ML Referee
Evaluate ML features beyond a baseline and test bounded feature repairs. Invite-only beta.
사용해야 할까요
품질 및 안전성
A
발견 사항 (2)
- LOWTool 'upload_campaign_shard' description lacks action verbupload_campaign_shard에서
- LOWTool 'upload_repair_matrix' description lacks action verbupload_repair_matrix에서
도구 정의와 프로토콜 준수에 대한 자동 분석을 기반으로 합니다.
컨텍스트 비용
~8,555토큰 (도구 정의)
~3.3 KB일반적인 응답 크기
상당한 주의 영향 (128k 컨텍스트의 6.68%)
이는 서버의 도구가 모델의 컨텍스트에 로드될 때마다 소비되는 대략적인 토큰 수입니다. 수치가 높을수록 다른 작업에 사용할 수 있는 주의가 줄어듭니다.
설치
원클릭 설치
`claude_desktop_config.json` 파일에 다음을 추가하세요:
{
"mcpServers": {
"mlreferee": {
"url": "https://api.mlreferee.com/mcp"
}
}
}원격 엔드포인트
https://api.mlreferee.com/mcpstreamable-http할 수 있는 일
도구 목록
도구 (26)
🟢 읽기 전용🟡 쓰기🔴 삭제⚪ 알 수 없음
🟢describe_referee
Discover capabilities, privacy, and the approved public catalog.
입력 스키마
{
"type": "object",
"properties": {},
"additionalProperties": false,
"title": "Empty"
}🟢list_verticals
List the three supported evaluation profiles and their discovery tags.
입력 스키마
{
"type": "object",
"properties": {},
"additionalProperties": false,
"title": "Empty"
}🟢get_referee_readiness
Report connection, credential, audit-credit, and submission readiness with the next action.
입력 스키마
{
"type": "object",
"properties": {},
"additionalProperties": false,
"title": "Empty"
}🟢get_referee_usage(window_days)
Read aggregate, privacy-preserving Referee activity and outcome evidence for the selected audience.
입력 스키마
{
"type": "object",
"properties": {
"window_days": {
"default": 90,
"maximum": 365,
"minimum": 1,
"title": "Window Days",
"type": "integer"
}
},
"additionalProperties": false,
"title": "UsageWindow"
}🟢get_input_schema(vertical)
Get neutral column roles and validation defaults for a vertical.
입력 스키마
{
"type": "object",
"properties": {
"vertical": {
"pattern": "^(general_ml|demand_forecasting|bioinformatics_research)$",
"title": "Vertical",
"type": "string"
}
},
"required": [
"vertical"
],
"additionalProperties": false,
"title": "Vertical"
}🟢get_repair_schema
Get the feature-only contract for diagnosing degradation and testing bounded repairs.
입력 스키마
{
"type": "object",
"properties": {},
"additionalProperties": false,
"title": "Empty"
}🟡upload_private_dataset(data, format, retention_seconds)
Encrypt a de-identified CSV/JSON payload for a short tenant-private retention window.
입력 스키마
{
"type": "object",
"properties": {
"data": {
"maxLength": 5000000,
"minLength": 1,
"title": "Data",
"type": "string"
},
"format": {
"default": "csv",
"pattern": "^(csv|json)$",
"title": "Format",
"type": "string"
},
"retention_seconds": {
"default": 86400,
"maximum": 86400,
"minimum": 60,
"title": "Retention Seconds",
"type": "integer"
}
},
"required": [
"data"
],
"additionalProperties": false,
"title": "Upload"
}🟡upload_campaign_shard(data, format, retention_seconds)
Encrypt one of up to 64 private feature shards for a large campaign; retain for at least 4,200 seconds.
입력 스키마
{
"type": "object",
"properties": {
"data": {
"maxLength": 5000000,
"minLength": 1,
"title": "Data",
"type": "string"
},
"format": {
"default": "csv",
"pattern": "^(csv|json)$",
"title": "Format",
"type": "string"
},
"retention_seconds": {
"default": 86400,
"maximum": 86400,
"minimum": 60,
"title": "Retention Seconds",
"type": "integer"
}
},
"required": [
"data"
],
"additionalProperties": false,
"title": "Upload"
}🟡upload_repair_matrix(data, format, retention_seconds)
Encrypt one complete de-identified repair matrix of up to 100 MB and 1,000 diagnostic features.
입력 스키마
{
"type": "object",
"properties": {
"data": {
"maxLength": 100000000,
"minLength": 1,
"title": "Data",
"type": "string"
},
"format": {
"default": "csv",
"pattern": "^(csv|json)$",
"title": "Format",
"type": "string"
},
"retention_seconds": {
"default": 7200,
"maximum": 86400,
"minimum": 4200,
"title": "Retention Seconds",
"type": "integer"
}
},
"required": [
"data"
],
"additionalProperties": false,
"title": "RepairUpload"
}🟡preflight_dataset(upload_id, vertical, contract)
Validate a tenant-private upload and its baseline evidence/row identity without spending a credit.
