@ggui-ai/ui-gen 0.28.0 → 0.29.0

package/dist/llm-evaluator-PN3WFCS4.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/evaluation/llm-evaluator.ts"],"names":[],"mappings":";;;;;;;;;;;;AA4CA,SAAS,kBAAkB,KAAA,EAA0B;AACnD,EAAA,IAAI,KAAA,CAAM,OAAA,CAAQ,KAAK,CAAA,EAAG;AACxB,IAAA,OAAO,KAAA,CAAM,OAAO,CAAC,CAAA,KAAmB,OAAO,CAAA,KAAM,QAAA,IAAY,CAAA,CAAE,MAAA,GAAS,CAAC,CAAA;AAAA,EAC/E;AACA,EAAA,IAAI,OAAO,UAAU,QAAA,EAAU;AAC7B,IAAA,IAAI;AACF,MAAA,MAAM,MAAA,GAAS,IAAA,CAAK,KAAA,CAAM,KAAK,CAAA;AAC/B,MAAA,IAAI,KAAA,CAAM,OAAA,CAAQ,MAAM,CAAA,EAAG;AACzB,QAAA,OAAO,MAAA,CAAO,OAAO,CAAC,CAAA,KAAmB,OAAO,CAAA,KAAM,QAAA,IAAY,CAAA,CAAE,MAAA,GAAS,CAAC,CAAA;AAAA,MAChF;AAAA,IACF,CAAA,CAAA,MAAQ;AAEN,MAAA,IAAI,KAAA,CAAM,MAAA,GAAS,CAAA,EAAG,OAAO,CAAC,KAAK,CAAA;AAAA,IACrC;AAAA,EACF;AACA,EAAA,OAAO,EAAC;AACV;AA2EA,IAAM,kBAAA,GAAqB,CAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,yGAAA,CAAA;AAQ3B,IAAM,yBAAA,GAA4B,CAAA;AAAA;AAAA;AAAA;AAAA;AAAA,yGAAA,CAAA;AAOlC,IAAM,4BAAA,GAA+B,CAAA;AAAA;AAAA;AAAA;AAAA,iDAAA,CAAA;AAMrC,IAAM,qBAAA,GAAwB,CAAA;AAAA;AAAA;AAAA;AAAA,4EAAA,CAAA;AAM9B,IAAM,gCAAA,GAAmC,CAAA;AAAA,uEAAA,CAAA;AAGzC,IAAM,yBAAA,GAA4B,CAAA;AAAA;AAAA,yCAAA,CAAA;AAI3B,SAAS,kBAAkB,GAAA,EAA0B;AAC1D,EAAA,MAAM,QAAkB,EAAC;AACzB,EAAA,MAAM,UAAA,GAAa,IAAI,UAAA,IAAc,mBAAA;AACrC,EAAA,MAAM,OAAO,UAAA,KAAe,MAAA;AAE5B,EAAA,KAAA,CAAM,IAAA;AAAA,IACJ,IAAA,GACI,CAAA;AAAA;AAAA,gEAAA,CAAA,GAGA,CAAA;AAAA;AAAA,gEAAA;AAAA,GAGN;AAEA,EAAA,KAAA,CAAM,IAAA,CAAK,CAAA;AAAA;AAAA;AAAA;;AAAA;AAAA;AAAA;AAAA;AAAA;;AAAA;;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,6MAAA,CAkB4L,CAAA;AAEvM,EAAA,KAAA,CAAM,IAAA,CAAK,CAAA;AAAA,EACX,GAAA,CAAI,cAAc,CAAA,CAAE,CAAA;AAEpB,EAAA,IAAI,IAAI,QAAA,EAAU;AAChB,IAAA,KAAA,CAAM,IAAA,CAAK,CAAA;;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;;AAAA;AAAA,EAuBb,IAAA,CAAK,UAAU,GAAA,CAAI,QAAQ,EAAE,KAAA,CAAM,CAAA,EAAG,GAAI,CAAC;AAAA,MAAA,CACtC,CAAA;AAAA,EACL;AAEA,EAAA,IAAI,IAAI,mBAAA,EAAqB;AAC3B,IAAA,KAAA,CAAM,IAAA,CAAK,CAAA;AAAA,EACb,GAAA,CAAI,mBAAmB,CAAA,CAAE,CAAA;AAAA,EACzB;AAEA,EAAA,KAAA,CAAM,IAAA,CAAK,CAAA,EAAG,IAAA,GAAO,kBAAA,GAAqB,yBAAyB;;AAAA;;AAAA;;AAAA;AAAA;AAAA;AAAA;;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;;AAAA;;AAAA;AAAA;AAAA;AAAA;AAAA;;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;;AAAA;AAAA;AAAA;AAAA;AAAA;;AAAA;AAAA;AAAA;AAAA;;AAAA,EA4CnE,IAAA,GAAO,wBAAwB,4BAA4B;;AAAA;;AAAA;AAAA;AAAA;AAAA;;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;;AAAA;AAAA,EAkC3D,IAAA,GAAO,4BAA4B,gCAAgC;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,6CAAA,CAavB,CAAA;AAE5C,EAAA,MAAM,YAAA,GAAe,6BAAA,CAA8B,GAAA,CAAI,OAAO,CAAA;AAC9D,EAAA,IAAI,YAAA,CAAa,MAAA,GAAS,CAAA,EAAG,KAAA,CAAM,KAAK,YAAY,CAAA;AACpD,EAAA,OAAO,KAAA,CAAM,KAAK,MAAM,CAAA;AAC1B;AAOA,SAAS,cAAc,SAAA,EAA+B;AACpD,EAAA,OAAO;AAAA,IACL,IAAA,EAAM,YAAY,SAAS,CAAA,CAAA;AAAA,IAC3B,WAAA,EAAa,gCAAgC,SAAS,CAAA,CAAA;AAAA,IACtD,UAAA,EAAY;AAAA,MACV,IAAA,EAAM,QAAA;AAAA,MACN,QAAA,EAAU,CAAC,MAAM,CAAA;AAAA,MACjB,UAAA,EAAY;AAAA,QACV,IAAA,EAAM,EAAE,IAAA,EAAM,SAAA,EAAU;AAAA,QACxB,MAAA,EAAQ;AAAA,UACN,IAAA,EAAM,OAAA;AAAA,UACN,KAAA,EAAO,EAAE,IAAA,EAAM,QAAA,EAAS;AAAA,UACxB,WAAA,EAAa;AAAA;AACf;AACF;AACF,GACF;AACF;AAGA,SAAS,cAAc,SAAA,EAA+B;AACpD,EAAA,OAAO;AAAA,IACL,IAAA,EAAM,YAAY,SAAS,CAAA,CAAA;AAAA,IAC3B,WAAA,EAAa,gCAAgC,SAAS,CAAA,CAAA;AAAA,IACtD,UAAA,EAAY;AAAA,MACV,IAAA,EAAM,QAAA;AAAA,MACN,QAAA,EAAU,CAAC,QAAQ,CAAA;AAAA,MACnB,UAAA,EAAY;AAAA,QACV,MAAA,EAAQ;AAAA,UACN,IAAA,EAAM,QAAA;AAAA,UACN,IAAA,EAAM,CAAC,MAAA,EAAQ,MAAA,EAAQ,MAAM;AAAA,SAC/B;AAAA,QACA,MAAA,EAAQ;AAAA,UACN,IAAA,EAAM,OAAA;AAAA,UACN,KAAA,EAAO,EAAE,IAAA,EAAM,QAAA,EAAS;AAAA,UACxB,WAAA,EAAa;AAAA;AACf;AACF;AACF,GACF;AACF;AAMA,SAAS,kBAAA,CACP,SAAA,EACA,cAAA,EACA,UAAA,EACQ;AACR,EAAA,MAAM,WAAA,GAAc;AAAA;AAAA;AAAA,EAAkD,cAAc;AAAA,MAAA,CAAA;AAGpF,EAAA,MAAM,YAAA,GAAe,gBAAA,CAAiB,SAAA,EAAW,UAAU,CAAA;AAC3D,EAAA,IAAI,YAAA,IAAgB,YAAA,CAAa,IAAA,GAAO,CAAA,EAAG;AACzC,IAAA,OAAO,CAAA,EAAG,aAAa,eAAe;AAAA,EACxC,WAAW;;AAAA,0CAAA,EAE+B,SAAS,CAAA,MAAA,CAAA;AAAA,EACnD;AAIA,EAAA,QAAQ,SAAA;AAAW,IACjB,KAAK,eAAA;AACH,MAAA,OAAO,CAAA;;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;;AAAA;;AAAA;;AAAA;AAAA,EAeX,WAAW;;AAAA;;AAAA;AAAA;AAAA;;AAAA,qEAAA,CAAA;AAAA,IAUT,KAAK,OAAA;AACH,MAAA,OAAO,CAAA;;AAAA;;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;;AAAA;AAAA,EAwBX,WAAW;;AAAA;;AAAA;AAAA;AAAA;;AAAA,gEAAA,CAAA;AAAA,IAUT,KAAK,eAAA;AACH,MAAA,OAAO,CAAA;;AAAA;;AAAA;;AAAA;AAAA,EAOX,WAAW;;AAAA;;AAAA;AAAA;AAAA;AAAA;;AAAA,mEAAA,CAAA;AAAA,IAWT,KAAK,eAAA;AACH,MAAA,OAAO,CAAA;;AAAA;;AAAA;AAAA,EAKX,WAAW;;AAAA;;AAAA;AAAA;AAAA;AAAA;;AAAA,0EAAA,CAAA;AAAA,IAWT,KAAK,QAAA;AACH,MAAA,OAAO,CAAA;;AAAA;;AAAA;AAAA,EAKX,WAAW;;AAAA;;AAAA;AAAA;AAAA;AAAA;;AAAA,0DAAA,CAAA;AAAA,IAWT,KAAK,SAAA;AACH,MAAA,OAAO,CAAA;;AAAA;;AAAA;AAAA;AAAA;;AAAA;AAAA,EASX,WAAW;;AAAA;;AAAA;AAAA;AAAA;AAAA,iIAAA,CAAA;AAAA,IAST,KAAK,QAAA;AACH,MAAA,OAAO,CAAA;;AAAA;;AAAA;AAAA,EAKX,WAAW;;AAAA;;AAAA;AAAA;AAAA;AAAA;;AAAA,sFAAA,CAAA;AAAA,IAWT;AACE,MAAA,OAAO,CAAA,SAAA,EAAY,SAAA,CAAU,WAAA,EAAa,IAAI,WAAW;;AAAA,0CAAA,EAAiD,SAAS,CAAA,MAAA,CAAA;AAAA;AAEzH;AAMA,SAAS,oBAAoB,QAAA,EAA0B;AACrD,EAAA,QAAQ,QAAA;AAAU,IAChB,KAAK,QAAA;AACH,MAAA,OAAO,kBAAA;AAAA,IACT,KAAK,QAAA;AACH,MAAA,OAAO,cAAA;AAAA,IACT,KAAK,QAAA;AACH,MAAA,OAAO,uBAAA;AAAA,IACT;AACE,MAAA,OAAO,kBAAA;AAAA;AAEb;AA4BA,SAAS,WAAW,QAAA,EAA0D;AAC5E,EAAA,OAAO;AAAA,IACL,IAAI,QAAA,CAAS,WAAA;AAAA,IACb,KAAK,QAAA,CAAS,YAAA;AAAA,IACd,GAAI,SAAS,eAAA,KAAoB,MAAA,GAAY,EAAE,SAAA,EAAW,QAAA,CAAS,eAAA,EAAgB,GAAI,EAAC;AAAA,IACxF,GAAI,SAAS,mBAAA,KAAwB,MAAA,GAAY,EAAE,aAAA,EAAe,QAAA,CAAS,mBAAA,EAAoB,GAAI;AAAC,GACtG;AACF;AAoBA,IAAM,eAAA,GAAkB,wBAAA,CAAyB,MAAA,CAAO,CAAC,CAAA,KAAM,CAAA,CAAE,IAAA,KAAS,CAAC,CAAA,CAAE,GAAA,CAAI,CAAC,CAAA,KAAM,EAAE,SAAS,CAAA;AACnG,IAAM,eAAA,GAAkB,wBAAA,CAAyB,MAAA,CAAO,CAAC,CAAA,KAAM,CAAA,CAAE,IAAA,KAAS,CAAC,CAAA,CAAE,GAAA,CAAI,CAAC,CAAA,KAAM,EAAE,SAAS,CAAA;AAEnG,eAAe,cACb,SAAA,EACA,IAAA,EACA,cACA,cAAA,EACA,KAAA,EACA,OACA,UAAA,EAC0B;AAC1B,EAAA,MAAM,KAAA,GAAQ,KAAK,GAAA,EAAI;AACvB,EAAA,MAAM,OAAO,IAAA,KAAS,CAAA,GAAI,cAAc,SAAS,CAAA,GAAI,cAAc,SAAS,CAAA;AAC5E,EAAA,MAAM,UAAA,GAAa,kBAAA,CAAmB,SAAA,EAAW,cAAA,EAAgB,UAAU,CAAA;AAE3E,EAAA,IAAI;AACF,IAAA,MAAM,QAAA,GAAW,MAAM,KAAA,CAAM,SAAA;AAAA,MAC3B,KAAA;AAAA,MACA,YAAA;AAAA,MACA,UAAA;AAAA,MACA,CAAC,IAAI,CAAA;AAAA,MACL;AAAA,KACF;AAEA,IAAA,MAAM,IAAA,GAAO,QAAA,CAAS,SAAA,CAAU,CAAC,CAAA;AACjC,IAAA,MAAM,UAAA,GAAa,IAAA,CAAK,GAAA,EAAI,GAAI,KAAA;AAEhC,IAAA,IAAI,CAAC,IAAA,EAAM;AACT,MAAA,OAAA,CAAQ,IAAA,CAAK,CAAA,OAAA,EAAU,SAAS,CAAA,uBAAA,CAAyB,CAAA;AACzD,MAAA,OAAO;AAAA,QACL,SAAA;AAAA,QACA,IAAA;AAAA,QACA,QAAQ,EAAC;AAAA,QACT,IAAA,EAAM,IAAA;AAAA,QACN,QAAA,EAAU,SAAA;AAAA,QACV,UAAA,EAAY,uBAAA;AAAA,QACZ,MAAA,EAAQ,WAAW,QAAQ,CAAA;AAAA,QAC3B;AAAA,OACF;AAAA,IACF;AAEA,IAAA,MAAM,MAAM,IAAA,CAAK,KAAA;AACjB,IAAA,MAAM,SAAsB,EAAC;AAC7B,IAAA,IAAI,IAAA;AAEJ,IAAA,IAAI,SAAS,CAAA,EAAG;AAEd,MAAA,IAAA,GAAO,IAAI,IAAA,KAAS,IAAA;AACpB,MAAA,MAAM,SAAA,GAAY,iBAAA,CAAkB,GAAA,CAAI,MAAM,CAAA;AAC9C,MAAA,KAAA,MAAW,QAAQ,SAAA,EAAW;AAC5B,QAAA,MAAA,CAAO,IAAA,CAAK;AAAA,UACV,IAAA,EAAM,CAAA;AAAA,UACN,MAAA,EAAQ,MAAA;AAAA,UACR,QAAA,EAAU,SAAA;AAAA,UACV,QAAA,EAAU,UAAA;AAAA,UACV,WAAA,EAAa,IAAA;AAAA,UACb,GAAA,EAAK;AAAA;AAAA,SACN,CAAA;AAAA,MACH;AAEA,MAAA,IAAI,CAAC,IAAA,IAAQ,MAAA,CAAO,MAAA,KAAW,CAAA,EAAG;AAChC,QAAA,MAAA,CAAO,IAAA,CAAK;AAAA,UACV,IAAA,EAAM,CAAA;AAAA,UACN,MAAA,EAAQ,MAAA;AAAA,UACR,QAAA,EAAU,SAAA;AAAA,UACV,QAAA,EAAU,UAAA;AAAA,UACV,WAAA,EAAa,GAAG,SAAS,CAAA,aAAA,CAAA;AAAA,UACzB,GAAA,EAAK,OAAO,SAAS,CAAA,OAAA;AAAA,SACtB,CAAA;AAAA,MACH;AAAA,IACF,CAAA,MAAO;AAEL,MAAA,MAAM,MAAA,GAAU,IAAI,MAAA,IAAqB,MAAA;AACzC,MAAA,IAAA,GAAO,MAAA,KAAW,MAAA;AAClB,MAAA,MAAM,SAAA,GAAY,iBAAA,CAAkB,GAAA,CAAI,MAAM,CAAA;AAC9C,MAAA,MAAM,OAAA,GAAU,MAAA,KAAW,MAAA,GAAS,MAAA,GAAS,MAAA;AAC7C,MAAA,KAAA,MAAW,QAAQ,SAAA,EAAW;AAC5B,QAAA,MAAA,CAAO,IAAA,CAAK;AAAA,UACV,IAAA,EAAM,CAAA;AAAA,UACN,MAAA,EAAQ,OAAA;AAAA,UACR,QAAA,EAAU,SAAA;AAAA,UACV,QAAA,EAAU,MAAA,KAAW,MAAA,GAAS,UAAA,GAAa,OAAA;AAAA,UAC3C,WAAA,EAAa,IAAA;AAAA,UACb,GAAA,EAAK;AAAA,SACN,CAAA;AAAA,MACH;AAAA,IACF;AAEA,IAAA,OAAO;AAAA,MACL,SAAA;AAAA,MACA,IAAA;AAAA,MACA,MAAA;AAAA,MACA,IAAA;AAAA,MACA,QAAA,EAAU,KAAA;AAAA,MACV,MAAA,EAAQ,WAAW,QAAQ,CAAA;AAAA,MAC3B;AAAA,KACF;AAAA,EACF,SAAS,CAAA,EAAG;AACV,IAAA,MAAM,UAAA,GAAa,IAAA,CAAK,GAAA,EAAI,GAAI,KAAA;AAChC,IAAA,MAAM,UAAU,CAAA,YAAa,KAAA,GAAQ,CAAA,CAAE,OAAA,GAAU,OAAO,CAAC,CAAA;AACzD,IAAA,OAAA,CAAQ,KAAA,CAAM,CAAA,OAAA,EAAU,SAAS,CAAA,eAAA,EAAa,OAAO,CAAA,CAAE,CAAA;AAKvD,IAAA,OAAO;AAAA,MACL,SAAA;AAAA,MACA,IAAA;AAAA,MACA,QAAQ,EAAC;AAAA,MACT,IAAA,EAAM,IAAA;AAAA,MACN,QAAA,EAAU,SAAA;AAAA,MACV,UAAA,EAAY,OAAA;AAAA,MACZ,MAAA,EAAQ,EAAE,EAAA,EAAI,CAAA,EAAG,KAAK,CAAA,EAAE;AAAA,MACxB;AAAA,KACF;AAAA,EACF;AACF;AAYA,eAAsB,WAAA,CACpB,SACA,OAAA,EAC+B;AAC/B,EAAA,MAAM,eAAe,iBAAA,CAAkB;AAAA,IACrC,gBAAgB,OAAA,CAAQ,cAAA;AAAA,IACxB,UAAU,OAAA,CAAQ,QAAA;AAAA,IAClB,qBAAqB,OAAA,CAAQ,aAAA;AAAA,IAC7B,YAAY,OAAA,CAAQ,UAAA;AAAA,IACpB,SAAS,OAAA,CAAQ;AAAA,GAClB,CAAA;AAED,EAAA,OAAO,EAAE,YAAA,EAAc,eAAA,EAAiB,EAAC,EAAE;AAC7C;AAOA,IAAM,oCAAoB,IAAI,GAAA,CAAY,CAAC,eAAA,EAAiB,OAAO,CAAC,CAAA;AAGpE,IAAM,iBAAA,uBAAwB,GAAA,CAAY;AAAA,EACxC,eAAA;AAAA,EACA,eAAA;AAAA,EACA,QAAA;AAAA,EACA,SAAA;AAAA,EACA;AACF,CAAC,CAAA;AAGD,IAAM,gBAAA,uBAAuB,GAAA,CAAY;AAAA,EACvC,GAAG,iBAAA;AAAA,EACH,GAAG;AACL,CAAC,CAAA;AAQM,SAAS,kBAAkB,GAAA,EAA6B;AAC7D,EAAA,MAAM,SAAA,GAAa,GAAA,CAAI,MAAA,IAAU,EAAC;AAClC,EAAA,MAAM,OAAA,GAAW,GAAA,CAAI,IAAA,IAAQ,EAAC;AAE9B,EAAA,MAAM,MAAA,GAAsB,SAAA,CACzB,MAAA,CAAO,CAAC,CAAA,KAAM;AACb,IAAA,MAAM,MAAM,CAAA,CAAE,QAAA;AACd,IAAA,OAAO,gBAAA,CAAiB,IAAI,GAAG,CAAA;AAAA,EACjC,CAAC,CAAA,CACA,GAAA,CAAI,CAAC,CAAA,KAAM;AACV,IAAA,MAAM,WAAW,CAAA,CAAE,QAAA;AACnB,IAAA,MAAM,OAAA,GAAU,iBAAA,CAAkB,GAAA,CAAI,QAAQ,CAAA;AAG9C,IAAA,MAAM,IAAA,GAAO,UAAU,CAAA,GAAI,CAAA;AAI3B,IAAA,MAAM,SAAS,OAAA,GACX,MAAA,GACE,CAAA,CAAE,MAAA,KAAsB,SAAS,MAAA,GAAS,MAAA;AAEhD,IAAA,OAAO;AAAA,MACL,IAAA;AAAA,MACA,MAAA;AAAA,MACA,QAAA;AAAA,MACA,QAAA,EAAW,CAAA,CAAE,QAAA,KAAsC,OAAA,GAAU,UAAA,GAAa,OAAA,CAAA;AAAA,MAC1E,WAAA,EAAc,EAAE,WAAA,IAA0B,EAAA;AAAA,MAC1C,GAAA,EAAM,EAAE,GAAA,IAAkB;AAAA,KAC5B;AAAA,EACF,CAAC,CAAA;AAEH,EAAA,OAAO;AAAA,IACL,MAAA;AAAA,IACA,IAAA,EAAM;AAAA,GACR;AACF;AAgBA,eAAsB,gBAAA,CACpB,OAAA,EACA,MAAA,EACA,gBAAA,EACyC;AACzC,EAAA,MAAM,eAAe,MAAA,CAAO,QAAA;AAC5B,EAAA,MAAM,KAAA,GAAQ,MAAA,CAAO,KAAA,IAAS,mBAAA,CAAoB,YAAY,CAAA;AAC9D,EAAA,MAAM,cAAA,GACJ,YAAA,KAAiB,QAAA,GACZ,WAAA,GACA,YAAA;AAGP,EAAA,MAAM,UAAA,GAAa,QAAQ,UAAA,IAAc,mBAAA;AACzC,EAAA,MAAM,YAAA,GAAe,gBAAA,EAAkB,YAAA,IAAgB,iBAAA,CAAkB;AAAA,IACvE,gBAAgB,OAAA,CAAQ,cAAA;AAAA,IACxB,UAAU,OAAA,CAAQ,QAAA;AAAA,IAClB,qBAAqB,OAAA,CAAQ,aAAA;AAAA,IAC7B,UAAA;AAAA,IACA,SAAS,OAAA,CAAQ;AAAA,GAClB,CAAA;AAGD,EAAA,MAAM,cAAA,GAAiB,QAAQ,UAAA,CAC5B,KAAA,CAAM,IAAI,CAAA,CACV,GAAA,CAAI,CAAC,IAAA,EAAM,CAAA,KAAM,CAAA,EAAG,OAAO,CAAA,GAAI,CAAC,CAAA,CAAE,QAAA,CAAS,CAAC,CAAC,MAAM,IAAI,CAAA,CAAE,CAAA,CACzD,IAAA,CAAK,IAAI,CAAA;AAMZ,EAAA,MAAM,QAAQ,WAAA,CAAY;AAAA,IACxB,QAAA,EAAU,cAAA;AAAA,IACV,KAAA;AAAA,IACA,eAAe,MAAA,CAAO,aAAA;AAAA,IACtB,SAAS,MAAA,CAAO;AAAA,GACjB,CAAA;AAMD,EAAA,MAAM,SAAA,GAAY,KAAK,GAAA,EAAI;AAE3B,EAAA,MAAM,kBAAA,GAAqB,eAAA,CAAgB,MAAA,GAAS,eAAA,CAAgB,MAAA;AAGpE,EAAA,MAAM,OAAA,GAAU,MAAM,OAAA,CAAQ,GAAA,CAAI;AAAA,IAChC,GAAG,eAAA,CAAgB,GAAA,CAAI,CAAA,CAAA,KAAK,aAAA,CAAc,CAAA,EAAG,CAAA,EAAG,YAAA,EAAc,cAAA,EAAgB,KAAA,EAAO,KAAA,EAAO,UAAU,CAAC,CAAA;AAAA,IACvG,GAAG,eAAA,CAAgB,GAAA,CAAI,CAAA,CAAA,KAAK,aAAA,CAAc,CAAA,EAAG,CAAA,EAAG,YAAA,EAAc,cAAA,EAAgB,KAAA,EAAO,KAAA,EAAO,UAAU,CAAC;AAAA,GACxG,CAAA;AAGD,EAAA,MAAM,YAAyB,EAAC;AAChC,EAAA,MAAM,iBAA2B,EAAC;AAClC,EAAA,IAAI,OAAA,GAAU,CAAA;AACd,EAAA,IAAI,QAAA,GAAW,CAAA;AAGf,EAAA,IAAI,SAAA;AACJ,EAAA,IAAI,aAAA;AAEJ,EAAA,KAAA,MAAW,KAAK,OAAA,EAAS;AACvB,IAAA,SAAA,CAAU,IAAA,CAAK,GAAG,CAAA,CAAE,MAAM,CAAA;AAC1B,IAAA,IAAI,EAAE,IAAA,EAAM;AACV,MAAA,cAAA,CAAe,IAAA,CAAK,EAAE,SAAS,CAAA;AAAA,IACjC;AACA,IAAA,OAAA,IAAW,EAAE,MAAA,CAAO,EAAA;AACpB,IAAA,QAAA,IAAY,EAAE,MAAA,CAAO,GAAA;AACrB,IAAA,IAAI,CAAA,CAAE,OAAO,SAAA,KAAc,MAAA,eAAwB,SAAA,IAAa,CAAA,IAAK,EAAE,MAAA,CAAO,SAAA;AAC9E,IAAA,IAAI,CAAA,CAAE,OAAO,aAAA,KAAkB,MAAA,mBAA4B,aAAA,IAAiB,CAAA,IAAK,EAAE,MAAA,CAAO,aAAA;AAAA,EAC5F;AAEA,EAAA,MAAM,MAAA,GAAS,IAAA,CAAK,GAAA,EAAI,GAAI,SAAA;AAK5B,EAAA,MAAM,gBAAA,GAAwC,OAAA,CAAQ,GAAA,CAAI,CAAC,CAAA,MAAO;AAAA,IAChE,WAAW,CAAA,CAAE,SAAA;AAAA,IACb,MAAM,CAAA,CAAE,IAAA;AAAA,IACR,QAAQ,CAAA,CAAE,QAAA;AAAA,IACV,GAAI,EAAE,UAAA,KAAe,MAAA,GAAY,EAAE,MAAA,EAAQ,CAAA,CAAE,UAAA,EAAW,GAAI;AAAC,GAC/D,CAAE,CAAA;AACF,EAAA,MAAM,YAAA,GAAe,iBAAiB,MAAA,CAAO,CAAC,MAAM,CAAA,CAAE,MAAA,KAAW,SAAS,CAAA,CAAE,MAAA;AAG5E,EAAA,KAAA,MAAW,KAAK,OAAA,EAAS;AACvB,IAAA,MAAM,UAAA,GAAa,EAAE,MAAA,CAAO,MAAA;AAC5B,IAAA,MAAM,MAAA,GAAS,EAAE,QAAA,KAAa,SAAA,GAC1B,YACA,CAAA,CAAE,IAAA,GAAO,SAAS,CAAA,CAAE,IAAA,KAAS,IAAI,MAAA,GAAU,CAAA,CAAE,OAAO,IAAA,CAAK,CAAA,CAAA,KAAK,EAAE,MAAA,KAAW,MAAM,IAAI,MAAA,GAAS,MAAA;AAClG,IAAA,MAAM,UAAA,GAAa,UAAA,GAAa,CAAA,GAAI,CAAA,EAAA,EAAK,UAAU,SAAS,UAAA,GAAa,CAAA,GAAI,GAAA,GAAM,EAAE,CAAA,CAAA,CAAA,GAAM,EAAA;AAC3F,IAAA,OAAA,CAAQ,GAAA,CAAI,CAAA,OAAA,EAAU,CAAA,CAAE,SAAS,KAAK,MAAM,CAAA,EAAG,UAAU,CAAA,CAAA,EAAA,CAAK,EAAE,UAAA,GAAa,GAAA,EAAM,OAAA,CAAQ,CAAC,CAAC,CAAA,CAAA,CAAG,CAAA;AAAA,EAClG;AACA,EAAA,OAAA,CAAQ,GAAA;AAAA,IACN,CAAA,cAAA,EAAA,CAAkB,MAAA,GAAS,GAAA,EAAM,OAAA,CAAQ,CAAC,CAAC,CAAA,QAAA,EAAW,kBAAkB,CAAA,gBAAA,EAClE,OAAO,CAAA,KAAA,EAAQ,QAAQ,CAAA,WAAA,EAAc,SAAA,IAAa,YAAY,CAAA,YAAA,EAAe,aAAA,IAAiB,YAAY,CAAA,UAAA,EACtG,SAAA,CAAU,MAAM,CAAA,MAAA,EAAS,cAAA,CAAe,MAAM,CAAA,CAAA,IACvD,YAAA,GAAe,CAAA,GAAI,CAAA,SAAA,EAAY,YAAY,CAAA,CAAA,GAAK,EAAA;AAAA,GACnD;AAEA,EAAA,OAAO;AAAA,IACL,MAAA,EAAQ,SAAA;AAAA,IACR,IAAA,EAAM,cAAA;AAAA,IACN,gBAAA;AAAA,IACA,WAAA,EAAa,OAAA;AAAA,IACb,YAAA,EAAc,QAAA;AAAA,IACd,GAAI,SAAA,KAAc,MAAA,GAAY,EAAE,eAAA,EAAiB,SAAA,KAAc,EAAC;AAAA,IAChE,GAAI,aAAA,KAAkB,MAAA,GAAY,EAAE,mBAAA,EAAqB,aAAA,KAAkB;AAAC,GAC9E;AACF","file":"llm-evaluator-PN3WFCS4.js","sourcesContent":["// packages/ui-gen/src/evaluation/llm-evaluator.ts\n//\n// Parallel per-criterion LLM evaluator with cached mother prompt.\n//\n// Architecture:\n//   1. Mother prompt (cached, sent once): user request + contract + design system + all criteria definitions\n//   2. 7 parallel per-criterion LLM calls, each reusing the cached mother prompt\n//   3. Each call returns a focused tool response for one criterion\n//\n// Tiers:\n//   Tier 1 — Critical (always blocks): functionality, crash → { pass: boolean, issues?: string[] }\n//   Tier 2 — Quality (LLM decides):    interactivity, accessibility, layout, loading, visual → { result: 'fail'|'warn'|'pass', issues?: string[] }\n//\n// Caching: Anthropic ephemeral cache, OpenAI auto-cache, Google explicit cache.\n// Same system prompt across all 7 calls → first call caches, 6 remaining hit cache.