GRANDET EVALUATION STANDARD

Evaluate models, suppliers, prices and quality separately

The historical directory has no measured endpoint-quality scores. Unmeasured values remain null and cannot be inferred from model names, original-creator benchmarks or prices. This document defines proposed measurement and publication rules; it does not report executed tests.

Define the objects and comparison scope

Official identifier groups establish naming evidence; domains identify sources. Neither proves delivery through a particular account, plan or channel. Snapshots, rolling aliases, modality, quantization and reasoning settings retain their distinctions.

Quote key: host × recorded name × group/plan × period × protocol/context/configuration × unit/billing basis. Quota points and panel dollars are not automatically cash. Cash comparison needs matching scope, proven conversion, mandatory fees, eligibility and fresh evidence; host count cannot substitute for independent suppliers.

Token scenario estimate = input share × input rate + output share × output rate using exact rational arithmetic. It is an estimate in the original unit; request, second and unknown billing stay separate. Unknown fees are not zero, and minimum top-up remains distinct from spending.

Model capability

Benchmarks are bound to the tested version, parameters and endpoint; they do not transfer to every relay using the same name.

Current: no publishable score

Endpoint service quality

Bound to the host, model claim, plan, region, workload and observation window.

Current: no publishable score

Documentation disclosure

Evaluates disclosure transparency; it does not establish uptime, model authenticity or speed.

Current: no publishable score

Price qualification

Cash conversion, mandatory fees, account conditions and freshness are checked separately and do not enter quality scores.

Current: no publishable score

Endpoint quality: 100 points, only with all six dimensions

Weights follow Grandet provider-quality-v1; sample gates, formulas and windows are proposed Grandet product policy, not an industry standard.

Current state: method defined; measurement pipeline not executed. Missing samples, observations, workload thresholds, revisions or independent evidence block scores. Known weights are not rescaled to 100 and unmeasured values are not zero.

Dimension / weight	Definition	Formula and minimum samples	Publication gate
Reliable delivery · 35Currently unmeasured	Complete protocol-valid responses within the timeout divided by eligible attempts, with sample size and a 95% Wilson interval; not annual uptime.	`100 * wilson95Lower(valid_completed, eligible_attempts)` { "eligibleAttempts": 100, "calendarDays": 7, "timeBinsPerDay": 3 }	Below sample or temporal-coverage gates, publish raw measurements only and leave the dimension score null.
Billing consistency · 20Currently unmeasured	Recompute expected charges from actual input, output, cache, reasoning and mandatory fees, then compare with actual billing. No bill means no score.	`100 * within_tolerance_bills / eligible_reconciled_bills` { "independentReconciledBills": 30, "allApplicableBillingTypesCovered": true }	Applicable billing types must be covered; zero expected charges must not create a relative-error denominator.
Capacity and rate behavior · 15Currently unmeasured	Successful requests meeting the response objective under preregistered load tiers. Without published service limits this describes only the tested load.	`100 * load_test_requests_meeting_registered_SLO / eligible_load_test_requests` { "preregisteredLoadTiers": 3, "attemptsPerTier": 30 }	Load tiers, per-tier allocation and response objectives must be fixed before testing; selecting only the fastest tier is prohibited.
Time to first token · 10Currently unmeasured	Elapsed time from request send to the first nonempty model event; first-answer latency is reported separately.	`100 * clamp((T_bad - p95_TTFT_seconds) / (T_bad - T_good), 0, 1)` { "successfulRequests": 100, "reliabilitySamplingGateMustPass": true }	Missing workload thresholds, samples or reliability coverage leave the score null; workloads and regions must not be pooled.
Generation speed · 10Currently unmeasured	Per-request native visible output tokens/s after the first token, with P50, P10 and full duration; hidden reasoning is not mixed into visible speed.	`100 * clamp((p10_visible_tokens_per_second - S_bad) / (S_good - S_bad), 0, 1)` { "successfulRequests": 100, "reliabilitySamplingGateMustPass": true, "fixedOutputLengthBin": true }	Tokenizer, output-length bin and configuration must match; aggregate concurrent throughput cannot substitute for single-request speed.
Evidence breadth and freshness · 10Currently unmeasured	Raw evidence, fixed methods, planned coverage, valid window and independent replication earn 20 points each; this evaluates evidence, not endpoint behavior.	`20 * verified_pass_check_count, only when all five checks have been completed` { "completedChecks": 5 }	A checked failure earns zero; an unchecked item remains unknown. Any unknown check leaves this dimension null.

