@dalmia/calibrate-mcp 0.0.23 → 0.0.25

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (63) hide show
  1. package/bin/mcp-server.js +50 -27
  2. package/bin/mcp-server.js.map +18 -18
  3. package/esm/funcs/agentTestsGetBenchmark.js +6 -1
  4. package/esm/funcs/agentTestsGetBenchmark.js.map +1 -1
  5. package/esm/funcs/agentTestsGetRun.js +6 -1
  6. package/esm/funcs/agentTestsGetRun.js.map +1 -1
  7. package/esm/funcs/annotationTasksGetSummary.js +2 -0
  8. package/esm/funcs/annotationTasksGetSummary.js.map +1 -1
  9. package/esm/funcs/evaluatorsGet.js +5 -1
  10. package/esm/funcs/evaluatorsGet.js.map +1 -1
  11. package/esm/landing-page.js +1 -1
  12. package/esm/lib/config.d.ts +2 -2
  13. package/esm/lib/config.js +2 -2
  14. package/esm/mcp-server/mcp-server.js +1 -1
  15. package/esm/mcp-server/server.js +1 -1
  16. package/esm/models/{evaluatordetailresponse.d.ts → evaluatordetailresponsecompact.d.ts} +17 -17
  17. package/esm/models/evaluatordetailresponsecompact.d.ts.map +1 -0
  18. package/esm/models/{evaluatordetailresponse.js → evaluatordetailresponsecompact.js} +14 -13
  19. package/esm/models/evaluatordetailresponsecompact.js.map +1 -0
  20. package/esm/models/{evaluatorversionresponse.d.ts → evaluatorversioncompact.d.ts} +4 -4
  21. package/esm/models/evaluatorversioncompact.d.ts.map +1 -0
  22. package/esm/models/{evaluatorversionresponse.js → evaluatorversioncompact.js} +3 -3
  23. package/esm/models/evaluatorversioncompact.js.map +1 -0
  24. package/esm/models/getagenttestrunstatusagenttestsruntaskidgetop.d.ts +2 -0
  25. package/esm/models/getagenttestrunstatusagenttestsruntaskidgetop.d.ts.map +1 -1
  26. package/esm/models/getagenttestrunstatusagenttestsruntaskidgetop.js +2 -0
  27. package/esm/models/getagenttestrunstatusagenttestsruntaskidgetop.js.map +1 -1
  28. package/esm/models/getbenchmarkstatusagenttestsbenchmarktaskidgetop.d.ts +2 -0
  29. package/esm/models/getbenchmarkstatusagenttestsbenchmarktaskidgetop.d.ts.map +1 -1
  30. package/esm/models/getbenchmarkstatusagenttestsbenchmarktaskidgetop.js +2 -0
  31. package/esm/models/getbenchmarkstatusagenttestsbenchmarktaskidgetop.js.map +1 -1
  32. package/esm/models/getevaluatorendpointevaluatorsevaluatoruuidgetop.d.ts +3 -2
  33. package/esm/models/getevaluatorendpointevaluatorsevaluatoruuidgetop.d.ts.map +1 -1
  34. package/esm/models/getevaluatorendpointevaluatorsevaluatoruuidgetop.js +3 -2
  35. package/esm/models/getevaluatorendpointevaluatorsevaluatoruuidgetop.js.map +1 -1
  36. package/esm/models/tasksummaryannotationtaskstaskuuidsummarygetop.d.ts +2 -0
  37. package/esm/models/tasksummaryannotationtaskstaskuuidsummarygetop.d.ts.map +1 -1
  38. package/esm/models/tasksummaryannotationtaskstaskuuidsummarygetop.js +2 -0
  39. package/esm/models/tasksummaryannotationtaskstaskuuidsummarygetop.js.map +1 -1
  40. package/esm/models/tasksummaryresponse.d.ts +2 -2
  41. package/esm/models/tasksummaryresponse.d.ts.map +1 -1
  42. package/esm/models/tasksummaryresponse.js +3 -2
  43. package/esm/models/tasksummaryresponse.js.map +1 -1
  44. package/package.json +1 -1
  45. package/src/funcs/agentTestsGetBenchmark.ts +6 -1
  46. package/src/funcs/agentTestsGetRun.ts +6 -1
  47. package/src/funcs/annotationTasksGetSummary.ts +2 -0
  48. package/src/funcs/evaluatorsGet.ts +5 -1
  49. package/src/landing-page.ts +1 -1
  50. package/src/lib/config.ts +2 -2
  51. package/src/mcp-server/mcp-server.ts +1 -1
  52. package/src/mcp-server/server.ts +1 -1
  53. package/src/models/{evaluatordetailresponse.ts → evaluatordetailresponsecompact.ts} +29 -28
  54. package/src/models/{evaluatorversionresponse.ts → evaluatorversioncompact.ts} +5 -5
  55. package/src/models/getagenttestrunstatusagenttestsruntaskidgetop.ts +8 -0
  56. package/src/models/getbenchmarkstatusagenttestsbenchmarktaskidgetop.ts +8 -0
  57. package/src/models/getevaluatorendpointevaluatorsevaluatoruuidgetop.ts +9 -5
  58. package/src/models/tasksummaryannotationtaskstaskuuidsummarygetop.ts +8 -0
  59. package/src/models/tasksummaryresponse.ts +4 -5
  60. package/esm/models/evaluatordetailresponse.d.ts.map +0 -1
  61. package/esm/models/evaluatordetailresponse.js.map +0 -1
  62. package/esm/models/evaluatorversionresponse.d.ts.map +0 -1
  63. package/esm/models/evaluatorversionresponse.js.map +0 -1
@@ -1 +1 @@
1