\n\nimport { buildStylingProfileJudgeBlock } from '../boilerplate/styling-profile.js';\nimport type { GenerationProfileInput } from '../boilerplate/styling-profile.js';\nimport type {\n  EvalIssue,\n  EvalResult,\n  EvalCategory,\n  CriterionCoverage,\n  CriterionRunStatus,\n} from './types-public.js';\nimport { LLM_EVAL_STATIC_CRITERIA } from './types-public.js';\nimport { createAgent } from '../harness/llm-router';\nimport type { AgentConfig, LLMToolDef, LLMAgent, LLMToolCallResponse } from '../harness/llm-router';\nimport type { JsonObject } from '@ggui-ai/protocol';\nimport { getCriterionById } from './types-public.js';\nimport { DEFAULT_DESIGN_MODE, type DesignMode } from '../design-mode.js';\n\n// =============================================================================\n// Helpers\n// =============================================================================\n\n/**\n * Safely coerce LLM tool output to string[].\n *\n * LLMs sometimes return `issues` as a JSON string instead of an array.\n * Iterating a string goes character-by-character, creating thousands of\n * single-char \"issues\". This helper handles: string[] (pass-through),\n * string (JSON parse), or anything else (empty array).\n */\nfunction coerceStringArray(value: unknown): string[] {\n  if (Array.isArray(value)) {\n    return value.filter((v): v is string => typeof v === 'string' && v.length > 1);\n  }\n  if (typeof value === 'string') {\n    try {\n      const parsed = JSON.parse(value);\n      if (Array.isArray(parsed)) {\n        return parsed.filter((v): v is string => typeof v === 'string' && v.length > 1);\n      }\n    } catch {\n      // Not JSON — treat the whole string as a single issue\n      if (value.length > 1) return [value];\n    }\n  }\n  return [];\n}\n\n// =============================================================================\n// Public interfaces\n// =============================================================================\n\nexport interface LLMEvalContext {\n  sourceCode: string;\n  originalPrompt: string;\n  designContext?: string;\n  contract?: unknown;\n  /**\n   * Which triad produced the source. Selects the criteria text\n   * (`criteriaFor`) and the mother prompt's styling rules: in `free`\n   * mode the `tokens` criterion narrows to color and the `visual`\n   * criterion + \"Design System Rules\" block become arm-neutral. Absent\n   * means `constrained` (byte-identical to the pre-`designMode` prompt).\n   */\n  designMode?: DesignMode;\n  /** The app's generation profile (#991) — the judge reads the same text the coder was given. */\n  profile?: GenerationProfileInput;\n}\n\nexport interface LLMEvalConfig {\n  provider: 'claude' | 'openai' | 'google' | 'openrouter';\n  model?: string;\n  /**\n   * Explicit per-call credentials/routing (#484) — see\n   * `AgentConfig.routeOverride`. Threaded onto the agent this\n   * evaluator constructs so it never falls back to reading\n   * `process.env`. Absent (default) preserves existing behavior.\n   */\n  routeOverride?: AgentConfig['routeOverride'];\n  /**\n   * Provider-429-retry observer (#489) — see `AgentConfig.onRetry`.\n   * Threaded onto the agent this evaluator constructs so a rate-limited\n   * retry inside an evaluation turn's `apiCall()` reaches the same\n   * caller the coding leg reaches. Absent (default) preserves existing\n   * behavior (retries still happen; no observer fires).\n   */\n  onRetry?: AgentConfig['onRetry'];\n}\n\n/** A request-specific evaluation criterion generated by the LLM. */\nexport interface DynamicCriterion {\n  name: string;\n  description: string;\n}\n\n/** Pre-warmed evaluation context (generated during coding, consumed at eval time). */\nexport interface PreWarmedEvalContext {\n  motherPrompt: string;\n  dynamicCriteria: DynamicCriterion[];\n}\n\n// =============================================================================\n// Mother prompt builder (cacheable context)\n// =============================================================================\n\nexport interface EvalContext {\n  originalPrompt: string;\n  contract?: unknown;\n  designSystemSummary?: string;\n  /** See `LLMEvalContext.designMode`. */\n  designMode?: DesignMode;\n  /** See `LLMEvalContext.profile`. */\n  profile?: GenerationProfileInput;\n}\n\n/**\n * Arm-neutral styling rules for `designMode: 'free'` — replaces the\n * constrained \"Design System Rules\" block. Mirrors the free prompt: color\n * on the closed token manifest, everything else may be literal, raw HTML\n * + CSS is a first-class choice.\n */\nconst FREE_STYLING_RULES = `## Styling Rules (important for evaluation)\n- Color: every color must be a bare \\`var(--ggui-color-*)\\` token — a hardcoded #hex, rgb()/hsl() or CSS named color is a fail, and so is a literal fallback inside \\`var(--ggui-*, …)\\`.\n- Accent text (eyebrow, tagline, link, label) painted with a bare \\`var(--ggui-color-primary-300…600)\\` is a readability fail — the accent ink is \\`var(--ggui-color-link)\\`; a tint as text on a primary fill and a deep stop (700–900) on a light tint are fine\n- Spacing, typography, radius, shadow geometry and layout values MAY be literals (\\`padding: '20px'\\`, \\`fontSize: 18\\`, \\`borderRadius: 12\\`) — NEVER flag them.\n- Raw HTML elements with inline styles or \\`<style>\\` blocks are a first-class choice — never flag \"not using the design system\" or reward using it.\n- When \\`@ggui-ai/design\\` primitives ARE used, their enum/scale props (\\`variant=\"primary\"\\`, \\`size=\"lg\"\\`, \\`gap=\"md\"\\`, \\`tone=\"muted\"\\`, \\`surface=\"accent\"\\`) are valid.\n- One \\`export default function Component\\` — helper components as named functions above it are fine`;\n\nconst CONSTRAINED_STYLING_RULES = `## Design System Rules (important for evaluation)\n- Spacing props (\\`gap\\`, \\`padding\\`, \\`margin\\`) take a t-shirt-scale name: \\`gap=\"md\"\\`, \\`padding=\"lg\"\\` (\\`none|xs|sm|md|lg|xl|2xl\\`). **These ARE design tokens** — each resolves to a \\`--ggui-spacing-*\\` variable. A raw \\`var(--ggui-spacing-*)\\` string is an accepted escape hatch. NEVER flag a scale name as \"hardcoded\". Only a numeric prop (\\`padding={24}\\`) or a raw CSS length (\\`gap=\"13px\"\\`) bypasses the scale and should be warned.\n- All colors must use CSS variables: \\`color=\"var(--ggui-color-onContainer)\"\\` — hardcoded #hex or rgb() is a fail\n- Accent text (eyebrow, tagline, link, label) painted with a bare \\`primary-300…600\\` stop is a readability fail — the accent ink is \\`var(--ggui-color-link)\\` (or \\`tone=\"emphasized\"\\`); tints as text on a primary fill and deep stops on a light tint are fine\n- Component props that take enum/scale values — \\`variant=\"primary\"\\`, \\`size=\"lg\"\\`, \\`gap=\"md\"\\`, \\`padding=\"lg\"\\`, \\`shadow=\"md\"\\`, \\`radius=\"lg\"\\`, \\`tone=\"muted\"\\`, \\`surface=\"accent\"\\` — are all design system tokens and are always valid\n- One \\`export default function Component\\` — helper components as named functions above it are fine`;\n\nconst CONSTRAINED_VISUAL_REFERENCE = `**visual**: Design system consistency.\n- Uses CSS variables (var(--ggui-*)) for colors, not hardcoded values\n- Consistent use of spacing tokens\n- Proper use of component variants (primary, outline, ghost)\n- Shadow, border-radius from design system tokens`;\n\nconst FREE_VISUAL_REFERENCE = `**visual**: Visual composition.