Total = Σ(dimension score × weight)/100. Proposed window and observation expiry: seven days. Recalculation does not refresh observation time. First model token and first answer token are separate; heartbeats do not count. Fallback success does not erase earlier failures. Generation speed binds the native tokenizer, visible output and fixed output lengths; aggregate throughput is not per-user speed.

Exact definitions, threshold registration and machine rules

{
  "scopeKey": [
    "retailOperatorIdOrUnknown",
    "host",
    "endpointId",
    "rawModelName",
    "canonicalModelIdOrUnknown",
    "plan",
    "protocol",
    "region",
    "workloadId",
    "parametersHash",
    "window"
  ],
  "commonGates": {
    "sameScopeRequired": true,
    "rawObservationEvidenceRequired": true,
    "methodAndParametersVersionRequired": true,
    "numericScoreOnlyIfAllSixDimensionsAvailable": true,
    "unknownBecomesZero": false,
    "rescaleKnownWeights": false,
    "minimumIndependentEvidenceSources": 3,
    "grandetOfficialEvidenceAlternative": true,
    "evidenceSourceGateDoesNotReplaceSampleGates": true,
    "proposedWindowDays": 7,
    "proposedObservationExpiryDays": 7,
    "recalculationRefreshesObservationTime": false,
    "sourceConfidenceIsStatisticalConfidence": false,
    "missingWorkloadThresholdsBlockScore": true
  },
  "measurementRules": {
    "eligibleAttempt": "A request conforms to the declared protocol, capability, account and known limits and was not deliberately cancelled by the user; excluded counts and reasons remain visible.",
    "success": "The request returns a complete, protocol-valid response before its predeclared timeout; success does not imply a correct task answer.",
    "failure": "Timeouts, incomplete streams, malformed output and errors count as failures. Rate errors inside declared capacity count as failures. Fallback success does not erase a previous endpoint failure.",
    "retry": "Each attempt is recorded separately; first-attempt performance and end-to-end experience remain distinct.",
    "firstToken": "The first nonempty model-generated output event; empty SSE events and heartbeats are excluded.",
    "firstAnswerToken": "The first nonempty answer event after any visible reasoning; reported separately from first model token.",
    "tokenComparability": "Generation speed uses an identified native tokenizer within the same model; visible answer tokens and hidden reasoning tokens remain separate. Cross-model comparisons use separately declared normalized-token or fixed-task measures.",
    "percentileConvention": "Nearest-rank quantile: sorted sample at index max(1,ceil(p*n)), one-based. Publish n and percentile definition.",
    "wilson95Lower": "For n>0, p=k/n, z=1.959963984540054: (p+z*z/(2*n)-z*sqrt(p*(1-p)/n+z*z/(4*n*n)))/(1+z*z/n).",
    "workloadThresholdRegistry": "TTFT good/bad seconds, generation good/bad tokens_per_second, timeouts, load tiers, input/output lengths, cache behavior and reasoning budget must be versioned and fixed before testing. No default threshold is silently supplied."
  },
  "aggregation": {
    "formula": "sum(weight_i * dimension_score_i) / 100",
    "weightTotal": 100,
    "minimumWeightedCoverage": 100,
    "preserveUnroundedValue": true,
    "grades": {
      "A": 90,
      "B": 80,
      "C": 70,
      "D": 60,
      "E": 50,
      "F": 0
    },
    "confirmedIdentityFraudCapFromExistingMethod": 20,
    "capDoesNotDetectIdentityFraud": true,
    "publicationWhenIncomplete": "NOT_SCORED_WITH_EXPLICIT_WEIGHTED_COVERAGE",
    "priceAffectsScore": false,
    "creatorBenchmarkAffectsEndpointScore": false,
    "identityNameMatchAffectsScore": false
  }
}

Documentation disclosure is scored separately

Retains the 20 document-score-v1 checks, five points each. Publication needs at least 50% coverage and completed P01, T02 and F01 checks. Unknown items retain intervals; disclosure scores do not certify service quality, authenticity or policy compliance.