- {"version":3,"file":"tasksummaryannotationtaskstaskuuidsummarygetop.js","sourceRoot":"","sources":["../../src/models/tasksummaryannotationtaskstaskuuidsummarygetop.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;AAClC,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;AAElC;;GAEG;AACH,MAAM,CAAC,MAAM,KAAK,GAAG;IACnB,GAAG,EAAE,KAAK;IACV,IAAI,EAAE,MAAM;CACJ,CAAC;AAMX,MAAM,CAAC,MAAM,eAAe,GAAG,CAAC,CAAC,IAAI,CAAC;IACpC,KAAK;IACL,MAAM;CACP,CAAC,CAAC,QAAQ,CAAC,gBAAgB,CAAC,CAAC;AAc9B,MAAM,CAAC,MAAM,6DAA6D,GACP,CAAC,CAAC,MAAM,CAAC;IACxE,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAC1B,oGAAoG,CACrG,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACvB,KAAK,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,QAAQ,CAAC,mCAAmC,CAAC;IACxE,SAAS,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC5C,+JAA+J,CAChK;IACD,MAAM,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,QAAQ,CACjC,kDAAkD,CACnD;IACD,KAAK,EAAE,eAAe,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,QAAQ,CAAC,gBAAgB,CAAC;IACjE,CAAC,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CACpB,uEAAuE,CACxE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACvB,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,YAAY,CAAC,CAAC,QAAQ,CAChD,0BAA0B,CAC3B;IACD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,2BAA2B,CAAC;IAC3D,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;CAC1C,CAAC,CAAC;AAML,MAAM,CAAC,MAAM,8DAA8D,GACP,CAAC,CAAC,KAAK,CAAC;IACxE,6BAA6B;IAC7B,6BAA6B;CAC9B,CAAC,CAAC"}
1
+ {"version":3,"file":"tasksummaryannotationtaskstaskuuidsummarygetop.js","sourceRoot":"","sources":["../../src/models/tasksummaryannotationtaskstaskuuidsummarygetop.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;AAClC,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;AAElC;;GAEG;AACH,MAAM,CAAC,MAAM,KAAK,GAAG;IACnB,GAAG,EAAE,KAAK;IACV,IAAI,EAAE,MAAM;CACJ,CAAC;AAMX,MAAM,CAAC,MAAM,eAAe,GAAG,CAAC,CAAC,IAAI,CAAC;IACpC,KAAK;IACL,MAAM;CACP,CAAC,CAAC,QAAQ,CAAC,gBAAgB,CAAC,CAAC;AAgB9B,MAAM,CAAC,MAAM,6DAA6D,GACP,CAAC,CAAC,MAAM,CAAC;IACxE,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC1C,qUAAqU,CACtU;IACD,iBAAiB,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CACpD,qFAAqF,CACtF;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAC1B,oGAAoG,CACrG,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACvB,KAAK,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,QAAQ,CAAC,mCAAmC,CAAC;IACxE,SAAS,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC5C,+JAA+J,CAChK;IACD,MAAM,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,QAAQ,CACjC,kDAAkD,CACnD;IACD,KAAK,EAAE,eAAe,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,QAAQ,CAAC,gBAAgB,CAAC;IACjE,CAAC,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CACpB,uEAAuE,CACxE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACvB,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,YAAY,CAAC,CAAC,QAAQ,CAChD,0BAA0B,CAC3B;IACD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,2BAA2B,CAAC;IAC3D,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;CAC1C,CAAC,CAAC;AAML,MAAM,CAAC,MAAM,8DAA8D,GACP,CAAC,CAAC,KAAK,CAAC;IACxE,6BAA6B;IAC7B,6BAA6B;CAC9B,CAAC,CAAC"}
@@ -32,9 +32,9 @@ export type TaskSummaryResponse = {
32
32
  rows: Array<{
33
33
  [k: string]: any;
34
34
  }>;
35
- item_comments: {
35
+ item_comments?: {
36
36
  [k: string]: any;
37
- };
37
+ } | null | undefined;
38
38
  pagination: PaginationMeta;
39
39
  AdditionalProperties?: {
40
40
  [k: string]: any;
@@ -1 +1 @@
1
- {"version":3,"file":"tasksummaryresponse.d.ts","sourceRoot":"","sources":["../../src/models/tasksummaryresponse.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EAAE,cAAc,EAA4B,MAAM,qBAAqB,CAAC;AAE/E;;GAEG;AACH,eAAO,MAAM,QAAQ;;;;;CAKX,CAAC;AACX;;GAEG;AACH,MAAM,MAAM,QAAQ,GAAG,UAAU,CAAC,OAAO,QAAQ,CAAC,CAAC;AAEnD,eAAO,MAAM,kBAAkB;;;;;EAKA,CAAC;AAEhC,MAAM,MAAM,mBAAmB,GAAG;IAChC,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,QAAQ,CAAC;IACpB,UAAU,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,CAAC;IACxC,UAAU,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,CAAC;IACxC,IAAI,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,CAAC;IAClC,aAAa,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC;IACpC,UAAU,EAAE,cAAc,CAAC;IAC3B,oBAAoB,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC;CAC7C,CAAC;AAEF,eAAO,MAAM,6BAA6B,EAAE,CAAC,CAAC,OAAO,CAAC,mBAAmB,CAqBtE,CAAC;AAEJ,eAAO,MAAM,qCAAqC,EAAE,CAAC,CAAC,OAAO,CAAC,GAAG,CAe/D,CAAC"}
1
+ {"version":3,"file":"tasksummaryresponse.d.ts","sourceRoot":"","sources":["../../src/models/tasksummaryresponse.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EAAE,cAAc,EAA4B,MAAM,qBAAqB,CAAC;AAE/E;;GAEG;AACH,eAAO,MAAM,QAAQ;;;;;CAKX,CAAC;AACX;;GAEG;AACH,MAAM,MAAM,QAAQ,GAAG,UAAU,CAAC,OAAO,QAAQ,CAAC,CAAC;AAEnD,eAAO,MAAM,kBAAkB;;;;;EAKA,CAAC;AAEhC,MAAM,MAAM,mBAAmB,GAAG;IAChC,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,QAAQ,CAAC;IACpB,UAAU,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,CAAC;IACxC,UAAU,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,CAAC;IACxC,IAAI,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,CAAC;IAClC,aAAa,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,GAAG,IAAI,GAAG,SAAS,CAAC;IACxD,UAAU,EAAE,cAAc,CAAC;IAC3B,oBAAoB,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC;CAC7C,CAAC;AAEF,eAAO,MAAM,6BAA6B,EAAE,CAAC,CAAC,OAAO,CAAC,mBAAmB,CAoBtE,CAAC;AAEJ,eAAO,MAAM,qCAAqC,EAAE,CAAC,CAAC,OAAO,CAAC,GAAG,CAe/D,CAAC"}
@@ -22,7 +22,8 @@ export const TaskType$zodSchema = z.enum([
22
22
  export const TaskSummaryResponse$zodSchema = collectExtraKeys$(z.object({
23
23
  annotators: z.array(z.record(z.string(), z.any())).describe("The annotators who labelled or commented on items in the task, stable across pages"),
24
24
  evaluators: z.array(z.record(z.string(), z.any())).describe("The task's evaluators, stable across pages, each with its live rubric"),
25
- item_comments: z.record(z.string(), z.any()).describe("Annotator comments, keyed by item ID"),
25
+ item_comments: z.record(z.string(), z.any()).nullable().optional()
26
+ .describe("Annotator comments, keyed by item ID"),
26
27
  pagination: PaginationMeta$zodSchema,
27
28
  rows: z.array(z.record(z.string(), z.any())).describe("One row per item, evaluator, and version on this page, with the evaluator's value and each annotator's label"),
28
29
  task_id: z.string().describe("ID of the task"),
@@ -32,7 +33,7 @@ export const TaskSummaryResponse$zodSchemaOutbound = z.object({
32
33
  AdditionalProperties: z.record(z.string(), z.any()).optional(),
33
34
  annotators: z.array(z.record(z.string(), z.any())),
34
35
  evaluators: z.array(z.record(z.string(), z.any())),
35
- item_comments: z.record(z.string(), z.any()),
36
+ item_comments: z.nullable(z.record(z.string(), z.any())).optional(),
36
37
  pagination: PaginationMeta$zodSchema,
37
38
  rows: z.array(z.record(z.string(), z.any())),
38
39
  task_id: z.string(),
@@ -1 +1 @@
1
- {"version":3,"file":"tasksummaryresponse.js","sourceRoot":"","sources":["../../src/models/tasksummaryresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,gBAAgB,IAAI,iBAAiB,EAAE,MAAM,mBAAmB,CAAC;AAE1E,OAAO,EAAkB,wBAAwB,EAAE,MAAM,qBAAqB,CAAC;AAE/E;;GAEG;AACH,MAAM,CAAC,MAAM,QAAQ,GAAG;IACtB,GAAG,EAAE,KAAK;IACV,GAAG,EAAE,KAAK;IACV,UAAU,EAAE,aAAa;IACzB,YAAY,EAAE,cAAc;CACpB,CAAC;AAMX,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC,CAAC,IAAI,CAAC;IACvC,KAAK;IACL,KAAK;IACL,aAAa;IACb,cAAc;CACf,CAAC,CAAC,QAAQ,CAAC,kBAAkB,CAAC,CAAC;AAahC,MAAM,CAAC,MAAM,6BAA6B,GACxC,iBAAiB,CACf,CAAC,CAAC,MAAM,CAAC;IACP,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,CACzD,oFAAoF,CACrF;IACD,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,CACzD,uEAAuE,CACxE;IACD,aAAa,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,CACnD,sCAAsC,CACvC;IACD,UAAU,EAAE,wBAAwB;IACpC,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,CACnD,8GAA8G,CAC/G;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,gBAAgB,CAAC;IAC9C,SAAS,EAAE,kBAAkB,CAAC,QAAQ,CAAC,kBAAkB,CAAC;CAC3D,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,GAAG,EAAE,CAAC,EACpB,sBAAsB,EACtB,IAAI,CACL,CAAC;AAEJ,MAAM,CAAC,MAAM,qCAAqC,GAAmB,CAAC,CAAC,MAAM,CAAC;IAC5E,oBAAoB,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE;IAC9D,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC;IAClD,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC;IAClD,aAAa,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC;IAC5C,UAAU,EAAE,wBAAwB;IACpC,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC;IAC5C,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE;IACnB,SAAS,EAAE,kBAAkB;CAC9B,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,EAAE;IACjB,MAAM,EAAE,oBAAoB,EAAE,GAAG,IAAI,EAAE,GAAG,CAAC,CAAC;IAC5C,OAAO;QACL,GAAG,IAAI;QACP,GAAG,oBAAoB;KACxB,CAAC;AACJ,CAAC,CAAC,CAAC"}
1
+ {"version":3,"file":"tasksummaryresponse.js","sourceRoot":"","sources":["../../src/models/tasksummaryresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,gBAAgB,IAAI,iBAAiB,EAAE,MAAM,mBAAmB,CAAC;AAE1E,OAAO,EAAkB,wBAAwB,EAAE,MAAM,qBAAqB,CAAC;AAE/E;;GAEG;AACH,MAAM,CAAC,MAAM,QAAQ,GAAG;IACtB,GAAG,EAAE,KAAK;IACV,GAAG,EAAE,KAAK;IACV,UAAU,EAAE,aAAa;IACzB,YAAY,EAAE,cAAc;CACpB,CAAC;AAMX,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC,CAAC,IAAI,CAAC;IACvC,KAAK;IACL,KAAK;IACL,aAAa;IACb,cAAc;CACf,CAAC,CAAC,QAAQ,CAAC,kBAAkB,CAAC,CAAC;AAahC,MAAM,CAAC,MAAM,6BAA6B,GACxC,iBAAiB,CACf,CAAC,CAAC,MAAM,CAAC;IACP,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,CACzD,oFAAoF,CACrF;IACD,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,CACzD,uEAAuE,CACxE;IACD,aAAa,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;SAC/D,QAAQ,CAAC,sCAAsC,CAAC;IACnD,UAAU,EAAE,wBAAwB;IACpC,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,CACnD,8GAA8G,CAC/G;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,gBAAgB,CAAC;IAC9C,SAAS,EAAE,kBAAkB,CAAC,QAAQ,CAAC,kBAAkB,CAAC;CAC3D,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,GAAG,EAAE,CAAC,EACpB,sBAAsB,EACtB,IAAI,CACL,CAAC;AAEJ,MAAM,CAAC,MAAM,qCAAqC,GAAmB,CAAC,CAAC,MAAM,CAAC;IAC5E,oBAAoB,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE;IAC9D,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC;IAClD,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC;IAClD,aAAa,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,EAAE;IACnE,UAAU,EAAE,wBAAwB;IACpC,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC;IAC5C,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE;IACnB,SAAS,EAAE,kBAAkB;CAC9B,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,EAAE;IACjB,MAAM,EAAE,oBAAoB,EAAE,GAAG,IAAI,EAAE,GAAG,CAAC,CAAC;IAC5C,OAAO;QACL,GAAG,IAAI;QACP,GAAG,oBAAoB;KACxB,CAAC;AACJ,CAAC,CAAC,CAAC"}
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@dalmia/calibrate-mcp",
3
- "version": "0.0.23",
3
+ "version": "0.0.25",
4
4
  "author": "Speakeasy",
5
5
  "type": "module",
6
6
  "sideEffects": false,
@@ -3,7 +3,7 @@
3
3
  */
4
4
 