\n- Clear hierarchy (one focal point, headings distinct from body), consistent spacing rhythm\n- Coherent palette drawn from the theme color tokens\n- Fit to the canvas: fluid width, nothing cramped or stretched\n- No reward or penalty for using (or not using) the design-system primitives`;\n\nconst CONSTRAINED_RAW_CLICK_TARGET_GAP = `- A raw \\`<div>\\` / \\`<span>\\` used as a click target without \\`as={Clickable}\\`\n  (no role, no keyboard) — or any hand-rolled interactive element.`;\n\nconst FREE_RAW_CLICK_TARGET_GAP = `- A raw \\`<div>\\` / \\`<span>\\` used as a click target with no \\`role=\"button\"\\`,\n  \\`tabIndex\\` and keyboard handler — a real \\`<button>\\` / \\`<a href>\\` is the\n  correct raw element and is never a gap.`;\n\nexport function buildMotherPrompt(ctx: EvalContext): string {\n  const parts: string[] = [];\n  const designMode = ctx.designMode ?? DEFAULT_DESIGN_MODE;\n  const free = designMode === 'free';\n\n  parts.push(\n    free\n      ? `You are evaluating a generated React component. It was composed freely — raw HTML elements, inline styles and <style> blocks are allowed and the ggui design package is optional — against a data contract and a theme-token color rule.\nThe code has already passed automated checks (compilation, security, imports, color tokens).\nYour job is to evaluate ONE specific quality criterion per call.`\n      : `You are evaluating a generated React component built with the ggui design system.\nThe code has already passed automated checks (compilation, security, imports, design tokens).\nYour job is to evaluate ONE specific quality criterion per call.`,\n  );\n\n  parts.push(`## About This Prompt\nThis system prompt has 2 sections:\n1. **Context** — the original user request and data contract\n2. **Criteria Reference** — definitions for all 7 evaluation criteria (your call will focus on one)\n\n## Input Messages\nYou will receive ONE user message per call containing:\n1. Which criterion to evaluate (e.g., \"Evaluate FUNCTIONALITY\")\n2. The source code with line numbers\n3. Example outputs showing the expected tool response format\n\nRespond with a single tool call (evaluate_{criterion}). Report ALL issues you find for that criterion, not just the first one. Each issue should be an actionable fix instruction.\n\n## Issue-array discipline (applies to EVERY criterion)\nThe \"issues\" array is for confirmed defects ONLY. Before adding an entry, it must be something you are CERTAIN is missing or broken. NEVER add:\n- a self-negating note (\"…actually this is fine\", \"so no issue here\", \"no actual crash scenarios are present\");\n- a speculative / conditional note (\"if the data is shaped X then…\", \"this may break when…\");\n- a \"should be verified\" / \"double-check that…\" note.\nIf you are not certain something is broken, it is a PASS — omit it entirely. An entry you describe as acceptable still counts against the score, so describing a non-issue is worse than saying nothing.`);\n\n  parts.push(`## Original Request\n${ctx.originalPrompt}`);\n\n  if (ctx.contract) {\n    parts.push(`## Data Contract (JSON Schema → TypeScript types)\n\nThe component's boilerplate was generated from this contract. All types are compiler-enforced:\n- \\`props\\` → \\`interface Props\\` — every field must be rendered in the UI\n- \\`actionSpec\\` (flat map keyed by action name) → \\`useAction<Payload>(name)\\` hooks — each must be wired to a UI element (button, form, etc.)\n- \\`actionSpec[name].nextStep\\` (optional) — names an agent-side MCP tool the agent SHOULD invoke on its\n  next turn after the action fires. Surfaces as advisory event metadata; the component does NOT call\n  the named tool. Use it to judge whether button labels/icons match the underlying action's intent.\n- \\`streamSpec\\` (flat map keyed by channel name) → \\`useStream<T>(name)\\` hooks — each must be consumed with null guard (\\`.latest && ...\\`)\n- \\`agentCapabilities.tools\\` is a READ-ONLY catalog — the AGENT invokes these. The component never calls them.\n  References surface only via \\`actionSpec[*].nextStep\\` and \\`streamSpec[*].source.tool\\`. Do NOT flag\n  missing component-side calls for these.\n- \\`clientCapabilities.gadgets\\` → a map keyed by npm PACKAGE name; each package maps export\n  names to a descriptor. Two kinds, both contract features that MUST be used:\n  • Built-in browser capabilities (\\`useGeolocation\\`, \\`useCamera\\`, \\`useClipboardWrite\\`,\n    \\`useMicrophone\\`, \\`useFilePicker\\`, \\`useClipboardPaste\\`, \\`useNotifications\\`) — imported\n    from \\`@ggui-ai/gadgets\\`.\n  • Registered third-party gadgets — hooks or components — imported from their OWN package\n    (e.g. \\`import { useChartTheme } from '@ggui-samples/gadget-chart'\\`). The boilerplate\n    emits that import. Any export listed under \\`clientCapabilities.gadgets\\` IS part of the\n    contract — NEVER flag it as \"not part of the contract\".\n\n\\`\\`\\`json\n${JSON.stringify(ctx.contract).slice(0, 6000)}\n\\`\\`\\``);\n  }\n\n  if (ctx.designSystemSummary) {\n    parts.push(`## Design System\n${ctx.designSystemSummary}`);\n  }\n\n  parts.push(`${free ? FREE_STYLING_RULES : CONSTRAINED_STYLING_RULES}\n\n## Evaluation Criteria Reference\n\n### Tier 1 — Critical (blocks delivery)\n\n**functionality**: Does the component implement ALL features in the original request?\n- Check every noun/verb in the request against the rendered output\n- Missing feature = fail, even if other features work perfectly\n- Partial implementation = fail (e.g., search UI exists but doesn't filter)\n\n**crash**: Are there runtime crash scenarios?\n- Uninitialized state used in render (e.g., items.map when items could be undefined)\n- Missing event handlers referenced in JSX\n- Destructuring undefined objects\n- Array access without bounds checking\n- Missing null/undefined guards before .map(), .filter(), .length\n\n### Tier 2 — Quality (fail if severe, warn if improvement)\n\n**interactivity**: Does it have interactive elements appropriate for its purpose?\n- Forms should have submit buttons\n- Lists should be navigable\n- Editable fields should have save/cancel\n- Interactive components should have hover/focus feedback\n\n**accessibility**: Semantic HTML, labels, alt text, keyboard navigation, ARIA state.\n- Form inputs must have associated labels (label prop or aria-label)\n- Images need alt text\n- Interactive elements need keyboard support\n- Interactive controls expose their state (aria-checked / aria-selected / aria-expanded / aria-pressed), never styling alone\n- Proper heading hierarchy (h1 > h2 > h3)\n\n**layout**: Spacing, alignment, visual grouping.\n- Consistent spacing between sections\n- Related items grouped visually\n- No overlapping or cramped elements\n- Proper use of containers and cards for grouping\n\n**loading**: Loading/error/empty states for dynamic data.