20 documentation checks and evidence rules

ID	Check	Met: 5 points	Partial: 2.5 points
P01 *	Model and input/output billing units	The model and both input and output billing units are explicit.	Only the model or the billing unit is explicit.
P02	Cash currency and credit conversion	The cash currency is explicit, with verifiable conversion rules for credits or an explicit statement that the amount is cash.	Only the currency label is explicit; the cash-to-credit relationship is unclear.
P03	Mandatory charges and billing conditions	All mandatory charges are explicit, or no additional charges are explicitly stated, and the applicable conditions are clear.	Only some charges or conditions are disclosed.
P04	Minimum deposit and payment conditions	The minimum top-up, including an explicit absence of a minimum, and payment conditions are stated.	Only one of these is stated.
P05	Quote scope and version	The applicable models/plans and price-update or validity rules are both explicit.	Only the applicability scope is explicit.
I01	Identifiable legal operator	A legal operator is named and has a registration identifier or registration evidence that can be cross-checked.	Only a brand or a legal name without cross-verification is available.
I02	Service and complaint contacts	Service contacts and a responsibility or complaints channel are explicit.	Only a general contact channel is available.
I03	Responsibility and jurisdiction disclosure	The responsible entity and applicable law or dispute jurisdiction are both named.	Only one of these is explicit.
T01	Key service terms	Locatable terms cover service, payment, suspension and the scope of responsibility.	Only some of these terms are covered.
T02 *	Complete data policy disclosure	Data types, purposes, retention and upstream sharing are all explained explicitly.	Only some of these matters are explained.
C01	Cross-page consistency	At least two key pages, such as pricing and terms, have been compared with no conflict affecting applicable conditions found.	Only a limited portion of the cross-check has been completed.
C02	Traceable terms revisions	A date or version and verifiable changes or historical records are available.	A date or version exists, but no change history is available.
F01 *	Unused-balance refund protection	Refunds of unused balances are explicitly allowed in ordinary circumstances, with a clear process and deduction conditions.	Refunds are limited to narrow exceptions, or the process is incomplete.
F02	Balance expiry and deduction limits	No expiry is explicit, or a fixed expiry was disclosed in advance, with no arbitrary forfeiture.	Expiry or deduction rules exist, but protection is incomplete.
F03	Balance handling after suspension	Notice or appeal procedures and protection for handling unused balances are available.	Only one of these is explicit.
D01	Request-content purpose limits	Content purposes are explicitly limited, and additional training or sale is prohibited.	Only some uses are restricted.
D02	Retention and deletion limits	A maximum content-retention period or no retention is explicit, together with deletion mechanisms and exceptions.	Only general retention principles or a deletion channel are provided.
D03	Upstream processing boundaries	Upstream participation and the applicable policies or processing limits are explicit.	Only a general statement about third-party participation is available.
V01	Price and material-terms change protection	Advance notice is explicit, and paid entitlements cannot be changed retroactively.	Only an announcement or notification mechanism is provided.
V02	Complaint and appeal process	An appeal channel and handling steps or time limits are explicit.	Only a channel is provided.

