481},"ברוכים הבאים","\u002Fhe\u002Fdocs\u002Fdemarrer\u002Fbienvenue",{"title":483,"_path":484},"יצירת חשבון","\u002Fhe\u002Fdocs\u002Fdemarrer\u002Fcreer-un-compte",{"title":486,"_path":487},"חיבור מפתח LLM","\u002Fhe\u002Fdocs\u002Fdemarrer\u002Fbrancher-llm",{"title":489,"_path":490,"children":491},"ערוצים","\u002Fhe\u002Fdocs\u002Fcanaux",[492,495,498,501,504,507,510,513],{"title":493,"_path":494},"סקירת הערוצים","\u002Fhe\u002Fdocs\u002Fcanaux\u002Fvue-densemble",{"title":496,"_path":497},"טלפוניה בזמן אמת","\u002Fhe\u002Fdocs\u002Fcanaux\u002Ftelephone",{"title":499,"_path":500},"אימייל","\u002Fhe\u002Fdocs\u002Fcanaux\u002Femail",{"title":502,"_path":503},"Chat widget לאתר","\u002Fhe\u002Fdocs\u002Fcanaux\u002Fchat-widget",{"title":505,"_path":506},"Webhook HTTP","\u002Fhe\u002Fdocs\u002Fcanaux\u002Fwebhook",{"title":508,"_path":509},"מסרים מיידיים (WhatsApp, Messenger, Telegram, Mattermost)","\u002Fhe\u002Fdocs\u002Fcanaux\u002Fmessageries",{"title":511,"_path":512},"Meta Lead Ads","\u002Fhe\u002Fdocs\u002Fcanaux\u002Ffacebook-leads",{"title":514,"_path":515},"תזמון (cron)","\u002Fhe\u002Fdocs\u002Fcanaux\u002Fschedule",{"title":517,"_path":518,"children":519},"פייפליינים","\u002Fhe\u002Fdocs\u002Fpipelines",[520,523,526,529,532,535,538,541],{"title":521,"_path":522},"העורך הויזואלי","\u002Fhe\u002Fdocs\u002Fpipelines\u002Fediteur-visuel",{"title":524,"_path":525},"סוכני AI","\u002Fhe\u002Fdocs\u002Fpipelines\u002Fagents",{"title":527,"_path":528},"אופרטורים דטרמיניסטיים","\u002Fhe\u002Fdocs\u002Fpipelines\u002Foperateurs",{"title":530,"_path":531},"בקרת זרימה","\u002Fhe\u002Fdocs\u002Fpipelines\u002Fcontrole-flux",{"title":533,"_path":534},"שלבי קוד","\u002Fhe\u002Fdocs\u002Fpipelines\u002Fcode-step",{"title":536,"_path":537},"תת-פייפליינים","\u002Fhe\u002Fdocs\u002Fpipelines\u002Fsous-pipelines",{"title":539,"_path":540},"פעולות ללא קוד","\u002Fhe\u002Fdocs\u002Fpipelines\u002Factions-no-code",{"title":542,"_path":543},"בסיס ידע","\u002Fhe\u002Fdocs\u002Fpipelines\u002Fbase-de-connaissances",{"title":545,"_path":546,"children":547},"פיקוח","\u002Fhe\u002Fdocs\u002Fsupervision",[548,551,554],{"title":549,"_path":550},"היסטוריית ביצועים","\u002Fhe\u002Fdocs\u002Fsupervision\u002Fexecutions",{"title":552,"_path":553},"קרדיטים ומדידת צריכה","\u002Fhe\u002Fdocs\u002Fsupervision\u002Fcredits-metering",{"title":8,"_path":4},{"title":556,"_path":557,"children":558},"אבטחה","\u002Fhe\u002Fdocs\u002Fsecurite",[559,562,565,568],{"title":560,"_path":561},"עקרונות אבטחה","\u002Fhe\u002Fdocs\u002Fsecurite\u002Fprincipes",{"title":563,"_path":564},"מולטי-טננט קפדני","\u002Fhe\u002Fdocs\u002Fsecurite\u002Fmulti-tenant",{"title":566,"_path":567},"ביקורת GDPR ויומן גישה","\u002Fhe\u002Fdocs\u002Fsecurite\u002Faudit-rgpd",{"title":569,"_path":570},"פריסה ריבונית","\u002Fhe\u002Fdocs\u002Fsecurite\u002Fdeploiement-souverain",{"title":572,"_path":573,"children":574},"מפתחים","\u002Fhe\u002Fdocs\u002Fdeveloppeurs",[575,578,581],{"title":576,"_path":577},"API