입력 스키마
{
"type": "object",
"properties": {
"upload_id": {
"pattern": "^[a-f0-9]{32}$",
"title": "Upload Id",
"type": "string"
},
"vertical": {
"enum": [
"general_ml",
"demand_forecasting",
"bioinformatics_research"
],
"title": "Vertical",
"type": "string"
},
"contract": {
"$ref": "#/$defs/FeatureAuditContract"
}
},
"required": [
"upload_id",
"vertical",
"contract"
],
"$defs": {
"FeatureAuditContract": {
"additionalProperties": false,
"description": "Directory-facing contract for feature-quality evidence.",
"properties": {
"task": {
"default": "regression",
"enum": [
"regression",
"binary_classification",
"time_series_ranking"
],
"title": "Task",
"type": "string"
},
"observation_time": {
"default": "observation_time",
"title": "Observation Time",
"type": "string"
},
"label_available_time": {
"default": "label_available_time",
"title": "Label Available Time",
"type": "string"
},
"entity": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Entity"
},
"outcome": {
"default": "outcome",
"title": "Outcome",
"type": "string"
},
"baseline": {
"items": {
"type": "string"
},
"maxItems": 32,
"title": "Baseline",
"type": "array"
},
"baseline_spec": {
"anyOf": [
{
"discriminator": {
"mapping": {
"generic_refit": "#/$defs/GenericRefitBaseline",
"provided_oof": "#/$defs/FeatureProvidedOOFBaseline"
},
"propertyName": "mode"
},
"oneOf": [
{
"$ref": "#/$defs/GenericRefitBaseline"
},
{
"$ref": "#/$defs/FeatureProvidedOOFBaseline"
}
]
},
{
"type": "null"
}
],
"default": null,
"title": "Baseline Spec"
},
"candidates": {
"items": {
"type": "string"
},
"maxItems": 32,
"minItems": 1,
"title": "Candidates",
"type": "array"
},
"sample_weight": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Sample Weight"
},
"batch": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Batch"
},
"representation_analysis": {
"default": false,
"description": "Opt-in alternative representation comparison and numerical fitting certificates. Binary classification with numeric generic-refit inputs only; requires baseline features. Descriptive, no audit decision changes. Single audits only; limited to 10,000 rows and 250,000 input feature cells.",
"title": "Representation Analysis",
"type": "boolean"
},
"utility": {
"default": "default",
"enum": [
"default",
"absolute_error",
"squared_error"
],
"title": "Utility",
"type": "string"
},
"split": {
"default": "expanding",
"enum": [
"expanding",
"rolling",
"grouped",
"purged"
],
"title": "Split",
"type": "string"
},
"folds": {
"default": 5,
"maximum": 8,
"minimum": 3,
"title": "Folds",
"type": "integer"
},
"embargo_seconds": {
"default": 0,
"maximum": 31536000,
"minimum": 0,
"title": "Embargo Seconds",
"type": "integer"
},
"train_window": {
"default": 100,
"maximum": 100000,
"minimum": 20,
"title": "Train Window",
"type": "integer"
},
"fdr_target": {
"default": 0.05,
"description": "Nominal false-discovery-rate target for the complete submitted family. Expected false fraction, not an individual false-finding probability. Set before evaluation.",
"exclusiveMaximum": 1,
"exclusiveMinimum": 0,
"title": "Fdr Target",
"type": "number"
},
"provisional_error_target": {
"default": 0.1,
"description": "Nominal family-wise error target for provisional evidence. Chance of at least one false rejection, conditional on valid tests. Legacy input name provisional_significance_level is also accepted. Set before evaluation.",
"exclusiveMaximum": 1,
"exclusiveMinimum": 0,
"title": "Provisional Error Target",
"type": "number"
},
"selection_error_target": {
"default": 0.05,
"description": "Nominal family-wise error target for the optional selection diagnostic over strict-scope candidates. Reported only; never changes decisions. Set before evaluation.",
"exclusiveMaximum": 1,
"exclusiveMinimum": 0,
"title": "Selection Error Target",
"type": "number"
},
"desired_power": {
"default": 0.8,
"exclusiveMaximum": 1,
"minimum": 0.5,
"title": "Desired Power",
"type": "number"
},
"target_effect": {
"anyOf": [
{
"exclusiveMinimum": 0,
"type": "number"
},
{
"type": "null"
}
],
"default": null,
"title": "Target Effect"
},
"research_only": {
"const": true,
"default": true,
"title": "Research Only",
"type": "boolean"
},
"deidentified": {
"default": true,
"title": "Deidentified",
"type": "boolean"
},
"scope": {
"const": "feature_quality_only",
"default": "feature_quality_only",
"title": "Scope",
"type": "string"
},
"provisional_significance_level": {
"deprecated": true,
"description": "Deprecated input alias for provisional_error_target; conflicting values are rejected.",
"exclusiveMaximum": 1,
"exclusiveMinimum": 0,
"title": "Provisional Significance Level",
"type": "number"
}
},
"title": "FeatureAuditContract",
"type": "object"
},
"FeatureProvidedOOFBaseline": {
"additionalProperties": false,
"description": "Public score evidence without eligibility, selections, or action policy.",
"properties": {
"mode": {
"const": "provided_oof",
"title": "Mode",
"type": "string"
},
"score": {
"maxLength": 256,
"minLength": 1,
"title": "Score",
"type": "string"
},
"score_higher_is_better": {
"default": true,
"title": "Score Higher Is Better",