{
  "methodVersion": "document-score-v1.0.0",
  "checkCount": 20,
  "weightPerCheck": 5,
  "outcomePoints": {
    "MET": 5,
    "PARTIAL": 2.5,
    "NOT_MET": 0,
    "NOT_DISCLOSED": 0,
    "NOT_CHECKED": null,
    "CONFLICT": null
  },
  "minimumCoveragePercent": 50,
  "criticalChecks": [
    "P01",
    "F01",
    "T02"
  ],
  "intervalMeaning": "Possible score range from unresolved check weights, not a statistical confidence interval",
  "filterUsesLowerBound": true,
  "currentPublicScore": null,
  "checks": [
    {
      "id": "P01",
      "label": {
        "zh": "明确模型及输入输出计费单位",
        "en": "Model and input/output billing units"
      },
      "category": "计费披露",
      "met": "模型及双向单位均明确",
      "partial": "仅模型或计费单位明确",
      "critical": true,
      "weight_bps": 500,
      "weightPoints": 5,
      "notMet": "明确相反条款，且保留可定位原文证据。",
      "notDisclosed": "完整检查全部相关文档后没有披露；必须保存引用与 COMPLETE_RELEVANT_DOCUMENTS 依据。",
      "notChecked": "无法完成核查；保持未知。",
      "conflict": "相互冲突的证据尚未裁定；保持未知。",
      "met_en": "The model and both input and output billing units are explicit.",
      "partial_en": "Only the model or the billing unit is explicit."
    },
    {
      "id": "P02",
      "label": {
        "zh": "现金币种与额度兑换关系",
        "en": "Cash currency and credit conversion"
      },
      "category": "计费披露",
      "met": "现金币种明确，积分另有可核验兑换规则或明确为现金",
      "partial": "仅币种标签明确，现金/积分兑换不明",
      "critical": false,
      "weight_bps": 500,
      "weightPoints": 5,
      "notMet": "明确相反条款，且保留可定位原文证据。",
      "notDisclosed": "完整检查全部相关文档后没有披露；必须保存引用与 COMPLETE_RELEVANT_DOCUMENTS 依据。",
      "notChecked": "无法完成核查；保持未知。",
      "conflict": "相互冲突的证据尚未裁定；保持未知。",
      "met_en": "The cash currency is explicit, with verifiable conversion rules for credits or an explicit statement that the amount is cash.",
      "partial_en": "Only the currency label is explicit; the cash-to-credit relationship is unclear."
    },
    {
      "id": "P03",
      "label": {
        "zh": "强制附加费用与计费条件",
        "en": "Mandatory charges and billing conditions"
      },
      "category": "计费披露",
      "met": "明确所有强制费用或明确无附加费，且适用条件清楚",
      "partial": "只披露部分费用或条件",
      "critical": false,
      "weight_bps": 500,
      "weightPoints": 5,
      "notMet": "明确相反条款，且保留可定位原文证据。",
      "notDisclosed": "完整检查全部相关文档后没有披露；必须保存引用与 COMPLETE_RELEVANT_DOCUMENTS 依据。",
      "notChecked": "无法完成核查；保持未知。",
      "conflict": "相互冲突的证据尚未裁定；保持未知。",
      "met_en": "All mandatory charges are explicit, or no additional charges are explicitly stated, and the applicable conditions are clear.",
      "partial_en": "Only some charges or conditions are disclosed."
    },
    {
      "id": "P04",
      "label": {
        "zh": "最低充值与付款条件",
        "en": "Minimum deposit and payment conditions"
      },
      "category": "计费披露",
      "met": "明确最低充值（含无最低）及付款条件",
      "partial": "只披露其中一项",
      "critical": false,
      "weight_bps": 500,
      "weightPoints": 5,
      "notMet": "明确相反条款，且保留可定位原文证据。",
      "notDisclosed": "完整检查全部相关文档后没有披露；必须保存引用与 COMPLETE_RELEVANT_DOCUMENTS 依据。",
      "notChecked": "无法完成核查；保持未知。",
      "conflict": "相互冲突的证据尚未裁定；保持未知。",
      "met_en": "The minimum top-up, including an explicit absence of a minimum, and payment conditions are stated.",
      "partial_en": "Only one of these is stated."
    },
    {
      "id": "P05",
      "label": {
        "zh": "报价适用范围与版本",
        "en": "Quote scope and version"
      },
      "category": "计费披露",
      "met": "模型/套餐适用范围及价格更新或有效期规则均明确",
      "partial": "只明确适用范围",
      "critical": false,
      "weight_bps": 500,
      "weightPoints": 5,
      "notMet": "明确相反条款，且保留可定位原文证据。",
      "notDisclosed": "完整检查全部相关文档后没有披露；必须保存引用与 COMPLETE_RELEVANT_DOCUMENTS 依据。",
      "notChecked": "无法完成核查；保持未知。",
      "conflict": "相互冲突的证据尚未裁定；保持未知。",
      "met_en": "The applicable models/plans and price-update or validity rules are both explicit.",