5
5
  import { CalibrateMcpCore } from "../core.js";
6
- import { encodeSimple } from "../lib/encodings.js";
6
+ import { encodeFormQuery, encodeSimple } from "../lib/encodings.js";
7
7
  import { compactMap } from "../lib/primitives.js";
8
8
  import { safeParse } from "../lib/schemas.js";
9
9
  import { RequestOptions } from "../lib/sdks.js";
@@ -96,6 +96,10 @@ async function $do(
96
96
  const path$ = pathToFunc("/agent-tests/benchmark/{task_id}")(
97
97
  pathParams$,
98
98
  );
99
+ const query$ = encodeFormQuery({
100
+ "compact": payload$.compact,
101
+ "only_failed": payload$.only_failed,
102
+ });
99
103
 
100
104
  const headers$ = new Headers(compactMap({
101
105
  Accept: "application/json",
@@ -132,6 +136,7 @@ async function $do(
132
136
  baseURL: options?.serverURL,
133
137
  path: path$,
134
138
  headers: headers$,
139
+ query: query$,
135
140
  body: body$,
136
141
  userAgent: client$._options.userAgent,
137
142
  timeoutMs: options?.timeoutMs || client$._options.timeoutMs
@@ -3,7 +3,7 @@
3
3
  */
4
4
 
5
5
  import { CalibrateMcpCore } from "../core.js";
6
- import { encodeSimple } from "../lib/encodings.js";
6
+ import { encodeFormQuery, encodeSimple } from "../lib/encodings.js";
7
7
  import { compactMap } from "../lib/primitives.js";
8
8
  import { safeParse } from "../lib/schemas.js";
9
9
  import { RequestOptions } from "../lib/sdks.js";
@@ -96,6 +96,10 @@ async function $do(
96
96
  const path$ = pathToFunc("/agent-tests/run/{task_id}")(
97
97
  pathParams$,
98
98
  );
99
+ const query$ = encodeFormQuery({
100
+ "compact": payload$.compact,
101
+ "only_failed": payload$.only_failed,
102
+ });
99
103
 
100
104
  const headers$ = new Headers(compactMap({
101
105
  Accept: "application/json",
@@ -132,6 +136,7 @@ async function $do(
132
136
  baseURL: options?.serverURL,
133
137
  path: path$,
134
138
  headers: headers$,
139
+ query: query$,
135
140
  body: body$,
136
141
  userAgent: client$._options.userAgent,
137
142
  timeoutMs: options?.timeoutMs || client$._options.timeoutMs
@@ -97,6 +97,8 @@ async function $do(
97
97
  pathParams$,
98
98
  );
99
99
  const query$ = encodeFormQuery({
100
+ "compact": payload$.compact,
101
+ "disagreement_only": payload$.disagreement_only,
100
102
  "item_id": payload$.item_id,
101
103
  "limit": payload$.limit,
102
104
  "live_only": payload$.live_only,
@@ -3,7 +3,7 @@
3
3
  */
4
4
 
5
5
  import { CalibrateMcpCore } from "../core.js";
6
- import { encodeSimple } from "../lib/encodings.js";
6
+ import { encodeFormQuery, encodeSimple } from "../lib/encodings.js";
7
7
  import { compactMap } from "../lib/primitives.js";
8
8
  import { safeParse } from "../lib/schemas.js";
9
9
  import { RequestOptions } from "../lib/sdks.js";
@@ -96,6 +96,9 @@ async function $do(
96
96
  const path$ = pathToFunc("/evaluators/{evaluator_uuid}")(
97
97
  pathParams$,
98
98
  );
99
+ const query$ = encodeFormQuery({
100
+ "compact": payload$.compact,
101
+ });
99
102
 
100
103
  const headers$ = new Headers(compactMap({
101
104
  Accept: "application/json",
@@ -132,6 +135,7 @@ async function $do(
132
135
  baseURL: options?.serverURL,
133
136
  path: path$,
134
137
  headers: headers$,
138
+ query: query$,
135
139
  body: body$,
136
140
  userAgent: client$._options.userAgent,
137
141
  timeoutMs: options?.timeoutMs || client$._options.timeoutMs
@@ -930,7 +930,7 @@ http_headers = { "api-key-auth" = "YOUR_API_KEY_AUTH" }`;
930
930
  <h1>Instructions</h1>
931
931
  <p>One-click installation for Claude Desktop users</p>
932
932
  <div class="instruction-item">
933
- <a href="https://github.com/dalmia/calibrate-mcp/releases/download/v0.0.23/mcp-server.mcpb" download="mcp-server.mcpb" class="action-button header-action" style="display: inline-flex; margin-bottom: 16px;">
933
+ <a href="https://github.com/dalmia/calibrate-mcp/releases/download/v0.0.25/mcp-server.mcpb" download="mcp-server.mcpb" class="action-button header-action" style="display: inline-flex; margin-bottom: 16px;">
934
934
  📥 Download MCP Bundle
935
935
  </a>
936
936
  </div>
package/src/lib/config.ts CHANGED
@@ -65,8 +65,8 @@ export function serverURLFromOptions(options: SDKOptions): URL | null {
65
65
  export const SDK_METADATA = {
66
66
  language: "typescript",
67
67
  openapiDocVersion: "0.1.0",
68
- sdkVersion: "0.0.23",
68
+ sdkVersion: "0.0.25",
69
69
  genVersion: "2.915.1",
70
70
  userAgent:
71
- "speakeasy-sdk/mcp-typescript 0.0.23 2.915.1 0.1.0 @dalmia/calibrate-mcp",
71
+ "speakeasy-sdk/mcp-typescript 0.0.25 2.915.1 0.1.0 @dalmia/calibrate-mcp",
72
72
  } as const;
@@ -21,7 +21,7 @@ const routes = buildRouteMap({
21
21
  export const app = buildApplication(routes, {
22
22
  name: "mcp",
23
23
  versionInfo: {
24
- currentVersion: "0.0.23",
24
+ currentVersion: "0.0.25",
25
25
  },
26
26
  });
27
27
 
@@ -64,7 +64,7 @@ export function createMCPServer(deps: {
64
64
  }) {
65
65
  const server = new McpServer({
66
66
  name: "CalibrateMcp",
67
- version: "0.0.23",
67
+ version: "0.0.25",
68
68
  });
69
69
 