"type": "boolean"
},
"evaluation_group": {
"items": {
"type": "string"
},
"maxItems": 8,
"title": "Evaluation Group",
"type": "array"
},
"row_id": {
"items": {
"type": "string"
},
"maxItems": 8,
"title": "Row Id",
"type": "array"
},
"candidate_scores": {
"additionalProperties": {
"type": "string"
},
"maxProperties": 32,
"title": "Candidate Scores",
"type": "object"
},
"model_version": {
"anyOf": [
{
"maxLength": 128,
"minLength": 1,
"pattern": "^[A-Za-z0-9][A-Za-z0-9._-]*$",
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Model Version"
},
"verification_digest": {
"pattern": "^[a-f0-9]{64}$",
"title": "Verification Digest",
"type": "string"
}
},
"required": [
"mode",
"score",
"verification_digest"
],
"title": "FeatureProvidedOOFBaseline",
"type": "object"
},
"GenericRefitBaseline": {
"additionalProperties": false,
"properties": {
"mode": {
"const": "generic_refit",
"default": "generic_refit",
"title": "Mode",
"type": "string"
},
"features": {
"items": {
"type": "string"
},
"maxItems": 32,
"minItems": 0,
"title": "Features",
"type": "array"
}
},
"required": [
"features"
],
"title": "GenericRefitBaseline",
"type": "object"
}
},
"additionalProperties": false,
"title": "InputRequest"
}⚪preflight_repair(upload_id, contract)
Validate a tenant-private feature-repair matrix without spending a credit.
입력 스키마
{
"type": "object",
"properties": {
"upload_id": {
"pattern": "^[a-f0-9]{32}$",
"title": "Upload Id",
"type": "string"
},
"contract": {
"$ref": "#/$defs/FeatureRepairContract"
}
},
"required": [
"upload_id",
"contract"
],
"$defs": {
"FeatureRepairColumns": {
"additionalProperties": false,
"description": "Column roles for repair based only on feature and label quality.",
"properties": {
"observation_time": {
"default": "observation_time",
"title": "Observation Time",
"type": "string"
},
"label_available_time": {
"default": "label_available_time",
"title": "Label Available Time",
"type": "string"
},
"outcome": {
"default": "outcome",
"title": "Outcome",
"type": "string"
},
"entity": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Entity"
},
"sample_weight": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Sample Weight"
}
},
"title": "FeatureRepairColumns",
"type": "object"
},
"FeatureRepairContract": {
"additionalProperties": false,
"description": "Directory-facing contract for label-based feature repair.",
"properties": {
"mode": {
"default": "known_break",
"enum": [
"known_break",
"rolling_monitor"
],
"title": "Mode",
"type": "string"
},
"task": {
"default": "regression",
"enum": [
"regression",
"binary_classification",
"time_series_ranking"
],
"title": "Task",
"type": "string"
},
"columns": {
"$ref": "#/$defs/FeatureRepairColumns"
},
"baseline_features": {
"items": {
"type": "string"
},
"maxItems": 32,
"title": "Baseline Features",
"type": "array"
},
"diagnostic_features": {
"items": {
"type": "string"
},
"maxItems": 1000,
"minItems": 1,
"title": "Diagnostic Features",
"type": "array"
},
"reference_period": {
"anyOf": [
{
"$ref": "#/$defs/RepairPeriod"
},
{
"type": "null"
}
],
"default": null
},
"comparison_period": {
"anyOf": [
{
"$ref": "#/$defs/RepairPeriod"
},
{
"type": "null"
}
],
"default": null
},
"rolling": {
"anyOf": [
{
"$ref": "#/$defs/FeatureRepairRollingWindow"
},
{
"type": "null"
}
],
"default": null
},
"asof": {
"anyOf": [
{
"format": "date",
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Asof"
},
"minimum_rows": {
"default": 40,
"maximum": 100000,
"minimum": 20,
"title": "Minimum Rows",
"type": "integer"
},
"minimum_improvement": {
"default": 0.01,
"maximum": 1,
"minimum": 0,
"title": "Minimum Improvement",
"type": "number"
},
"correlation_threshold": {
"default": 0.85,
"exclusiveMaximum": 1,
"minimum": 0.5,
"title": "Correlation Threshold",
"type": "number"
},
"seed": {
"default": 1729,
"maximum": 2147483647,
"minimum": 0,
"title": "Seed",
"type": "integer"
},
"research_only": {
"const": true,
"default": true,
"title": "Research Only",
"type": "boolean"
},
"scope": {
"const": "feature_quality_repair_only",
"default": "feature_quality_repair_only",
"title": "Scope",
"type": "string"
}
},
"required": [
"diagnostic_features"
],
"title": "FeatureRepairContract",
"type": "object"
},
"FeatureRepairRollingWindow": {
"additionalProperties": false,
"properties": {
"reference_days": {
"default": 120,
"maximum": 1000,
"minimum": 40,
"title": "Reference Days",
"type": "integer"
},
"comparison_days": {
"default": 40,
"maximum": 500,
"minimum": 20,
"title": "Comparison Days",
"type": "integer"
}
},
"title": "FeatureRepairRollingWindow",
"type": "object"
},
"RepairPeriod": {
"additionalProperties": false,
"properties": {
"start": {
"format": "date",
"title": "Start",
"type": "string"
},
"end_exclusive": {
"format": "date",
"title": "End Exclusive",
"type": "string"
}
},
"required": [
"start",
"end_exclusive"
],
"title": "RepairPeriod",
"type": "object"
}
},
"additionalProperties": false,
"title": "RepairInputRequest"
}🟡quote_audit(upload_id, vertical, contract, idempotency_key)
Create a contract-bound quote for one audit credit; baseline mode and evidence cannot be substituted.