      "partial_en": "Only the applicability scope is explicit."
    },
    {
      "id": "I01",
      "label": {
        "zh": "运营主体可识别",
        "en": "Identifiable legal operator"
      },
      "category": "主体责任",
      "met": "具名法律主体且有可交叉核验注册标识或登记证据",
      "partial": "只有品牌或未交叉核实的法律名称",
      "critical": false,
      "weight_bps": 500,
      "weightPoints": 5,
      "notMet": "明确相反条款，且保留可定位原文证据。",
      "notDisclosed": "完整检查全部相关文档后没有披露；必须保存引用与 COMPLETE_RELEVANT_DOCUMENTS 依据。",
      "notChecked": "无法完成核查；保持未知。",
      "conflict": "相互冲突的证据尚未裁定；保持未知。",
      "met_en": "A legal operator is named and has a registration identifier or registration evidence that can be cross-checked.",
      "partial_en": "Only a brand or a legal name without cross-verification is available."
    },
    {
      "id": "I02",
      "label": {
        "zh": "可用责任联系渠道",
        "en": "Service and complaint contacts"
      },
      "category": "主体责任",
      "met": "明确服务联系渠道及责任/投诉渠道",
      "partial": "只有一般联系渠道",
      "critical": false,
      "weight_bps": 500,
      "weightPoints": 5,
      "notMet": "明确相反条款，且保留可定位原文证据。",
      "notDisclosed": "完整检查全部相关文档后没有披露；必须保存引用与 COMPLETE_RELEVANT_DOCUMENTS 依据。",
      "notChecked": "无法完成核查；保持未知。",
      "conflict": "相互冲突的证据尚未裁定；保持未知。",
      "met_en": "Service contacts and a responsibility or complaints channel are explicit.",
      "partial_en": "Only a general contact channel is available."
    },
    {
      "id": "I03",
      "label": {
        "zh": "法律责任和管辖披露",
        "en": "Responsibility and jurisdiction disclosure"
      },
      "category": "主体责任",
      "met": "责任主体及适用法律/争议管辖均具名",
      "partial": "只明确其中一项",
      "critical": false,
      "weight_bps": 500,
      "weightPoints": 5,
      "notMet": "明确相反条款，且保留可定位原文证据。",
      "notDisclosed": "完整检查全部相关文档后没有披露；必须保存引用与 COMPLETE_RELEVANT_DOCUMENTS 依据。",
      "notChecked": "无法完成核查；保持未知。",
      "conflict": "相互冲突的证据尚未裁定；保持未知。",
      "met_en": "The responsible entity and applicable law or dispute jurisdiction are both named.",
      "partial_en": "Only one of these is explicit."
    },
    {
      "id": "T01",
      "label": {
        "zh": "服务条款覆盖关键事项",
        "en": "Key service terms"
      },
      "category": "条款数据披露",
      "met": "可定位条款包含服务、付款、停用和责任范围",
      "partial": "只有部分条款",
      "critical": false,
      "weight_bps": 500,
      "weightPoints": 5,
      "notMet": "明确相反条款，且保留可定位原文证据。",
      "notDisclosed": "完整检查全部相关文档后没有披露；必须保存引用与 COMPLETE_RELEVANT_DOCUMENTS 依据。",
      "notChecked": "无法完成核查；保持未知。",
      "conflict": "相互冲突的证据尚未裁定；保持未知。",
      "met_en": "Locatable terms cover service, payment, suspension and the scope of responsibility.",
      "partial_en": "Only some of these terms are covered."
    },
    {
      "id": "T02",
      "label": {
        "zh": "数据政策披露完整",
        "en": "Complete data policy disclosure"
      },
      "category": "条款数据披露",
      "met": "数据类型、用途、留存及上游共享均有明确说明",
      "partial": "只有部分说明",
      "critical": true,
      "weight_bps": 500,
      "weightPoints": 5,
      "notMet": "明确相反条款，且保留可定位原文证据。",
      "notDisclosed": "完整检查全部相关文档后没有披露；必须保存引用与 COMPLETE_RELEVANT_DOCUMENTS 依据。",
      "notChecked": "无法完成核查；保持未知。",
      "conflict": "相互冲突的证据尚未裁定；保持未知。",
      "met_en": "Data types, purposes, retention and upstream sharing are all explained explicitly.",
      "partial_en": "Only some of these matters are explained."
    },
    {
      "id": "C01",
      "label": {
        "zh": "跨页面关键声明一致",
        "en": "Cross-page consistency"
      },
      "category": "一致与可追溯",
      "met": "已比较价格页与条款等至少两处关键页面且未发现影响条件的冲突",
      "partial": "只完成有限部分交叉检查",
      "critical": false,
      "weight_bps": 500,
      "weightPoints": 5,
      "notMet": "明确相反条款，且保留可定位原文证据。",
      "notDisclosed": "完整检查全部相关文档后没有披露；必须保存引用与 COMPLETE_RELEVANT_DOCUMENTS 依据。",
      "notChecked": "无法完成核查；保持未知。",
      "conflict": "相互冲突的证据尚未裁定；保持未知。",