\n- Components receiving async data should show loading state\n- Error boundaries or error messages for failure cases\n- Empty states (\"No items found\") when data arrays could be empty\n\n${free ? FREE_VISUAL_REFERENCE : CONSTRAINED_VISUAL_REFERENCE}\n\n## Primitive Accessibility (built-in — do NOT flag as missing)\n\nggui design-system primitives ship their own ARIA. The code you are\nreading imports them from \\`@ggui-ai/design\\`; the source does NOT show\ntheir internal markup. NEVER report a primitive below as missing a\nrole, aria-*, label, or keyboard support — it is already there:\n\n- \\`Input\\` / \\`Select\\` / \\`TextArea\\` — render a real \\`<label htmlFor>\\` from\n  the \\`label\\` prop, plus \\`aria-invalid\\` + \\`aria-describedby\\` + an error\n  \\`role=\"alert\"\\`. A primitive that has a \\`label\\` prop IS labelled.\n- \\`RadioGroup\\` — \\`role=\"radiogroup\"\\` + \\`aria-labelledby\\` from its \\`label\\`.\n- \\`Checkbox\\` — wrapped in its own \\`<label>\\`.\n- \\`Toggle\\` — \\`role=\"switch\"\\` + \\`aria-checked\\` + \\`aria-label\\`.\n- \\`Progress\\` — \\`role=\"progressbar\"\\` + \\`aria-valuenow/min/max\\` + \\`aria-label\\`.\n- \\`Slider\\` — full \\`aria-valuenow/min/max\\` slider semantics.\n- \\`Spinner\\` — \\`role=\"status\"\\` + \\`aria-label=\"Loading\"\\`.\n- \\`Skeleton\\` — \\`aria-hidden\\` (correctly decorative).\n- \\`Tabs\\` — full \\`role=\"tablist\"/\"tab\"/\"tabpanel\"\\` + \\`aria-selected/controls\\`.\n- \\`Stepper\\` — \\`<nav aria-label=\"Progress\">\\` + \\`aria-current=\"step\"\\` on the\n  active step; steps become real buttons when \\`onStepClick\\` is passed. It is\n  display-only by design — the component owning the step index in its own\n  \\`useState\\` is CORRECT, never a gap.\n- \\`Accordion\\` — \\`aria-expanded\\` + \\`aria-controls\\` + \\`role=\"region\"\\`.\n- \\`Alert\\` — \\`role=\"alert\"\\`; \\`Toast\\` — \\`role=\"alert\"\\` + \\`aria-live\\`.\n- \\`Tooltip\\` — \\`role=\"tooltip\"\\`.\n- \\`Clickable\\` (trait, applied via \\`as={Clickable}\\`) — \\`role=\"button\"\\` +\n  \\`tabIndex={0}\\` + Enter/Space keyboard activation.\n- \\`Icon\\` — decorative by default (\\`aria-hidden\\`); exposes \\`role=\"img\"\\` +\n  the label ONLY when an \\`aria-label\\` prop is passed. A bare \\`<Icon>\\` next\n  to text is CORRECTLY silent — never ask for \\`aria-hidden\\`, it is there.\n\nREAL accessibility gaps worth flagging:\n${free ? FREE_RAW_CLICK_TARGET_GAP : CONSTRAINED_RAW_CLICK_TARGET_GAP}\n- A hand-rolled control whose state (checked / selected / expanded /\n  pressed) is conveyed ONLY by styling — e.g. a styled-div \"checkbox\"\n  with a check icon + strikethrough label but no \\`role=\"checkbox\"\\` /\n  \\`aria-checked\\`. Note \\`as={Clickable}\\` adds button + keyboard\n  semantics but NOT state — \\`aria-checked\\` / \\`aria-pressed\\` /\n  \\`aria-selected\\` / \\`aria-expanded\\` must be passed explicitly.\n- An \\`<img>\\` / \\`<Image>\\` with no alt text.\n- \\`<Input>\\` / \\`<Select>\\` / \\`<TextArea>\\` used with NO \\`label\\` prop.\n- An icon-only \\`<Button>\\` (no text child) with no \\`aria-label\\`.\n- Live or streaming data (stream \\`.latest\\`, counters, real-time clocks)\n  rendered in a plain element with no \\`aria-live\\` / \\`role=\"status\"\\`.\n- Inverted or skipped heading levels (an \\`<Heading level={3}>\\` rendered\n  before the page's \\`<Heading level={1}>\\`).`);\n\n  const profileBlock = buildStylingProfileJudgeBlock(ctx.profile);\n  if (profileBlock.length > 0) parts.push(profileBlock);\n  return parts.join('\\n\\n');\n}\n\n// =============================================================================\n// Per-criterion tool definitions\n// =============================================================================\n\n/** Tier 1 tool schema: binary pass/fail with issues list */\nfunction makeTier1Tool(criterion: string): LLMToolDef {\n  return {\n    name: `evaluate_${criterion}`,\n    description: `Submit evaluation result for ${criterion}`,\n    parameters: {\n      type: 'object',\n      required: ['pass'],\n      properties: {\n        pass: { type: 'boolean' },\n        issues: {\n          type: 'array',\n          items: { type: 'string' },\n          description: 'List of specific issues found',\n        },\n      },\n    },\n  };\n}\n\n/** Tier 2 tool schema: fail/warn/pass with issues list */\nfunction makeTier2Tool(criterion: string): LLMToolDef {\n  return {\n    name: `evaluate_${criterion}`,\n    description: `Submit evaluation result for ${criterion}`,\n    parameters: {\n      type: 'object',\n      required: ['result'],\n      properties: {\n        result: {\n          type: 'string',\n          enum: ['fail', 'warn', 'pass'],\n        },\n        issues: {\n          type: 'array',\n          items: { type: 'string' },\n          description: 'List of specific issues found',\n        },\n      },\n    },\n  };\n}\n\n// =============================================================================\n// Per-criterion user prompts\n// =============================================================================\n\nfunction getCriterionPrompt(\n  criterion: string,\n  numberedSource: string,\n  designMode: DesignMode,\n): string {\n  const sourceBlock = `\\nSource code (with line numbers):\\n\\`\\`\\`tsx\\n${numberedSource}\\n\\`\\`\\``;\n\n  // Use criteria.ts as single source of truth when available\n  const criterionDef = getCriterionById(criterion, designMode);\n  if (criterionDef && criterionDef.tier > 0) {\n    return `${criterionDef.evalInstruction}\n${sourceBlock}\n\nReturn your evaluation using the evaluate_${criterion} tool.`;\n  }\n\n  // Fallback for criteria not yet in criteria.ts\n\n  switch (criterion) {\n    case 'functionality':\n      return `Evaluate FUNCTIONALITY: Does this component implement ALL features from the request AND the data contract?\n\nCheck against BOTH sources:\n1. **Original request** — each feature must be coded AND rendered in JSX\n2. **Data contract** (if present in the context above) — verify:\n   - ALL Props fields are rendered somewhere in the UI (not just accepted as params)\n   - ALL useAction hooks are wired to clickable UI elements (buttons, forms, cards)\n   - ALL useStream hooks have their \\`.latest\\` data rendered (with null guard)\n   - ALL clientCapabilities hooks (useGeolocation / useCamera / etc.) are called in the component body and their values either render directly or thread into a context / action payload\n\nA contract hook that is declared but never used in JSX is a MISSING feature.\n\n**CRITICAL RULE:** The \"issues\" array must ONLY contain MISSING or BROKEN features. Do NOT list features that ARE implemented.\n\nDesign token usage (hardcoded colors, numeric spacing) is NOT a functionality issue — ignore it here.\n${sourceBlock}\n\nReturn your evaluation using the evaluate_functionality tool.\n\nExample outputs:\n- All features present: {\"pass\": true}\n- Missing features: {\"pass\": false, \"issues\": [\"Add Task form is not rendered — only state logic exists\", \"useAction('refresh') is declared but never wired to any button\"]}\n\nOnly missing or broken features go in the issues array. Nothing else.`;\n\n    case 'crash':\n      return `Evaluate CRASH SAFETY: Are there ACTUAL runtime crash scenarios in this component?