입력 스키마
{
"type": "object",
"properties": {
"upload_id": {
"pattern": "^[a-f0-9]{32}$",
"title": "Upload Id",
"type": "string"
},
"vertical": {
"enum": [
"general_ml",
"demand_forecasting",
"bioinformatics_research"
],
"title": "Vertical",
"type": "string"
},
"contract": {
"$ref": "#/$defs/FeatureAuditContract"
},
"idempotency_key": {
"maxLength": 128,
"minLength": 8,
"pattern": "^[A-Za-z0-9_-]+$",
"title": "Idempotency Key",
"type": "string"
}
},
"required": [
"upload_id",
"vertical",
"contract",
"idempotency_key"
],
"$defs": {
"FeatureAuditContract": {
"additionalProperties": false,
"description": "Directory-facing contract for feature-quality evidence.",
"properties": {
"task": {
"default": "regression",
"enum": [
"regression",
"binary_classification",
"time_series_ranking"
],
"title": "Task",
"type": "string"
},
"observation_time": {
"default": "observation_time",
"title": "Observation Time",
"type": "string"
},
"label_available_time": {
"default": "label_available_time",
"title": "Label Available Time",
"type": "string"
},
"entity": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Entity"
},
"outcome": {
"default": "outcome",
"title": "Outcome",
"type": "string"
},
"baseline": {
"items": {
"type": "string"
},
"maxItems": 32,
"title": "Baseline",
"type": "array"
},
"baseline_spec": {
"anyOf": [
{
"discriminator": {
"mapping": {
"generic_refit": "#/$defs/GenericRefitBaseline",
"provided_oof": "#/$defs/FeatureProvidedOOFBaseline"
},
"propertyName": "mode"
},
"oneOf": [
{
"$ref": "#/$defs/GenericRefitBaseline"
},
{
"$ref": "#/$defs/FeatureProvidedOOFBaseline"
}
]
},
{
"type": "null"
}
],
"default": null,
"title": "Baseline Spec"
},
"candidates": {
"items": {
"type": "string"
},
"maxItems": 32,
"minItems": 1,
"title": "Candidates",
"type": "array"
},
"sample_weight": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Sample Weight"
},
"batch": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Batch"
},
"representation_analysis": {
"default": false,
"description": "Opt-in alternative representation comparison and numerical fitting certificates. Binary classification with numeric generic-refit inputs only; requires baseline features. Descriptive, no audit decision changes. Single audits only; limited to 10,000 rows and 250,000 input feature cells.",
"title": "Representation Analysis",
"type": "boolean"
},
"utility": {
"default": "default",
"enum": [
"default",
"absolute_error",
"squared_error"
],
"title": "Utility",
"type": "string"
},
"split": {
"default": "expanding",
"enum": [
"expanding",
"rolling",
"grouped",
"purged"
],
"title": "Split",
"type": "string"
},
"folds": {
"default": 5,
"maximum": 8,
"minimum": 3,
"title": "Folds",
"type": "integer"
},
"embargo_seconds": {
"default": 0,
"maximum": 31536000,
"minimum": 0,
"title": "Embargo Seconds",
"type": "integer"
},
"train_window": {
"default": 100,
"maximum": 100000,
"minimum": 20,
"title": "Train Window",
"type": "integer"
},
"fdr_target": {
"default": 0.05,
"description": "Nominal false-discovery-rate target for the complete submitted family. Expected false fraction, not an individual false-finding probability. Set before evaluation.",
"exclusiveMaximum": 1,
"exclusiveMinimum": 0,
"title": "Fdr Target",
"type": "number"
},
"provisional_error_target": {
"default": 0.1,
"description": "Nominal family-wise error target for provisional evidence. Chance of at least one false rejection, conditional on valid tests. Legacy input name provisional_significance_level is also accepted. Set before evaluation.",
"exclusiveMaximum": 1,
"exclusiveMinimum": 0,
"title": "Provisional Error Target",
"type": "number"
},
"selection_error_target": {
"default": 0.05,
"description": "Nominal family-wise error target for the optional selection diagnostic over strict-scope candidates. Reported only; never changes decisions. Set before evaluation.",
"exclusiveMaximum": 1,
"exclusiveMinimum": 0,
"title": "Selection Error Target",
"type": "number"
},
"desired_power": {
"default": 0.8,
"exclusiveMaximum": 1,
"minimum": 0.5,
"title": "Desired Power",
"type": "number"
},
"target_effect": {
"anyOf": [
{
"exclusiveMinimum": 0,
"type": "number"
},
{
"type": "null"
}
],
"default": null,
"title": "Target Effect"
},
"research_only": {
"const": true,
"default": true,
"title": "Research Only",
"type": "boolean"
},
"deidentified": {
"default": true,
"title": "Deidentified",
"type": "boolean"
},
"scope": {
"const": "feature_quality_only",
"default": "feature_quality_only",
"title": "Scope",
"type": "string"