      "met_en": "At least two key pages, such as pricing and terms, have been compared with no conflict affecting applicable conditions found.",
      "partial_en": "Only a limited portion of the cross-check has been completed."
    },
    {
      "id": "C02",
      "label": {
        "zh": "条款修订可追溯",
        "en": "Traceable terms revisions"
      },
      "category": "一致与可追溯",
      "met": "有日期/版本及可核查变更或历史记录",
      "partial": "有日期/版本但无变更记录",
      "critical": false,
      "weight_bps": 500,
      "weightPoints": 5,
      "notMet": "明确相反条款，且保留可定位原文证据。",
      "notDisclosed": "完整检查全部相关文档后没有披露；必须保存引用与 COMPLETE_RELEVANT_DOCUMENTS 依据。",
      "notChecked": "无法完成核查；保持未知。",
      "conflict": "相互冲突的证据尚未裁定；保持未知。",
      "met_en": "A date or version and verifiable changes or historical records are available.",
      "partial_en": "A date or version exists, but no change history is available."
    },
    {
      "id": "F01",
      "label": {
        "zh": "未用余额退款保障",
        "en": "Unused-balance refund protection"
      },
      "category": "余额退款停用",
      "met": "明确允许正常情形退还未用余额且流程及扣费条件清楚",
      "partial": "仅有限例外退款或流程不完整",
      "critical": true,
      "weight_bps": 500,
      "weightPoints": 5,
      "notMet": "明确相反条款，且保留可定位原文证据。",
      "notDisclosed": "完整检查全部相关文档后没有披露；必须保存引用与 COMPLETE_RELEVANT_DOCUMENTS 依据。",
      "notChecked": "无法完成核查；保持未知。",
      "conflict": "相互冲突的证据尚未裁定；保持未知。",
      "met_en": "Refunds of unused balances are explicitly allowed in ordinary circumstances, with a clear process and deduction conditions.",
      "partial_en": "Refunds are limited to narrow exceptions, or the process is incomplete."
    },
    {
      "id": "F02",
      "label": {
        "zh": "余额失效与扣除约束",
        "en": "Balance expiry and deduction limits"
      },
      "category": "余额退款停用",
      "met": "明确不失效或有事先披露的固定期限且无任意没收",
      "partial": "有期限/扣除规则但保护不完整",
      "critical": false,
      "weight_bps": 500,
      "weightPoints": 5,
      "notMet": "明确相反条款，且保留可定位原文证据。",
      "notDisclosed": "完整检查全部相关文档后没有披露；必须保存引用与 COMPLETE_RELEVANT_DOCUMENTS 依据。",
      "notChecked": "无法完成核查；保持未知。",
      "conflict": "相互冲突的证据尚未裁定；保持未知。",
      "met_en": "No expiry is explicit, or a fixed expiry was disclosed in advance, with no arbitrary forfeiture.",
      "partial_en": "Expiry or deduction rules exist, but protection is incomplete."
    },
    {
      "id": "F03",
      "label": {
        "zh": "账户停用余额处理",
        "en": "Balance handling after suspension"
      },
      "category": "余额退款停用",
      "met": "有通知/申诉流程及未用余额处理保障",
      "partial": "只明确其中一项",
      "critical": false,
      "weight_bps": 500,
      "weightPoints": 5,
      "notMet": "明确相反条款，且保留可定位原文证据。",
      "notDisclosed": "完整检查全部相关文档后没有披露；必须保存引用与 COMPLETE_RELEVANT_DOCUMENTS 依据。",
      "notChecked": "无法完成核查；保持未知。",
      "conflict": "相互冲突的证据尚未裁定；保持未知。",
      "met_en": "Notice or appeal procedures and protection for handling unused balances are available.",
      "partial_en": "Only one of these is explicit."
    },
    {
      "id": "D01",
      "label": {
        "zh": "请求内容处理范围限制",
        "en": "Request-content purpose limits"
      },
      "category": "数据保障",
      "met": "明确内容用途限定及不作额外训练/出售",
      "partial": "只对部分用途作限制",
      "critical": false,
      "weight_bps": 500,
      "weightPoints": 5,
      "notMet": "明确相反条款，且保留可定位原文证据。",
      "notDisclosed": "完整检查全部相关文档后没有披露；必须保存引用与 COMPLETE_RELEVANT_DOCUMENTS 依据。",
      "notChecked": "无法完成核查；保持未知。",
      "conflict": "相互冲突的证据尚未裁定；保持未知。",
      "met_en": "Content purposes are explicitly limited, and additional training or sale is prohibited.",
      "partial_en": "Only some uses are restricted."
    },
    {
      "id": "D02",
      "label": {
        "zh": "留存与删除约束",
        "en": "Retention and deletion limits"
      },
      "category": "数据保障",
      "met": "明确内容留存上限/不留存及删除机制和例外",
      "partial": "仅一般期限原则或只有删除渠道",
      "critical": false,
      "weight_bps": 500,
      "weightPoints": 5,