\n\n**CRITICAL RULE:** The \"issues\" array must ONLY contain code that WILL crash. Do NOT include safe code in the issues array — not even to explain why it's safe. If you checked a line and it has a guard, do not mention it at all.\n\n**These WILL crash (include in issues):**\n- .map() / .filter() / .length on a variable that is never initialized\n- Accessing a property of a value that could be undefined without a guard\n- Calling a function that is never defined\n- useStream().latest.field WITHOUT null guard — .latest is T | null\n- Optional Props field (has \\`?\\` in interface) accessed as props.field.x without guard — WILL crash if undefined\n- Array item field that could be undefined: items.map(item => item.priority.toUpperCase()) when priority is optional\n\n**These are SAFE (do NOT include in issues):**\n- Optional chaining: props.items?.map() — SAFE, skip it\n- Fallback values: items || [] — SAFE, skip it\n- Default parameters: (items = []) — SAFE, skip it\n- useState initializer: useState([]) — SAFE, skip it\n- Null check: items && items.map() — SAFE, skip it\n- stream.latest && stream.latest.field — SAFE, guarded\n- stream.all.map(...) — SAFE, .all is always an array (never null)\n- stream.all.length — SAFE, always a number\n- toolResult.data && toolResult.data.field — SAFE, guarded\n\nIf everything is safe, return {\"pass\": true} with no issues array.\n${sourceBlock}\n\nReturn your evaluation using the evaluate_crash tool.\n\nExample outputs:\n- Everything safe: {\"pass\": true}\n- Has crash: {\"pass\": false, \"issues\": [\"Line 41: selectedItem.name — selectedItem comes from .find() which returns undefined when no match, no guard\"]}\n\nOnly unguarded crash risks go in the issues array. Nothing else.`;\n\n    case 'interactivity':\n      return `Evaluate INTERACTIVITY: Does this component have sufficient interactive elements for its purpose?\n\nConsider: forms need submit buttons, lists need selection/navigation, editable content needs save/cancel, interactive elements need hover/focus feedback.\n\n**Contract actions** (if present): every useAction hook from the data contract should be triggered by a visible, clickable UI element. A declared action with no triggering button/link/form is a missing interaction.\n\n**Only list MISSING interactive elements in the issues array.** Do not list interactions that are present and working.\n${sourceBlock}\n\nReturn your evaluation using the evaluate_interactivity tool.\n\nExample outputs:\n- Sufficient: {\"result\": \"pass\"}\n- Minor gap: {\"result\": \"warn\", \"issues\": [\"No hover feedback on list items — users can't tell items are clickable\"]}\n- Severe gap: {\"result\": \"fail\", \"issues\": [\"Form has no submit button — users cannot complete the workflow\"]}\n\nUse 'fail' only for issues that block the component's core purpose.`;\n\n    case 'accessibility':\n      return `Evaluate ACCESSIBILITY: Check for missing labels, alt text, semantic HTML, keyboard navigation.\n\nLook for: form inputs without labels, images without alt text, non-semantic elements used for interactive purposes, missing aria attributes, heading hierarchy issues.\n\n**Only list MISSING accessibility features in the issues array.** Do not list things that are already accessible.\n${sourceBlock}\n\nReturn your evaluation using the evaluate_accessibility tool.\n\nExample outputs:\n- Clean: {\"result\": \"pass\"}\n- Minor issues: {\"result\": \"warn\", \"issues\": [\"Email input missing label prop\", \"Task edit uses double-click with no keyboard alternative\"]}\n- Severe: {\"result\": \"fail\", \"issues\": [\"No form inputs have labels — entire form is inaccessible to screen readers\"]}\n\nUse 'fail' only if the issue blocks delivery. Use 'warn' for improvements.`;\n\n    case 'layout':\n      return `Evaluate LAYOUT: Check spacing, alignment, visual grouping, and overall composition.\n\nLook for: inconsistent spacing, cramped elements, missing visual grouping (cards/sections), overlapping concerns in single containers, poor responsive patterns.\n\n**Only list ACTUAL layout problems in the issues array.** Do not list layout aspects that are working correctly.\n${sourceBlock}\n\nReturn your evaluation using the evaluate_layout tool.\n\nExample outputs:\n- Clean: {\"result\": \"pass\"}\n- Minor issues: {\"result\": \"warn\", \"issues\": [\"Header and content area have no visual separation\", \"Button group lacks consistent spacing\"]}\n- Severe: {\"result\": \"fail\", \"issues\": [\"All content rendered in a flat list with no visual structure — needs cards or sections\"]}\n\nUse 'fail' only for layouts that are fundamentally broken.`;\n\n    case 'loading':\n      return `Evaluate LOADING/EMPTY/ERROR STATES: Does the component handle async data, empty collections, and error cases?\n\nLook for: arrays that could be empty but have no empty state message, missing error handling for operations that can fail.\n\n**Contract-specific checks** (if present):\n- useStream hooks deliver data over time — consider showing a \"waiting for data\" state before .latest arrives\n- clientCapabilities hooks (e.g. useGeolocation) may return permission denials or undefined values — defensive null guards expected before threading into JSX\n\n**Only list MISSING states in the issues array.** Do not list states that are already handled. Components that receive all data via props (no streams, no async capabilities) do NOT need loading states — return pass.\n${sourceBlock}\n\nReturn your evaluation using the evaluate_loading tool.\n\nExample outputs:\n- Not applicable (props-driven): {\"result\": \"pass\"}\n- Missing states: {\"result\": \"warn\", \"issues\": [\"No empty state when items array is empty — should show 'No items found'\"]}\n- Severe: {\"result\": \"fail\", \"issues\": [\"Component shows blank screen when data hasn't loaded yet — needs loading spinner\"]}`;\n\n    case 'visual':\n      return `Evaluate VISUAL CONSISTENCY: Is the component using the design system correctly?\n\nLook for: hardcoded colors instead of CSS variables, numeric/raw-CSS spacing props that bypass the t-shirt scale, style objects that bypass the design system.\n\n**Only list ACTUAL design system violations in the issues array.** Do not list correct token usage. A t-shirt-scale spacing name (\\`gap=\"md\"\\`, \\`padding=\"lg\"\\`) IS correct token usage — never flag it as hardcoded; only numeric (\\`padding={16}\\`) or raw-CSS-length (\\`gap=\"13px\"\\`) spacing is a violation. Note: intentional custom colors (e.g., status indicators, category colors) are acceptable when no semantic token fits.\n${sourceBlock}\n\nReturn your evaluation using the evaluate_visual tool.\n\nExample outputs:\n- Clean: {\"result\": \"pass\"}\n- Minor issues: {\"result\": \"warn\", \"issues\": [\"Line 45: hardcoded '#f3f4f6' should use var(--ggui-color-sunken)\", \"Line 80: numeric padding={16} should use a scale name like padding='md'\"]}\n- Severe: {\"result\": \"fail\", \"issues\": [\"Component uses entirely custom styling with no design system tokens\"]}\n\nUse 'fail' only for pervasive design system violations. Use 'warn' for isolated cases.`;\n\n    default:\n      return `Evaluate ${criterion.toUpperCase()}.${sourceBlock}\\n\\nReturn your evaluation using the evaluate_${criterion} tool.