70
70
  const getClient = deps.getSDK || (() =>
@@ -5,9 +5,9 @@
5
5
  import * as z from "zod";
6
6
  import { ClosedEnum } from "../types/enums.js";
7
7
  import {
8
- EvaluatorVersionResponse,
9
- EvaluatorVersionResponse$zodSchema,
10
- } from "./evaluatorversionresponse.js";
8
+ EvaluatorVersionCompact,
9
+ EvaluatorVersionCompact$zodSchema,
10
+ } from "./evaluatorversioncompact.js";
11
11
 
12
12
  /**
13
13
  * What the evaluator judges:
@@ -20,7 +20,7 @@ import {
20
20
  * - `llm-general`: a standalone input and output pair
21
21
  * - `conversation`: a full conversation
22
22
  */
23
- export const EvaluatorDetailResponseEvaluatorType = {
23
+ export const EvaluatorDetailResponseCompactEvaluatorType = {
24
24
  Tts: "tts",
25
25
  Stt: "stt",
26
26
  Llm: "llm",
@@ -38,11 +38,11 @@ export const EvaluatorDetailResponseEvaluatorType = {
38
38
  * - `llm-general`: a standalone input and output pair
39
39
  * - `conversation`: a full conversation
40
40
  */
41
- export type EvaluatorDetailResponseEvaluatorType = ClosedEnum<
42
- typeof EvaluatorDetailResponseEvaluatorType
41
+ export type EvaluatorDetailResponseCompactEvaluatorType = ClosedEnum<
42
+ typeof EvaluatorDetailResponseCompactEvaluatorType
43
43
  >;
44
44
 
45
- export const EvaluatorDetailResponseEvaluatorType$zodSchema = z.enum([
45
+ export const EvaluatorDetailResponseCompactEvaluatorType$zodSchema = z.enum([
46
46
  "tts",
47
47
  "stt",
48
48
  "llm",
@@ -60,7 +60,7 @@ export const EvaluatorDetailResponseEvaluatorType$zodSchema = z.enum([
60
60
  * - `text`
61
61
  * - `audio`
62
62
  */
63
- export const EvaluatorDetailResponseDataType = {
63
+ export const EvaluatorDetailResponseCompactDataType = {
64
64
  Text: "text",
65
65
  Audio: "audio",
66
66
  } as const;
@@ -72,11 +72,11 @@ export const EvaluatorDetailResponseDataType = {
72
72
  * - `text`
73
73
  * - `audio`
74
74
  */
75
- export type EvaluatorDetailResponseDataType = ClosedEnum<
76
- typeof EvaluatorDetailResponseDataType
75
+ export type EvaluatorDetailResponseCompactDataType = ClosedEnum<
76
+ typeof EvaluatorDetailResponseCompactDataType
77
77
  >;
78
78
 
79
- export const EvaluatorDetailResponseDataType$zodSchema = z.enum([
79
+ export const EvaluatorDetailResponseCompactDataType$zodSchema = z.enum([
80
80
  "text",
81
81
  "audio",
82
82
  ]).describe("The modality the judge reads:\n\n- `text`\n- `audio`\n");
@@ -89,7 +89,7 @@ export const EvaluatorDetailResponseDataType$zodSchema = z.enum([
89
89
  * - `binary`: pass or fail
90
90
  * - `rating`: a numeric score, using the scale in `output_config`
91
91
  */
92
- export const EvaluatorDetailResponseOutputType = {
92
+ export const EvaluatorDetailResponseCompactOutputType = {
93
93
  Binary: "binary",
94
94
  Rating: "rating",
95
95
  } as const;
@@ -101,47 +101,48 @@ export const EvaluatorDetailResponseOutputType = {
101
101
  * - `binary`: pass or fail
102
102
  * - `rating`: a numeric score, using the scale in `output_config`
103
103
  */
104
- export type EvaluatorDetailResponseOutputType = ClosedEnum<
105
- typeof EvaluatorDetailResponseOutputType
104
+ export type EvaluatorDetailResponseCompactOutputType = ClosedEnum<
105
+ typeof EvaluatorDetailResponseCompactOutputType
106
106
  >;
107
107
 
108
- export const EvaluatorDetailResponseOutputType$zodSchema = z.enum([
108
+ export const EvaluatorDetailResponseCompactOutputType$zodSchema = z.enum([
109
109
  "binary",
110
110
  "rating",
111
111
  ]).describe(
112
112
  "How the evaluator scores:\n\n- `binary`: pass or fail\n- `rating`: a numeric score, using the scale in `output_config`\n",
113
113
  );
114
114
 
115
- export type EvaluatorDetailResponse = {
115
+ export type EvaluatorDetailResponseCompact = {
116
116
  uuid: string;
117
117
  name: string;
118
118
  description?: string | null | undefined;
119
- evaluator_type: EvaluatorDetailResponseEvaluatorType;
120
- data_type: EvaluatorDetailResponseDataType;
121
- output_type: EvaluatorDetailResponseOutputType;
119
+ evaluator_type: EvaluatorDetailResponseCompactEvaluatorType;
120
+ data_type: EvaluatorDetailResponseCompactDataType;
121
+ output_type: EvaluatorDetailResponseCompactOutputType;
122
122
  is_default: boolean;
123
123
  slug?: string | null | undefined;
124
124
  live_version_id?: string | null | undefined;
125
125
  created_at: string;
126
126
  updated_at: string;
127
- versions: Array<EvaluatorVersionResponse>;
127
+ versions: Array<EvaluatorVersionCompact>;
128
128
  };
129
129
 