      "notMet": "明确相反条款，且保留可定位原文证据。",
      "notDisclosed": "完整检查全部相关文档后没有披露；必须保存引用与 COMPLETE_RELEVANT_DOCUMENTS 依据。",
      "notChecked": "无法完成核查；保持未知。",
      "conflict": "相互冲突的证据尚未裁定；保持未知。",
      "met_en": "A maximum content-retention period or no retention is explicit, together with deletion mechanisms and exceptions.",
      "partial_en": "Only general retention principles or a deletion channel are provided."
    },
    {
      "id": "D03",
      "label": {
        "zh": "上游处理边界",
        "en": "Upstream processing boundaries"
      },
      "category": "数据保障",
      "met": "明确上游参与及适用政策/处理限制",
      "partial": "仅笼统说明第三方参与",
      "critical": false,
      "weight_bps": 500,
      "weightPoints": 5,
      "notMet": "明确相反条款，且保留可定位原文证据。",
      "notDisclosed": "完整检查全部相关文档后没有披露；必须保存引用与 COMPLETE_RELEVANT_DOCUMENTS 依据。",
      "notChecked": "无法完成核查；保持未知。",
      "conflict": "相互冲突的证据尚未裁定；保持未知。",
      "met_en": "Upstream participation and the applicable policies or processing limits are explicit.",
      "partial_en": "Only a general statement about third-party participation is available."
    },
    {
      "id": "V01",
      "label": {
        "zh": "价格和重大条款变更保障",
        "en": "Price and material-terms change protection"
      },
      "category": "变更与争议",
      "met": "明确提前通知且不追溯更改已付权益",
      "partial": "只有公告/通知机制",
      "critical": false,
      "weight_bps": 500,
      "weightPoints": 5,
      "notMet": "明确相反条款，且保留可定位原文证据。",
      "notDisclosed": "完整检查全部相关文档后没有披露；必须保存引用与 COMPLETE_RELEVANT_DOCUMENTS 依据。",
      "notChecked": "无法完成核查；保持未知。",
      "conflict": "相互冲突的证据尚未裁定；保持未知。",
      "met_en": "Advance notice is explicit, and paid entitlements cannot be changed retroactively.",
      "partial_en": "Only an announcement or notification mechanism is provided."
    },
    {
      "id": "V02",
      "label": {
        "zh": "投诉申诉程序",
        "en": "Complaint and appeal process"
      },
      "category": "变更与争议",
      "met": "明确申诉渠道及处理步骤/时限",
      "partial": "只有渠道",
      "critical": false,
      "weight_bps": 500,
      "weightPoints": 5,
      "notMet": "明确相反条款，且保留可定位原文证据。",
      "notDisclosed": "完整检查全部相关文档后没有披露；必须保存引用与 COMPLETE_RELEVANT_DOCUMENTS 依据。",
      "notChecked": "无法完成核查；保持未知。",
      "conflict": "相互冲突的证据尚未裁定；保持未知。",
      "met_en": "An appeal channel and handling steps or time limits are explicit.",
      "partial_en": "Only a channel is provided."
    }
  ],
  "rulesSha256": "6a7ed66d63eecdf93b03f628f90b8d10e08086b6ca65d183d490858bcb8d6ed5",
  "sourceRulesFileSha256": "6bd191729dea7b25c8ea105888fecc834ce175e830b63234812521be93f5c7a8",
  "notDisclosedRequires": "Complete relevant documents examined; full-source reference and absence_basis required",
  "sourceBoundary": "Provider statements score documented conditions only; never runtime truth",
  "scoreFormula": "L = sum checked MET/PARTIAL points; U = sum NOT_CHECKED/CONFLICT weights; report [L,L+U] only when publication gates pass.",
  "criticalUnknownBlocksScore": true,
  "expiryBlocksScore": true,
  "noArbitraryNotApplicable": true,
  "evidenceRequirements": [
    "Each check scope_id must equal the assessment scope_id.",
    "Every rule must occur exactly once; fixed total weight is 100.",
    "Checked outcomes require evidence IDs registered with URL, retrieved_at, source kind, SHA256, locator and excerpt.",
    "NOT_CHECKED cannot claim checked evidence.",
    "Derived analysis alone cannot independently support checked outcomes.",
    "Evidence observation cannot postdate the assessment.",
    "Machine byte/pointer verification proves locatability, not semantic correctness.",
    "The existing source method requires separate semantic/context review.",
    "Recalculation cannot refresh the source observation time.",
    "A MET conclusion must satisfy every conjunct in the met criterion.",
    "No arbitrary N/A outcome may shrink the denominator."
  ]
}