},
"provisional_significance_level": {
"deprecated": true,
"description": "Deprecated input alias for provisional_error_target; conflicting values are rejected.",
"exclusiveMaximum": 1,
"exclusiveMinimum": 0,
"title": "Provisional Significance Level",
"type": "number"
}
},
"title": "FeatureAuditContract",
"type": "object"
},
"FeatureProvidedOOFBaseline": {
"additionalProperties": false,
"description": "Public score evidence without eligibility, selections, or action policy.",
"properties": {
"mode": {
"const": "provided_oof",
"title": "Mode",
"type": "string"
},
"score": {
"maxLength": 256,
"minLength": 1,
"title": "Score",
"type": "string"
},
"score_higher_is_better": {
"default": true,
"title": "Score Higher Is Better",
"type": "boolean"
},
"evaluation_group": {
"items": {
"type": "string"
},
"maxItems": 8,
"title": "Evaluation Group",
"type": "array"
},
"row_id": {
"items": {
"type": "string"
},
"maxItems": 8,
"title": "Row Id",
"type": "array"
},
"candidate_scores": {
"additionalProperties": {
"type": "string"
},
"maxProperties": 32,
"title": "Candidate Scores",
"type": "object"
},
"model_version": {
"anyOf": [
{
"maxLength": 128,
"minLength": 1,
"pattern": "^[A-Za-z0-9][A-Za-z0-9._-]*$",
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Model Version"
},
"verification_digest": {
"pattern": "^[a-f0-9]{64}$",
"title": "Verification Digest",
"type": "string"
}
},
"required": [
"mode",
"score",
"verification_digest"
],
"title": "FeatureProvidedOOFBaseline",
"type": "object"
},
"GenericRefitBaseline": {
"additionalProperties": false,
"properties": {
"mode": {
"const": "generic_refit",
"default": "generic_refit",
"title": "Mode",
"type": "string"
},
"features": {
"items": {
"type": "string"
},
"maxItems": 32,
"minItems": 0,
"title": "Features",
"type": "array"
}
},
"required": [
"features"
],
"title": "GenericRefitBaseline",
"type": "object"
}
},
"additionalProperties": false,
"title": "QuoteRequest"
}⚪quote_repair(upload_id, contract, idempotency_key)
Preflight and quote label-based feature repair; large jobs return a measured ETA and credit requirement.
입력 스키마
{
"type": "object",
"properties": {
"upload_id": {
"pattern": "^[a-f0-9]{32}$",
"title": "Upload Id",
"type": "string"
},
"contract": {
"$ref": "#/$defs/FeatureRepairContract"
},
"idempotency_key": {
"maxLength": 128,
"minLength": 8,
"pattern": "^[A-Za-z0-9_-]+$",
"title": "Idempotency Key",
"type": "string"
}
},
"required": [
"upload_id",
"contract",
"idempotency_key"
],
"$defs": {
"FeatureRepairColumns": {
"additionalProperties": false,
"description": "Column roles for repair based only on feature and label quality.",
"properties": {
"observation_time": {
"default": "observation_time",
"title": "Observation Time",
"type": "string"
},
"label_available_time": {
"default": "label_available_time",
"title": "Label Available Time",
"type": "string"
},
"outcome": {
"default": "outcome",
"title": "Outcome",
"type": "string"
},
"entity": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Entity"
},
"sample_weight": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Sample Weight"
}
},
"title": "FeatureRepairColumns",
"type": "object"
},
"FeatureRepairContract": {
"additionalProperties": false,
"description": "Directory-facing contract for label-based feature repair.",
"properties": {
"mode": {
"default": "known_break",
"enum": [
"known_break",
"rolling_monitor"
],
"title": "Mode",
"type": "string"
},
"task": {
"default": "regression",
"enum": [
"regression",
"binary_classification",
"time_series_ranking"
],
"title": "Task",
"type": "string"
},
"columns": {
"$ref": "#/$defs/FeatureRepairColumns"
},
"baseline_features": {
"items": {
"type": "string"
},
"maxItems": 32,
"title": "Baseline Features",
"type": "array"
},
"diagnostic_features": {
"items": {
"type": "string"
},
"maxItems": 1000,
"minItems": 1,
"title": "Diagnostic Features",
"type": "array"
},
"reference_period": {
"anyOf": [
{
"$ref": "#/$defs/RepairPeriod"
},
{
"type": "null"
}
],
"default": null
},
"comparison_period": {
"anyOf": [
{
"$ref": "#/$defs/RepairPeriod"
},
{
"type": "null"
}
],
"default": null
},
"rolling": {
"anyOf": [
{
"$ref": "#/$defs/FeatureRepairRollingWindow"
},
{
"type": "null"
}
],
"default": null
},
"asof": {
"anyOf": [
{
"format": "date",
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Asof"
},
"minimum_rows": {
"default": 40,
"maximum": 100000,
"minimum": 20,
"title": "Minimum Rows",
"type": "integer"
},
"minimum_improvement": {
"default": 0.01,
"maximum": 1,
"minimum": 0,