admin","\u002Fhe\u002Fdocs\u002Fdeveloppeurs\u002Fapi-admin",{"title":579,"_path":580},"Webhooks יוצאים","\u002Fhe\u002Fdocs\u002Fdeveloppeurs\u002Fwebhooks-sortants",{"title":582,"_path":310},"שליטה ב-betool מתוך Claude (MCP)",{"title":584,"_path":585,"children":586},"אינטגרציות","\u002Fhe\u002Fdocs\u002Fintegrations",[587,590,593,596,599,602,605,608,611,614,617,620,623,626,629,632,635,638,641],{"title":588,"_path":589},"סקירת האינטגרציות","\u002Fhe\u002Fdocs\u002Fintegrations\u002Fvue-densemble",{"title":591,"_path":592},"Stripe","\u002Fhe\u002Fdocs\u002Fintegrations\u002Fstripe",{"title":594,"_path":595},"GitHub","\u002Fhe\u002Fdocs\u002Fintegrations\u002Fgithub",{"title":597,"_path":598},"Google","\u002Fhe\u002Fdocs\u002Fintegrations\u002Fgoogle",{"title":600,"_path":601},"LinkedIn Ads","\u002Fhe\u002Fdocs\u002Fintegrations\u002Flinkedin-ads",{"title":603,"_path":604},"Kavkom","\u002Fhe\u002Fdocs\u002Fintegrations\u002Fkavkom",{"title":606,"_path":607},"Jira","\u002Fhe\u002Fdocs\u002Fintegrations\u002Fjira",{"title":609,"_path":610},"Slack","\u002Fhe\u002Fdocs\u002Fintegrations\u002Fslack",{"title":612,"_path":613},"SMS","\u002Fhe\u002Fdocs\u002Fintegrations\u002Fsms",{"title":615,"_path":616},"Enable Banking (open banking)","\u002Fhe\u002Fdocs\u002Fintegrations\u002Fenable-banking",{"title":618,"_path":619},"CRM, שיווק ופרודוקטיביות","\u002Fhe\u002Fdocs\u002Fintegrations\u002Fcrm-marketing",{"title":621,"_path":622},"HeyGen (וידאו)","\u002Fhe\u002Fdocs\u002Fintegrations\u002Fheygen",{"title":624,"_path":625},"La Poste","\u002Fhe\u002Fdocs\u002Fintegrations\u002Fla-poste",{"title":627,"_path":628},"CRM עסקי (beTool Application, Qhare)","\u002Fhe\u002Fdocs\u002Fintegrations\u002Fcrm-metier",{"title":630,"_path":631},"Mitti (בדיקות שטח)","\u002Fhe\u002Fdocs\u002Fintegrations\u002Fmitti",{"title":633,"_path":634},"Meta (Messenger, Pages, Meta Ads)","\u002Fhe\u002Fdocs\u002Fintegrations\u002Fmeta",{"title":636,"_path":637},"Onoff Business (Beta)","\u002Fhe\u002Fdocs\u002Fintegrations\u002Fonoff",{"title":639,"_path":640},"Smart Life \u002F Tuya (מכשירים מחוברים)","\u002Fhe\u002Fdocs\u002Fintegrations\u002Fsmart-life",{"title":642,"_path":643},"מוצרי beTool (Desktop, Kiosk, Terminal)","\u002Fhe\u002Fdocs\u002Fintegrations\u002Fproduits-betool",{"title":645,"_path":646,"children":647},"תרחישי שימוש","\u002Fhe\u002Fdocs\u002Fcas-usage",[648,651,654,657,660,663],{"title":649,"_path":650},"סוכנים עסקיים לדוגמה","\u002Fhe\u002Fdocs\u002Fcas-usage\u002Fvue-densemble",{"title":652,"_path":653},"CFO IA","\u002Fhe\u002Fdocs\u002Fcas-usage\u002Fcfo-ia",{"title":655,"_path":656},"COO IA","\u002Fhe\u002Fdocs\u002Fcas-usage\u002Fcoo-ia",{"title":658,"_path":659},"CTO IA","\u002Fhe\u002Fdocs\u002Fcas-usage\u002Fcto-ia",{"title":661,"_path":662},"Product Manager IA","\u002Fhe\u002Fdocs\u002Fcas-usage\u002Fproduct-manager-ia",{"title":664,"_path":665},"עוזר אישי","\u002Fhe\u002Fdocs\u002Fcas-usage\u002Fassistant-personnel"]