130
- export const EvaluatorDetailResponse$zodSchema: z.ZodType<
131
- EvaluatorDetailResponse
130
+ export const EvaluatorDetailResponseCompact$zodSchema: z.ZodType<
131
+ EvaluatorDetailResponseCompact
132
132
  > = z.object({
133
133
  created_at: z.string().describe(
134
134
  "When the evaluator was created (ISO 8601 UTC)",
135
135
  ),
136
- data_type: EvaluatorDetailResponseDataType$zodSchema.describe(
136
+ data_type: EvaluatorDetailResponseCompactDataType$zodSchema.describe(
137
137
  "The modality the judge reads:\n\n- `text`\n- `audio`\n",
138
138
  ),
139
139
  description: z.string().nullable().optional().describe(
140
140
  "What the evaluator checks",
141
141
  ),
142
- evaluator_type: EvaluatorDetailResponseEvaluatorType$zodSchema.describe(
143
- "What the evaluator judges:\n\n- `tts`: TTS audio\n- `stt`: one transcript\n- `llm`: a reply with its conversation history\n- `llm-general`: a standalone input and output pair\n- `conversation`: a full conversation\n",
144
- ),
142
+ evaluator_type: EvaluatorDetailResponseCompactEvaluatorType$zodSchema
143
+ .describe(
144
+ "What the evaluator judges:\n\n- `tts`: TTS audio\n- `stt`: one transcript\n- `llm`: a reply with its conversation history\n- `llm-general`: a standalone input and output pair\n- `conversation`: a full conversation\n",
145
+ ),
145
146
  is_default: z.boolean().describe(
146
147
  "True for a built-in default evaluator, which you can't edit. False for an evaluator you created, which you can edit and add versions to",
147
148
  ),
@@ -149,7 +150,7 @@ export const EvaluatorDetailResponse$zodSchema: z.ZodType<
149
150
  "ID of the version that is currently live",
150
151
  ),
151
152
  name: z.string().describe("Evaluator name"),
152
- output_type: EvaluatorDetailResponseOutputType$zodSchema.describe(
153
+ output_type: EvaluatorDetailResponseCompactOutputType$zodSchema.describe(
153
154
  "How the evaluator scores:\n\n- `binary`: pass or fail\n- `rating`: a numeric score, using the scale in `output_config`\n",
154
155
  ),
155
156
  slug: z.string().nullable().optional().describe(
@@ -159,7 +160,7 @@ export const EvaluatorDetailResponse$zodSchema: z.ZodType<
159
160
  "When the evaluator was last updated (ISO 8601 UTC)",
160
161
  ),
161
162
  uuid: z.string().describe("Evaluator ID"),
162
- versions: z.array(EvaluatorVersionResponse$zodSchema).describe(
163
+ versions: z.array(EvaluatorVersionCompact$zodSchema).describe(
163
164
  "Full version history, oldest first",
164
165
  ),
165
166
  });
@@ -6,18 +6,18 @@ import * as z from "zod";
6
6
  import { OutputConfig, OutputConfig$zodSchema } from "./outputconfig.js";
7
7
  import { VariableSpec, VariableSpec$zodSchema } from "./variablespec.js";
8
8
 
9
- export type EvaluatorVersionResponse = {
9
+ export type EvaluatorVersionCompact = {
10
10
  uuid: string;
11
11
  version_number: number;
12
12
  judge_model: string;
13
- system_prompt: string;
13
+ system_prompt?: string | null | undefined;
14
14
  output_config?: OutputConfig | null | undefined;
15
15
  variables?: Array<VariableSpec> | null | undefined;
16
16
  created_at: string;
17
17
  };
18
18
 
19
- export const EvaluatorVersionResponse$zodSchema: z.ZodType<
20
- EvaluatorVersionResponse
19
+ export const EvaluatorVersionCompact$zodSchema: z.ZodType<
20
+ EvaluatorVersionCompact
21
21
  > = z.object({
22
22
  created_at: z.string().describe(
23
23
  "When the version was created (ISO 8601 UTC)",
@@ -28,7 +28,7 @@ export const EvaluatorVersionResponse$zodSchema: z.ZodType<
28
28
  output_config: OutputConfig$zodSchema.nullable().optional().describe(
29
29
  "The scale points and their labels. Required for a `rating` evaluator. A `binary` evaluator uses the default Correct/Wrong labels unless you set your own",
30
30
  ),
31
- system_prompt: z.string().describe(
31
+ system_prompt: z.string().nullable().optional().describe(
32
32
  "Judge system prompt, with `{{\"{{\"}}variable}}` placeholders unrendered",
33
33
  ),
34
34
  uuid: z.string().describe("Version ID"),
@@ -14,11 +14,19 @@ import {
14
14
 
15
15
  export type GetAgentTestRunStatusAgentTestsRunTaskIdGetRequest = {
16
16
  task_id: string;
17
+ only_failed?: boolean | undefined;
18
+ compact?: boolean | undefined;
17
19
  xAPIKey?: string | null | undefined;
18
20
  };
19
21
 
20
22
  export const GetAgentTestRunStatusAgentTestsRunTaskIdGetRequest$zodSchema:
21
23
  z.ZodType<GetAgentTestRunStatusAgentTestsRunTaskIdGetRequest> = z.object({
24
+ compact: z.boolean().default(false).describe(
25
+ "Return a compact response that omits heavy detail fields (`results.output`, `results.test_case`, `results.judge_results`, `results.reasoning`, `evaluators.output_config`), keeping only the lightweight decision fields. Omit for full detail",
26
+ ),
27
+ only_failed: z.boolean().default(false).describe(
28
+ "Return only failing test cases. Omit to return every case",
29
+ ),
22
30
  task_id: z.string().describe("Test run to poll for status and results"),
23
31
  xAPIKey: z.string().nullable().optional(),
24
32
  });
@@ -14,11 +14,19 @@ import {
14
14
 
15
15
  export type GetBenchmarkStatusAgentTestsBenchmarkTaskIdGetRequest = {
16
16
  task_id: string;
17
+ only_failed?: boolean | undefined;
18
+ compact?: boolean | undefined;
17
19
  xAPIKey?: string | null | undefined;
18
20
  };
19
21
 
20
22
  export const GetBenchmarkStatusAgentTestsBenchmarkTaskIdGetRequest$zodSchema:
21
23
  z.ZodType<GetBenchmarkStatusAgentTestsBenchmarkTaskIdGetRequest> = z.object({
24
+ compact: z.boolean().default(false).describe(
25
+ "Return a compact response that omits heavy detail fields (`model_results.test_results`, `evaluators.output_config`), keeping only the lightweight decision fields. Omit for full detail",
26
+ ),
27
+ only_failed: z.boolean().default(false).describe(
28
+ "Return only failing test cases for each model. Omit to return every case",
29
+ ),
22
30
  task_id: z.string().describe(
23
31
  "Benchmark run to poll for status and results",
24
32
  ),
@@ -4,9 +4,9 @@
4
4
 
5
5
  import * as z from "zod";
6
6
  import {
7
- EvaluatorDetailResponse,
8
- EvaluatorDetailResponse$zodSchema,
9
- } from "./evaluatordetailresponse.js";
7
+ EvaluatorDetailResponseCompact,
8
+ EvaluatorDetailResponseCompact$zodSchema,
9
+ } from "./evaluatordetailresponsecompact.js";
10
10
  import {
11
11
  HTTPValidationError,
12
12
  HTTPValidationError$zodSchema,
@@ -14,21 +14,25 @@ import {
14
14
 