"title": "Minimum Improvement",
"type": "number"
},
"correlation_threshold": {
"default": 0.85,
"exclusiveMaximum": 1,
"minimum": 0.5,
"title": "Correlation Threshold",
"type": "number"
},
"seed": {
"default": 1729,
"maximum": 2147483647,
"minimum": 0,
"title": "Seed",
"type": "integer"
},
"research_only": {
"const": true,
"default": true,
"title": "Research Only",
"type": "boolean"
},
"scope": {
"const": "feature_quality_repair_only",
"default": "feature_quality_repair_only",
"title": "Scope",
"type": "string"
}
},
"required": [
"diagnostic_features"
],
"title": "FeatureRepairContract",
"type": "object"
},
"FeatureRepairRollingWindow": {
"additionalProperties": false,
"properties": {
"reference_days": {
"default": 120,
"maximum": 1000,
"minimum": 40,
"title": "Reference Days",
"type": "integer"
},
"comparison_days": {
"default": 40,
"maximum": 500,
"minimum": 20,
"title": "Comparison Days",
"type": "integer"
}
},
"title": "FeatureRepairRollingWindow",
"type": "object"
},
"RepairPeriod": {
"additionalProperties": false,
"properties": {
"start": {
"format": "date",
"title": "Start",
"type": "string"
},
"end_exclusive": {
"format": "date",
"title": "End Exclusive",
"type": "string"
}
},
"required": [
"start",
"end_exclusive"
],
"title": "RepairPeriod",
"type": "object"
}
},
"additionalProperties": false,
"title": "RepairQuoteRequest"
}🟢quote_campaign(vertical, shards, idempotency_key)
Preflight all private feature shards, time a pilot, and quote credits and ETA for up to 1,000 candidates in one corrected search family.
입력 스키마
{
"type": "object",
"properties": {
"vertical": {
"pattern": "^(general_ml|demand_forecasting|bioinformatics_research)$",
"title": "Vertical",
"type": "string"
},
"shards": {
"items": {
"$ref": "#/$defs/FeatureCampaignShard"
},
"maxItems": 64,
"minItems": 2,
"title": "Shards",
"type": "array"
},
"idempotency_key": {
"maxLength": 128,
"minLength": 8,
"pattern": "^[A-Za-z0-9_-]+$",
"title": "Idempotency Key",
"type": "string"
}
},
"required": [
"vertical",
"shards",
"idempotency_key"
],
"$defs": {
"FeatureAuditContract": {
"additionalProperties": false,
"description": "Directory-facing contract for feature-quality evidence.",
"properties": {
"task": {
"default": "regression",
"enum": [
"regression",
"binary_classification",
"time_series_ranking"
],
"title": "Task",
"type": "string"
},
"observation_time": {
"default": "observation_time",
"title": "Observation Time",
"type": "string"
},
"label_available_time": {
"default": "label_available_time",
"title": "Label Available Time",
"type": "string"
},
"entity": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Entity"
},
"outcome": {
"default": "outcome",
"title": "Outcome",
"type": "string"
},
"baseline": {
"items": {
"type": "string"
},
"maxItems": 32,
"title": "Baseline",
"type": "array"
},
"baseline_spec": {
"anyOf": [
{
"discriminator": {
"mapping": {
"generic_refit": "#/$defs/GenericRefitBaseline",
"provided_oof": "#/$defs/FeatureProvidedOOFBaseline"
},
"propertyName": "mode"
},
"oneOf": [
{
"$ref": "#/$defs/GenericRefitBaseline"
},
{
"$ref": "#/$defs/FeatureProvidedOOFBaseline"
}
]
},
{
"type": "null"
}
],
"default": null,
"title": "Baseline Spec"
},
"candidates": {
"items": {
"type": "string"
},
"maxItems": 32,
"minItems": 1,
"title": "Candidates",
"type": "array"
},
"sample_weight": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Sample Weight"
},
"batch": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Batch"
},
"representation_analysis": {
"default": false,
"description": "Opt-in alternative representation comparison and numerical fitting certificates. Binary classification with numeric generic-refit inputs only; requires baseline features. Descriptive, no audit decision changes. Single audits only; limited to 10,000 rows and 250,000 input feature cells.",
"title": "Representation Analysis",
"type": "boolean"
},
"utility": {
"default": "default",
"enum": [
"default",
"absolute_error",
"squared_error"
],
"title": "Utility",
"type": "string"
},
"split": {
"default": "expanding",
"enum": [
"expanding",
"rolling",
"grouped",
"purged"
],
"title": "Split",
"type": "string"
},
"folds": {
"default": 5,
"maximum": 8,
"minimum": 3,
"title": "Folds",
"type": "integer"
},
"embargo_seconds": {
"default": 0,
"maximum": 31536000,
"minimum": 0,
"title": "Embargo Seconds",
"type": "integer"
},
"train_window": {
"default": 100,
"maximum": 100000,
"minimum": 20,
"title": "Train Window",
"type": "integer"
},
"fdr_target": {
"default": 0.05,