`;\n  }\n}\n\n// =============================================================================\n// Default models\n// =============================================================================\n\nfunction getDefaultEvalModel(provider: string): string {\n  switch (provider) {\n    case 'claude':\n      return 'claude-haiku-5-5';\n    case 'openai':\n      return 'gpt-5.6-luna';\n    case 'google':\n      return 'gemini-3.5-flash-lite';\n    default:\n      return 'claude-haiku-5-5';\n  }\n}\n\n// =============================================================================\n// Per-criterion evaluation call\n// =============================================================================\n\ninterface CriterionResult {\n  criterion: string;\n  tier: 1 | 2;\n  issues: EvalIssue[];\n  pass: boolean;\n  /**\n   * Did this criterion actually produce a verdict? The evaluator\n   * fail-opens (`pass: true`) on LLM errors and missing tool calls so a\n   * flaky call never blocks generation — this field is what keeps that\n   * fail-open honest: `skipped` means `pass` carries zero evidence.\n   */\n  coverage: CriterionRunStatus;\n  /** Populated for `skipped` — why the criterion produced no verdict. */\n  skipReason?: string;\n  /** This call's token accounting. The cache counts are present only when the\n   *  provider reported them — absent means unreported, never zero (ggui#1281). */\n  tokens: { in: number; out: number; cacheRead?: number; cacheCreation?: number };\n  durationMs: number;\n}\n\n/** A call's token accounting, carrying the prompt-cache counts only when the\n *  provider reported them (the router's rule: absent is \"unreported\"). */\nfunction callTokens(response: LLMToolCallResponse): CriterionResult['tokens'] {\n  return {\n    in: response.inputTokens,\n    out: response.outputTokens,\n    ...(response.cacheReadTokens !== undefined ? { cacheRead: response.cacheReadTokens } : {}),\n    ...(response.cacheCreationTokens !== undefined ? { cacheCreation: response.cacheCreationTokens } : {}),\n  };\n}\n\n/**\n * The evaluation's summed token accounting. `inputTokens` is the NON-cached\n * input (the router's convention); the cache counts are the sums of what the\n * calls reported, and stay absent when no call reported one — so a caller can\n * tell \"no cache reads\" from \"the provider does not say\" (ggui#1281).\n */\nexport interface LLMEvalTokenUsage {\n  inputTokens: number;\n  outputTokens: number;\n  cacheReadTokens?: number;\n  cacheCreationTokens?: number;\n}\n\n/**\n * All valid categories for LLM evaluation — derived from the ONE static\n * list in types-public so the coverage stamp here and the harness's\n * bypass stamp (`notApplicableCoverage`) can never disagree on the set.\n */\nconst TIER_1_CRITERIA = LLM_EVAL_STATIC_CRITERIA.filter((c) => c.tier === 1).map((c) => c.criterion);\nconst TIER_2_CRITERIA = LLM_EVAL_STATIC_CRITERIA.filter((c) => c.tier === 2).map((c) => c.criterion);\n\nasync function evalCriterion(\n  criterion: string,\n  tier: 1 | 2,\n  motherPrompt: string,\n  numberedSource: string,\n  agent: LLMAgent,\n  model: string,\n  designMode: DesignMode,\n): Promise<CriterionResult> {\n  const start = Date.now();\n  const tool = tier === 1 ? makeTier1Tool(criterion) : makeTier2Tool(criterion);\n  const userPrompt = getCriterionPrompt(criterion, numberedSource, designMode);\n\n  try {\n    const response = await agent.callTools(\n      model,\n      motherPrompt,\n      userPrompt,\n      [tool],\n      'required',\n    );\n\n    const call = response.toolCalls[0];\n    const durationMs = Date.now() - start;\n\n    if (!call) {\n      console.warn(`[eval] ${criterion}: no tool call returned`);\n      return {\n        criterion,\n        tier,\n        issues: [],\n        pass: true,\n        coverage: 'skipped',\n        skipReason: 'no tool call returned',\n        tokens: callTokens(response),\n        durationMs,\n      };\n    }\n\n    const raw = call.input as JsonObject;\n    const issues: EvalIssue[] = [];\n    let pass: boolean;\n\n    if (tier === 1) {\n      // Tier 1: binary pass/fail\n      pass = raw.pass === true;\n      const rawIssues = coerceStringArray(raw.issues);\n      for (const desc of rawIssues) {\n        issues.push({\n          tier: 1,\n          result: 'fail',\n          category: criterion as EvalCategory,\n          severity: 'critical',\n          description: desc,\n          fix: desc, // The issue description often includes the fix for tier 1\n        });\n      }\n      // If pass is false but no issues listed, add a generic one\n      if (!pass && issues.length === 0) {\n        issues.push({\n          tier: 1,\n          result: 'fail',\n          category: criterion as EvalCategory,\n          severity: 'critical',\n          description: `${criterion} check failed`,\n          fix: `Fix ${criterion} issues`,\n        });\n      }\n    } else {\n      // Tier 2: fail/warn/pass\n      const result = (raw.result as string) ?? 'pass';\n      pass = result === 'pass';\n      const rawIssues = coerceStringArray(raw.issues);\n      const outcome = result === 'fail' ? 'fail' : 'warn';\n      for (const desc of rawIssues) {\n        issues.push({\n          tier: 2,\n          result: outcome as 'fail' | 'warn',\n          category: criterion as EvalCategory,\n          severity: result === 'fail' ? 'critical' : 'major',\n          description: desc,\n          fix: desc,\n        });\n      }\n    }\n\n    return {\n      criterion,\n      tier,\n      issues,\n      pass,\n      coverage: 'ran',\n      tokens: callTokens(response),\n      durationMs,\n    };\n  } catch (e) {\n    const durationMs = Date.now() - start;\n    const message = e instanceof Error ? e.message : String(e);\n    console.error(`[eval] ${criterion}: error — ${message}`);\n    // On error, treat as pass to avoid blocking on flaky LLM calls.\n    // `coverage: 'skipped'` discloses the fail-open: this criterion\n    // produced NO verdict, and consumers must not read its `pass`\n    // membership as evidence.\n    return {\n      criterion,\n      tier,\n      issues: [],\n      pass: true,\n      coverage: 'skipped',\n      skipReason: message,\n      tokens: { in: 0, out: 0 },\n      durationMs,\n    };\n  }\n}\n\n// =============================================================================\n// Pre-warming\n// =============================================================================\n\n/**\n * Pre-warm evaluation context during the coding phase.\n *\n * Builds the mother prompt during coding so it's ready when eval starts.\n * The result can be passed to runLLMEvaluation() to avoid redundant work.