15
15
  export type GetEvaluatorEndpointEvaluatorsEvaluatorUuidGetRequest = {
16
16
  evaluator_uuid: string;
17
+ compact?: boolean | undefined;
17
18
  xAPIKey?: string | null | undefined;
18
19
  };
19
20
 
20
21
  export const GetEvaluatorEndpointEvaluatorsEvaluatorUuidGetRequest$zodSchema:
21
22
  z.ZodType<GetEvaluatorEndpointEvaluatorsEvaluatorUuidGetRequest> = z.object({
23
+ compact: z.boolean().default(false).describe(
24
+ "Return a compact response that omits heavy detail fields (`versions.system_prompt`, `versions.output_config`, `versions.variables`), keeping only the lightweight decision fields. Omit for full detail",
25
+ ),
22
26
  evaluator_uuid: z.string().describe("Evaluator to retrieve"),
23
27
  xAPIKey: z.string().nullable().optional(),
24
28
  });
25
29
 
26
30
  export type GetEvaluatorEndpointEvaluatorsEvaluatorUuidGetResponse =
27
- | EvaluatorDetailResponse
31
+ | EvaluatorDetailResponseCompact
28
32
  | HTTPValidationError;
29
33
 
30
34
  export const GetEvaluatorEndpointEvaluatorsEvaluatorUuidGetResponse$zodSchema:
31
35
  z.ZodType<GetEvaluatorEndpointEvaluatorsEvaluatorUuidGetResponse> = z.union([
32
- EvaluatorDetailResponse$zodSchema,
36
+ EvaluatorDetailResponseCompact$zodSchema,
33
37
  HTTPValidationError$zodSchema,
34
38
  ]);
@@ -34,16 +34,24 @@ export type TaskSummaryAnnotationTasksTaskUuidSummaryGetRequest = {
34
34
  task_uuid: string;
35
35
  item_id?: string | null | undefined;
36
36
  live_only?: boolean | undefined;
37
+ disagreement_only?: boolean | undefined;
37
38
  q?: string | null | undefined;
38
39
  sort_by?: string | undefined;
39
40
  order?: Order | undefined;
40
41
  limit?: number | undefined;
41
42
  offset?: number | undefined;
43
+ compact?: boolean | undefined;
42
44
  xAPIKey?: string | null | undefined;
43
45
  };
44
46
 