"description": "Nominal false-discovery-rate target for the complete submitted family. Expected false fraction, not an individual false-finding probability. Set before evaluation.",
"exclusiveMaximum": 1,
"exclusiveMinimum": 0,
"title": "Fdr Target",
"type": "number"
},
"provisional_error_target": {
"default": 0.1,
"description": "Nominal family-wise error target for provisional evidence. Chance of at least one false rejection, conditional on valid tests. Legacy input name provisional_significance_level is also accepted. Set before evaluation.",
"exclusiveMaximum": 1,
"exclusiveMinimum": 0,
"title": "Provisional Error Target",
"type": "number"
},
"selection_error_target": {
"default": 0.05,
"description": "Nominal family-wise error target for the optional selection diagnostic over strict-scope candidates. Reported only; never changes decisions. Set before evaluation.",
"exclusiveMaximum": 1,
"exclusiveMinimum": 0,
"title": "Selection Error Target",
"type": "number"
},
"desired_power": {
"default": 0.8,
"exclusiveMaximum": 1,
"minimum": 0.5,
"title": "Desired Power",
"type": "number"
},
"target_effect": {
"anyOf": [
{
"exclusiveMinimum": 0,
"type": "number"
},
{
"type": "null"
}
],
"default": null,
"title": "Target Effect"
},
"research_only": {
"const": true,
"default": true,
"title": "Research Only",
"type": "boolean"
},
"deidentified": {
"default": true,
"title": "Deidentified",
"type": "boolean"
},
"scope": {
"const": "feature_quality_only",
"default": "feature_quality_only",
"title": "Scope",
"type": "string"
},
"provisional_significance_level": {
"deprecated": true,
"description": "Deprecated input alias for provisional_error_target; conflicting values are rejected.",
"exclusiveMaximum": 1,
"exclusiveMinimum": 0,
"title": "Provisional Significance Level",
"type": "number"
}
},
"title": "FeatureAuditContract",
"type": "object"
},
"FeatureCampaignShard": {
"additionalProperties": false,
"properties": {
"upload_id": {
"pattern": "^[a-f0-9]{32}$",
"title": "Upload Id",
"type": "string"
},
"contract": {
"$ref": "#/$defs/FeatureAuditContract"
}
},
"required": [
"upload_id",
"contract"
],
"title": "FeatureCampaignShard",
"type": "object"
},
"FeatureProvidedOOFBaseline": {
"additionalProperties": false,
"description": "Public score evidence without eligibility, selections, or action policy.",
"properties": {
"mode": {
"const": "provided_oof",
"title": "Mode",
"type": "string"
},
"score": {
"maxLength": 256,
"minLength": 1,
"title": "Score",
"type": "string"
},
"score_higher_is_better": {
"default": true,
"title": "Score Higher Is Better",
"type": "boolean"
},
"evaluation_group": {
"items": {
"type": "string"
},
"maxItems": 8,
"title": "Evaluation Group",
"type": "array"
},
"row_id": {
"items": {
"type": "string"
},
"maxItems": 8,
"title": "Row Id",
"type": "array"
},
"candidate_scores": {
"additionalProperties": {
"type": "string"
},
"maxProperties": 32,
"title": "Candidate Scores",
"type": "object"
},
"model_version": {
"anyOf": [
{
"maxLength": 128,
"minLength": 1,
"pattern": "^[A-Za-z0-9][A-Za-z0-9._-]*$",
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Model Version"
},
"verification_digest": {
"pattern": "^[a-f0-9]{64}$",
"title": "Verification Digest",
"type": "string"
}
},
"required": [
"mode",
"score",
"verification_digest"
],
"title": "FeatureProvidedOOFBaseline",
"type": "object"
},
"GenericRefitBaseline": {
"additionalProperties": false,
"properties": {
"mode": {
"const": "generic_refit",
"default": "generic_refit",
"title": "Mode",
"type": "string"
},
"features": {
"items": {
"type": "string"
},
"maxItems": 32,
"minItems": 0,
"title": "Features",
"type": "array"
}
},
"required": [
"features"
],
"title": "GenericRefitBaseline",
"type": "object"
}
},
"additionalProperties": false,
"description": "Directory-facing campaign request containing feature-only contracts.",
"title": "FeatureCampaignQuoteRequest"
}🟢get_credits
Read beta or paid audit-credit availability and usage.
입력 스키마
{
"type": "object",
"properties": {},
"additionalProperties": false,
"title": "Empty"
}⚪redeem_beta_invite(invite_code)
Redeem one operator-issued invite for free beta audit credits before January 1, 2027.
입력 스키마
{
"type": "object",
"properties": {
"invite_code": {
"pattern": "^mlr_beta_[A-Za-z0-9_-]{32}$",
"title": "Invite Code",
"type": "string"
}
},
"required": [
"invite_code"
],
"additionalProperties": false,
"title": "BetaInvite"
}⚪request_beta_access(lane, use_case, idempotency_key)
Request operator-reviewed beta access without purchasing or granting credits.