\n */\nexport async function preWarmEval(\n  context: LLMEvalContext,\n  _config: LLMEvalConfig,\n): Promise<PreWarmedEvalContext> {\n  const motherPrompt = buildMotherPrompt({\n    originalPrompt: context.originalPrompt,\n    contract: context.contract,\n    designSystemSummary: context.designContext,\n    designMode: context.designMode,\n    profile: context.profile,\n  });\n\n  return { motherPrompt, dynamicCriteria: [] };\n}\n\n// =============================================================================\n// Response parsing (backward compat — used by tests)\n// =============================================================================\n\n/** Tier 1 categories — always tier 1 regardless of what LLM returns */\nconst TIER_1_CATEGORIES = new Set<string>(['functionality', 'crash']);\n\n/** Tier 2 categories */\nconst TIER_2_CATEGORIES = new Set<string>([\n  'interactivity',\n  'accessibility',\n  'layout',\n  'loading',\n  'visual',\n]);\n\n/** All valid categories for LLM evaluation */\nconst VALID_CATEGORIES = new Set<string>([\n  ...TIER_1_CATEGORIES,\n  ...TIER_2_CATEGORIES,\n]);\n\n/**\n * Parse the raw tool call input into a typed EvalResult.\n *\n * Exported for testing — not part of the public API.\n * Retained for backward compatibility with existing tests.\n */\nexport function parseEvalResponse(raw: JsonObject): EvalResult {\n  const rawIssues = (raw.issues ?? []) as Array<JsonObject>;\n  const rawPass = (raw.pass ?? []) as string[];\n\n  const issues: EvalIssue[] = rawIssues\n    .filter((i) => {\n      const cat = i.category as string;\n      return VALID_CATEGORIES.has(cat);\n    })\n    .map((i) => {\n      const category = i.category as string;\n      const isTier1 = TIER_1_CATEGORIES.has(category);\n\n      // Enforce correct tier based on category\n      const tier = isTier1 ? 1 : 2;\n\n      // Tier 1 issues are always 'fail' (critical checks)\n      // Tier 2 issues: LLM decides 'fail' or 'warn'\n      const result = isTier1\n        ? 'fail'\n        : ((i.result as string) === 'fail' ? 'fail' : 'warn');\n\n      return {\n        tier: tier as 1 | 2,\n        result: result as 'fail' | 'warn',\n        category: category as EvalCategory,\n        severity: (i.severity as 'critical' | 'major') || (isTier1 ? 'critical' : 'major'),\n        description: (i.description as string) || '',\n        fix: (i.fix as string) || '',\n      };\n    });\n\n  return {\n    issues,\n    pass: rawPass,\n  };\n}\n\n// =============================================================================\n// Main evaluation function\n// =============================================================================\n\n/**\n * Run LLM evaluation with parallel per-criterion calls.\n *\n * Runs 7 static criteria (2 tier 1 + 5 tier 2) plus 0-3 dynamic criteria\n * generated from the user's request. All run in parallel with a shared\n * mother prompt (cached by the provider). Wall time ~2s for all criteria.\n *\n * @param preWarmedContext - Optional pre-warmed context from `preWarmEval()`.\n *   If provided, skips mother prompt building and dynamic criteria generation.\n */\nexport async function runLLMEvaluation(\n  context: LLMEvalContext,\n  config: LLMEvalConfig,\n  preWarmedContext?: PreWarmedEvalContext | null,\n): Promise<EvalResult & LLMEvalTokenUsage> {\n  const providerName = config.provider;\n  const model = config.model || getDefaultEvalModel(providerName);\n  const routerProvider =\n    providerName === 'claude'\n      ? ('anthropic' as const)\n      : (providerName as 'openai' | 'google');\n\n  // Use pre-warmed context if available, otherwise build fresh\n  const designMode = context.designMode ?? DEFAULT_DESIGN_MODE;\n  const motherPrompt = preWarmedContext?.motherPrompt ?? buildMotherPrompt({\n    originalPrompt: context.originalPrompt,\n    contract: context.contract,\n    designSystemSummary: context.designContext,\n    designMode,\n    profile: context.profile,\n  });\n\n  // Add line numbers to source\n  const numberedSource = context.sourceCode\n    .split('\\n')\n    .map((line, i) => `${String(i + 1).padStart(4)} | ${line}`)\n    .join('\\n');\n\n  // Create a single agent — all calls share the same client + cache.\n  // Pass the full config (not just the bare provider string) so a\n  // caller-supplied `routeOverride` (#484) reaches the constructed\n  // `LLMAgent` instance and it never falls back to `process.env`.\n  const agent = createAgent({\n    provider: routerProvider,\n    model,\n    routeOverride: config.routeOverride,\n    onRetry: config.onRetry,\n  });\n\n  // Dynamic criteria removed — they generated overly strict edge-case checks\n  // that wasted coding turns on nitpicks. Will be replaced by a holistic\n  // reviewer in auto-improve mode.\n\n  const evalStart = Date.now();\n\n  const totalCriteriaCount = TIER_1_CRITERIA.length + TIER_2_CRITERIA.length;\n\n  // Run ALL criteria in parallel: 7 static (2 tier 1 + 5 tier 2)\n  const results = await Promise.all([\n    ...TIER_1_CRITERIA.map(c => evalCriterion(c, 1, motherPrompt, numberedSource, agent, model, designMode)),\n    ...TIER_2_CRITERIA.map(c => evalCriterion(c, 2, motherPrompt, numberedSource, agent, model, designMode)),\n  ]);\n\n  // Merge results\n  const allIssues: EvalIssue[] = [];\n  const passCategories: string[] = [];\n  let totalIn = 0;\n  let totalOut = 0;\n  // Measured, not estimated: summed from what the calls reported; absent\n  // while no call has reported one (ggui#1281).\n  let cacheRead: number | undefined;\n  let cacheCreation: number | undefined;\n\n  for (const r of results) {\n    allIssues.push(...r.issues);\n    if (r.pass) {\n      passCategories.push(r.criterion);\n    }\n    totalIn += r.tokens.in;\n    totalOut += r.tokens.out;\n    if (r.tokens.cacheRead !== undefined) cacheRead = (cacheRead ?? 0) + r.tokens.cacheRead;\n    if (r.tokens.cacheCreation !== undefined) cacheCreation = (cacheCreation ?? 0) + r.tokens.cacheCreation;\n  }\n\n  const wallMs = Date.now() - evalStart;\n\n  // Per-criterion coverage — one row per criterion, ALWAYS all of them,\n  // so a consumer can tell \"ran clean\" from \"silently didn't run\"\n  // (the fail-open paths above return pass:true with zero evidence).\n  const criteriaCoverage: CriterionCoverage[] = results.map((r) => ({\n    criterion: r.criterion,\n    tier: r.tier,\n    status: r.coverage,\n    ...(r.skipReason !== undefined ? { reason: r.skipReason } : {}),\n  }));\n  const skippedCount = criteriaCoverage.filter((c) => c.status === 'skipped').length;\n\n  // Log per-criterion results\n  for (const r of results) {\n    const issueCount = r.issues.length;\n    const status = r.coverage === 'skipped'\n      ? 'SKIPPED'\n      : r.pass ? 'PASS' : r.tier === 1 ? 'FAIL' : (r.issues.some(i => i.result === 'fail') ? 'FAIL' : 'WARN');\n    const issueLabel = issueCount > 0 ? ` (${issueCount} issue${issueCount > 1 ? 's' : ''})` : '';\n    console.log(`[eval] ${r.criterion}: ${status}${issueLabel} ${(r.durationMs / 1000).toFixed(1)}s`);\n  }\n  console.log(\n    `[eval] total: ${(wallMs / 1000).toFixed(1)}s wall (${totalCriteriaCount} parallel) | ` +\n    `in=${totalIn} out=${totalOut} cacheRead=${cacheRead ?? 'unreported'} cacheWrite=${cacheCreation ?? 'unreported'} | ` +\n    `issues=${allIssues.length} pass=${passCategories.length}` +\n    (skippedCount > 0 ? ` skipped=${skippedCount}` : ''),\n  );\n\n  return {\n    issues: allIssues,\n    pass: passCategories,\n    criteriaCoverage,\n    inputTokens: totalIn,\n    outputTokens: totalOut,\n    ...(cacheRead !== undefined ? { cacheReadTokens: cacheRead } : {}),\n    ...(cacheCreation !== undefined ? { cacheCreationTokens: cacheCreation } : {}),\n  };\n}\n"]}