45
47
  export const TaskSummaryAnnotationTasksTaskUuidSummaryGetRequest$zodSchema:
46
48
  z.ZodType<TaskSummaryAnnotationTasksTaskUuidSummaryGetRequest> = z.object({
49
+ compact: z.boolean().default(false).describe(
50
+ "Return a compact response that omits heavy detail fields (`rows.payload`, `rows.evaluator_reasoning`, `rows.annotations.reasoning`, `evaluators.versions.system_prompt`, `evaluators.versions.output_config`, `evaluators.versions.variables`, `item_comments`), keeping only the lightweight decision fields. Omit for full detail",
51
+ ),
52
+ disagreement_only: z.boolean().default(false).describe(
53
+ "When true, keep only rows where the evaluator disagreed with at least one annotator",
54
+ ),
47
55
  item_id: z.string().describe(
48
56
  "Filter rows to a single item. The full task-wide annotator union is still returned in `annotators`",
49
57
  ).nullable().optional(),
@@ -34,7 +34,7 @@ export type TaskSummaryResponse = {
34
34
  evaluators: Array<{ [k: string]: any }>;
35
35
  annotators: Array<{ [k: string]: any }>;
36
36
  rows: Array<{ [k: string]: any }>;
37
- item_comments: { [k: string]: any };
37
+ item_comments?: { [k: string]: any } | null | undefined;
38
38
  pagination: PaginationMeta;
39
39
  AdditionalProperties?: { [k: string]: any };
40
40
  };
@@ -48,9 +48,8 @@ export const TaskSummaryResponse$zodSchema: z.ZodType<TaskSummaryResponse> =
48
48
  evaluators: z.array(z.record(z.string(), z.any())).describe(
49
49
  "The task's evaluators, stable across pages, each with its live rubric",
50
50
  ),
51
- item_comments: z.record(z.string(), z.any()).describe(
52
- "Annotator comments, keyed by item ID",
53
- ),
51
+ item_comments: z.record(z.string(), z.any()).nullable().optional()
52
+ .describe("Annotator comments, keyed by item ID"),
54
53
  pagination: PaginationMeta$zodSchema,
55
54
  rows: z.array(z.record(z.string(), z.any())).describe(
56
55
  "One row per item, evaluator, and version on this page, with the evaluator's value and each annotator's label",
@@ -66,7 +65,7 @@ export const TaskSummaryResponse$zodSchemaOutbound: z.ZodType<any> = z.object({
66
65
  AdditionalProperties: z.record(z.string(), z.any()).optional(),
67
66
  annotators: z.array(z.record(z.string(), z.any())),
68
67
  evaluators: z.array(z.record(z.string(), z.any())),
69
- item_comments: z.record(z.string(), z.any()),
68
+ item_comments: z.nullable(z.record(z.string(), z.any())).optional(),
70
69
  pagination: PaginationMeta$zodSchema,
71
70
  rows: z.array(z.record(z.string(), z.any())),
72
71
  task_id: z.string(),
@@ -1 +0,0 @@
1
- {"version":3,"file":"evaluatordetailresponse.d.ts","sourceRoot":"","sources":["../../src/models/evaluatordetailresponse.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EACL,wBAAwB,EAEzB,MAAM,+BAA+B,CAAC;AAEvC;;;;;;;;;;GAUG;AACH,eAAO,MAAM,oCAAoC;;;;;;CAMvC,CAAC;AACX;;;;;;;;;;GAUG;AACH,MAAM,MAAM,oCAAoC,GAAG,UAAU,CAC3D,OAAO,oCAAoC,CAC5C,CAAC;AAEF,eAAO,MAAM,8CAA8C;;;;;;EAQ1D,CAAC;AAEF;;;;;;;GAOG;AACH,eAAO,MAAM,+BAA+B;;;CAGlC,CAAC;AACX;;;;;;;GAOG;AACH,MAAM,MAAM,+BAA+B,GAAG,UAAU,CACtD,OAAO,+BAA+B,CACvC,CAAC;AAEF,eAAO,MAAM,yCAAyC;;;EAGe,CAAC;AAEtE;;;;;;;GAOG;AACH,eAAO,MAAM,iCAAiC;;;CAGpC,CAAC;AACX;;;;;;;GAOG;AACH,MAAM,MAAM,iCAAiC,GAAG,UAAU,CACxD,OAAO,iCAAiC,CACzC,CAAC;AAEF,eAAO,MAAM,2CAA2C;;;EAKvD,CAAC;AAEF,MAAM,MAAM,uBAAuB,GAAG;IACpC,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACxC,cAAc,EAAE,oCAAoC,CAAC;IACrD,SAAS,EAAE,+BAA+B,CAAC;IAC3C,WAAW,EAAE,iCAAiC,CAAC;IAC/C,UAAU,EAAE,OAAO,CAAC;IACpB,IAAI,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACjC,eAAe,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IAC5C,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,KAAK,CAAC,wBAAwB,CAAC,CAAC;CAC3C,CAAC;AAEF,eAAO,MAAM,iCAAiC,EAAE,CAAC,CAAC,OAAO,CACvD,uBAAuB,CAkCvB,CAAC"}
@@ -1 +0,0 @@
1
- {"version":3,"file":"evaluatordetailresponse.js","sourceRoot":"","sources":["../../src/models/evaluatordetailresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,OAAO,EAEL,kCAAkC,GACnC,MAAM,+BAA+B,CAAC;AAEvC;;;;;;;;;;GAUG;AACH,MAAM,CAAC,MAAM,oCAAoC,GAAG;IAClD,GAAG,EAAE,KAAK;IACV,GAAG,EAAE,KAAK;IACV,GAAG,EAAE,KAAK;IACV,UAAU,EAAE,aAAa;IACzB,YAAY,EAAE,cAAc;CACpB,CAAC;AAgBX,MAAM,CAAC,MAAM,8CAA8C,GAAG,CAAC,CAAC,IAAI,CAAC;IACnE,KAAK;IACL,KAAK;IACL,KAAK;IACL,aAAa;IACb,cAAc;CACf,CAAC,CAAC,QAAQ,CACT,yNAAyN,CAC1N,CAAC;AAEF;;;;;;;GAOG;AACH,MAAM,CAAC,MAAM,+BAA+B,GAAG;IAC7C,IAAI,EAAE,MAAM;IACZ,KAAK,EAAE,OAAO;CACN,CAAC;AAaX,MAAM,CAAC,MAAM,yCAAyC,GAAG,CAAC,CAAC,IAAI,CAAC;IAC9D,MAAM;IACN,OAAO;CACR,CAAC,CAAC,QAAQ,CAAC,wDAAwD,CAAC,CAAC;AAEtE;;;;;;;GAOG;AACH,MAAM,CAAC,MAAM,iCAAiC,GAAG;IAC/C,MAAM,EAAE,QAAQ;IAChB,MAAM,EAAE,QAAQ;CACR,CAAC;AAaX,MAAM,CAAC,MAAM,2CAA2C,GAAG,CAAC,CAAC,IAAI,CAAC;IAChE,QAAQ;IACR,QAAQ;CACT,CAAC,CAAC,QAAQ,CACT,0HAA0H,CAC3H,CAAC;AAiBF,MAAM,CAAC,MAAM,iCAAiC,GAE1C,CAAC,CAAC,MAAM,CAAC;IACX,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAC7B,+CAA+C,CAChD;IACD,SAAS,EAAE,yCAAyC,CAAC,QAAQ,CAC3D,wDAAwD,CACzD;IACD,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACpD,2BAA2B,CAC5B;IACD,cAAc,EAAE,8CAA8C,CAAC,QAAQ,CACrE,yNAAyN,CAC1N;IACD,UAAU,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,CAC9B,yIAAyI,CAC1I;IACD,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACxD,0CAA0C,CAC3C;IACD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,gBAAgB,CAAC;IAC3C,WAAW,EAAE,2CAA2C,CAAC,QAAQ,CAC/D,0HAA0H,CAC3H;IACD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC7C,8CAA8C,CAC/C;IACD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAC7B,oDAAoD,CACrD;IACD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,cAAc,CAAC;IACzC,QAAQ,EAAE,CAAC,CAAC,KAAK,CAAC,kCAAkC,CAAC,CAAC,QAAQ,CAC5D,oCAAoC,CACrC;CACF,CAAC,CAAC"}
@@ -1 +0,0 @@
1
- {"version":3,"file":"evaluatorversionresponse.d.ts","sourceRoot":"","sources":["../../src/models/evaluatorversionresponse.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,YAAY,EAA0B,MAAM,mBAAmB,CAAC;AACzE,OAAO,EAAE,YAAY,EAA0B,MAAM,mBAAmB,CAAC;AAEzE,MAAM,MAAM,wBAAwB,GAAG;IACrC,IAAI,EAAE,MAAM,CAAC;IACb,cAAc,EAAE,MAAM,CAAC;IACvB,WAAW,EAAE,MAAM,CAAC;IACpB,aAAa,EAAE,MAAM,CAAC;IACtB,aAAa,CAAC,EAAE,YAAY,GAAG,IAAI,GAAG,SAAS,CAAC;IAChD,SAAS,CAAC,EAAE,KAAK,CAAC,YAAY,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACnD,UAAU,EAAE,MAAM,CAAC;CACpB,CAAC;AAEF,eAAO,MAAM,kCAAkC,EAAE,CAAC,CAAC,OAAO,CACxD,wBAAwB,CAqBxB,CAAC"}
@@ -1 +0,0 @@
1
- {"version":3,"file":"evaluatorversionresponse.js","sourceRoot":"","sources":["../../src/models/evaluatorversionresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAgB,sBAAsB,EAAE,MAAM,mBAAmB,CAAC;AACzE,OAAO,EAAgB,sBAAsB,EAAE,MAAM,mBAAmB,CAAC;AAYzE,MAAM,CAAC,MAAM,kCAAkC,GAE3C,CAAC,CAAC,MAAM,CAAC;IACX,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAC7B,6CAA6C,CAC9C;IACD,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAC9B,6HAA6H,CAC9H;IACD,aAAa,EAAE,sBAAsB,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAClE,0JAA0J,CAC3J;IACD,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAChC,0EAA0E,CAC3E;IACD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,YAAY,CAAC;IACvC,SAAS,EAAE,CAAC,CAAC,KAAK,CAAC,sBAAsB,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACvE,2BAA2B,CAC5B;IACD,cAAc,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,QAAQ,CAC9B,2GAA2G,CAC5G;CACF,CAAC,CAAC"}