입력 스키마
{
"type": "object",
"properties": {
"lane": {
"enum": [
"referee",
"repair",
"both"
],
"title": "Lane",
"type": "string"
},
"use_case": {
"enum": [
"evaluate_candidate_features",
"repair_degraded_features",
"both",
"explore_fit"
],
"title": "Use Case",
"type": "string"
},
"idempotency_key": {
"maxLength": 128,
"minLength": 8,
"pattern": "^[A-Za-z0-9_-]+$",
"title": "Idempotency Key",
"type": "string"
}
},
"required": [
"lane",
"use_case",
"idempotency_key"
],
"additionalProperties": false,
"title": "BetaAccessRequest"
}🟢get_beta_request_status(request_id)
Read the authenticated tenant's beta-access request status.
입력 스키마
{
"type": "object",
"properties": {
"request_id": {
"pattern": "^[a-f0-9]{32}$",
"title": "Request Id",
"type": "string"
}
},
"required": [
"request_id"
],
"additionalProperties": false,
"title": "BetaRequestId"
}🟡submit_audit(quote_id, idempotency_key)
Accept a quote and atomically reserve its disclosed audit credits.
입력 스키마
{
"type": "object",
"properties": {
"quote_id": {
"pattern": "^[a-f0-9]{32}$",
"title": "Quote Id",
"type": "string"
},
"idempotency_key": {
"maxLength": 128,
"minLength": 8,
"pattern": "^[A-Za-z0-9_-]+$",
"title": "Idempotency Key",
"type": "string"
}
},
"required": [
"quote_id",
"idempotency_key"
],
"additionalProperties": false,
"title": "SubmitRequest"
}⚪record_audit_followup(audit_id, kind, value)
Record whether a completed audit changed a decision or was confirmed prospectively.
입력 스키마
{
"type": "object",
"properties": {
"audit_id": {
"pattern": "^[a-f0-9]{32}$",
"title": "Audit Id",
"type": "string"
},
"kind": {
"enum": [
"decision_impact",
"prospective_outcome"
],
"title": "Kind",
"type": "string"
},
"value": {
"enum": [
"changed_research_plan",
"confirmed_existing_plan",
"no_decision_change",
"unknown",
"confirmed",
"not_confirmed",
"inconclusive",
"not_run"
],
"title": "Value",
"type": "string"
}
},
"required": [
"audit_id",
"kind",
"value"
],
"additionalProperties": false,
"title": "FollowupRequest"
}🟢get_audit_status(audit_id)
Read the state of an audit owned by the authenticated tenant.
입력 스키마
{
"type": "object",
"properties": {
"audit_id": {
"pattern": "^[a-f0-9]{32}$",
"title": "Audit Id",
"type": "string"
}
},
"required": [
"audit_id"
],
"additionalProperties": false,
"title": "Audit"
}🟢get_audit_result(audit_id)
Read a completed tenant-private result and signed output manifest.
입력 스키마
{
"type": "object",
"properties": {
"audit_id": {
"pattern": "^[a-f0-9]{32}$",
"title": "Audit Id",
"type": "string"
}
},
"required": [
"audit_id"
],
"additionalProperties": false,
"title": "Audit"
}🟢get_audit_receipt(audit_id)
Read a tenant-private report fingerprint retained after report expiry or deletion. Older reports may have no record. This does not retrieve report contents.
입력 스키마
{
"type": "object",
"properties": {
"audit_id": {
"pattern": "^[a-f0-9]{32}$",
"title": "Audit Id",
"type": "string"
}
},
"required": [
"audit_id"
],
"additionalProperties": false,
"title": "Audit"
}🟢verify_audit_report(audit_id, sha256, representation)
Compare a locally computed SHA-256 with the retained tenant-private receipt. Use the exact downloaded result JSON bytes (not the API envelope), or UTF-8 result.report text with representation report_text. Never copy the expected hash as the candidate: compute it from the supplied artifact. A digest match does not validate a summary, factual correctness, or absence of harm. No report is uploaded.
입력 스키마
{
"type": "object",
"properties": {
"audit_id": {
"pattern": "^[a-f0-9]{32}$",
"title": "Audit Id",
"type": "string"
},
"sha256": {
"pattern": "^[a-fA-F0-9]{64}$",
"title": "Sha256",
"type": "string"
},
"representation": {
"default": "result_json",
"enum": [
"result_json",
"report_text"
],
"title": "Representation",
"type": "string"
}
},
"required": [
"audit_id",
"sha256"
],
"additionalProperties": false,
"title": "VerifyReportRequest"
}🔴delete_audit(audit_id)
Delete retained inputs and results for an audit owned by this tenant.
입력 스키마
{
"type": "object",
"properties": {
"audit_id": {
"pattern": "^[a-f0-9]{32}$",
"title": "Audit Id",
"type": "string"
}
},
"required": [
"audit_id"
],
"additionalProperties": false,
"title": "Audit"
}🔴delete_upload(upload_id)
Delete a tenant-private upload that is no longer needed.
입력 스키마
{
"type": "object",
"properties": {
"upload_id": {
"pattern": "^[a-f0-9]{32}$",
"title": "Upload Id",
"type": "string"
}
},
"required": [
"upload_id"
],
"additionalProperties": false,
"title": "UploadId"
}커뮤니티
증거
최근 관측
검증됨버전이 기록되지 않음도구 26개