Complete documentation scoring criteria JSON

Website acceptance: 20 checks × 5 points

This grades correctness and usability of the local historical preview, not provider quality. Pass = 5, fail = 0; unchecked items prevent a total. This cycle requires at least 95, zero open P0/P1 issues and independent review. Unmeasured service and unresolved identity stay explicit.

ID	Check	Points	Required pass criteria
I01	Model identity boundaries	5	Recorded-name counts, canonical-model counts and configuration/rule counts are labelled separately; unresolved items are not advertised as independent models.
I02	Attribution evidence	5	Every mapping has an explicit status and an evidence or rule version. String similarity alone permits only a candidate attribution and cannot certify actual supplier delivery.
I03	Non-model entries and versions	5	Tests, operations and wildcard rules are separate; snapshots, modalities, context, budgets and channel parameters are not silently stripped, and conflicts are retained.
I04	Two-way lookup	5	A canonical model leads to all associated declarations and quotes; an exact recorded name retrieves its attribution and all original prices; unresolved items remain accessible.
S01	Retail operators and domains	5	Source hosts, brands and verified legal operators are distinguished; domain counts are not presented as independent-supplier counts.
S02	Upstream and channels	5	Declared upstreams and verified upstreams remain distinct; a source prefix does not certify a supply chain, and plans/groups are retained.
S03	Supplier coverage	5	Every captured host is searchable or locatable; detail views do not omit data because of the old 20-model allowlist, and count scopes are explicit.
P01	Original prices and billing fidelity	5	Cash labels, credits, per-request values, formulas and unknown original fields are all locatable. Unknowns are not replaced with zero; original numeric lexemes and units are retained.
P02	Comparison scope	5	Comparability requires evidence for model identity, protocol, configuration, context, capture period, cash conversion and fee completeness; original-value sorting states its limits.
P03	Calculation correctness	5	Input/output shares and units are correct within the same scope; rational arithmetic prevents misordering; per-request and special billing are not processed as Token prices; cache and fee conditions are visible.
P04	Time and qualification	5	Observation dates and capture periods are visible; historical records are not presented as current quotes. No formal cash rank is generated without verified cash, account, fee and independent-operator requirements.
Q01	Separate four kinds of conclusions	5	Original-creator model capability, endpoint measurements, documentation disclosure and price qualification are presented and cited separately; reputation is not used to infer endpoint performance.
Q02	Missing data and provenance	5	Missing samples, expired evidence and insufficient coverage remain unknown with null scores; every published score is traceable to samples, method and observation window.
Q03	Executable methodology	5	Documentation checks and thresholds, and quality dimensions, measurement definitions, formulas, sample gates and publication gates are public; the current unmeasured state and method status are explicit.
D01	Complete source-to-export coverage	5	An independent read of PART2/PART3 source rows and configuration pointers matches exported stable-ID sets; missing entries and unintended duplicates both equal zero.
D02	Complete export-to-display coverage	5	Expected expanded IDs for raw records, rates, configurations, legacy catalog entries and orphan records equal those produced by boardRecords; every filter can be cleared, and unresolved entries are not omitted by default.
D03	Evidence completeness	5	URLs, source excerpts or safe original files, JSON pointers, hashes and observation times are traceable; unknown and empty sources are explicit; credentials are not exposed.
U01	Core interactions	5	Desktop and mobile checks verify model/host search, switching, pagination, direct refresh, filter reset, details and downloads; failed loads are retryable and do not appear complete.
U02	Client and server consistency	5	The persistent preview service works; server-rendered and client names, data and results agree; core browser flows have no uncaught errors; build and regression checks pass.
G01	Readable and citable output	5	Server-rendered pages expose readable identities, prices and status; page methodology and machine data use consistent concepts; Chinese and English wording agrees; local noindex and undeployed status are truthful.

References and machine data

Full quality method JSON · Website review rubric JSON