> ## Documentation Index > Fetch the complete documentation index at: https://docs.erdo.ai/llms.txt > Use this file to discover all available pages before exploring further. # DecisionScorecardAPI mirrors erdo_decision_scorecard: raw aggregates with their > denominators, stratified by decision class and evidence kind, with no eligibility verdict and no pooled outcome rate. ## OpenAPI ````yaml /api/openapi.json get /v1/decisions-scorecard openapi: 3.0.0 info: title: Erdo API description: >- Erdo's REST API: query and write datasets, run agents, manage threads, integrations, pages, evals, workstreams, and experiments. Authenticate with a Bearer API key (erdo_api_...) or scoped token (erdo_token_...). version: '2026-08-07' servers: - url: https://api.erdo.ai description: Production security: - bearerAuth: [] paths: /v1/decisions-scorecard: get: summary: > DecisionScorecardAPI mirrors erdo_decision_scorecard: raw aggregates with their description: > denominators, stratified by decision class and evidence kind, with no eligibility verdict and no pooled outcome rate. operationId: GET:mcp.DecisionScorecardAPI parameters: - allowEmptyValue: true explode: true in: query name: workstream_slug required: false schema: type: string style: form - allowEmptyValue: true explode: true in: query name: source required: false schema: type: string style: form - allowEmptyValue: true explode: true in: query name: decision_class required: false schema: type: string style: form - allowEmptyValue: true explode: true in: query name: since required: false schema: type: string style: form - allowEmptyValue: true explode: true in: query name: until required: false schema: type: string style: form responses: '200': content: application/json: schema: properties: decide_latency: $ref: '#/components/schemas/types.DecisionDecideLatency' deciders: $ref: '#/components/schemas/types.DecisionDeciderBreakdown' execution: $ref: '#/components/schemas/types.DecisionExecutionCoverage' measurement: $ref: '#/components/schemas/types.DecisionMeasurementCoverage' outcomes_by_class_and_evidence: description: > deterministic postcondition and an observational before/after are not comparable evidence, and one campaign class's results say nothing about another's. items: $ref: '#/components/schemas/types.DecisionOutcomeStratum' title: > OutcomesByClassAndEvidence is the outcome split, stratified. Never pooled: a type: array outcomes_by_source_and_evidence: description: > family that recorded them — the actor/policy regime axis. A record where the engine gate's decisions settle differently from a human's is a fact the class cut cannot show. items: $ref: '#/components/schemas/types.DecisionOutcomeStratum' title: > OutcomesBySourceAndEvidence cuts the same settled effects by the producer type: array totals: $ref: '#/components/schemas/types.DecisionTotals' window: $ref: '#/components/schemas/types.DecisionScorecardWindow' type: object description: Success response default: $ref: '#/components/responses/APIError' components: schemas: types.DecisionDecideLatency: description: | being authorized, in seconds. properties: decided: description: | reach authorized\_at — so this describes the time to a yes. format: int64 title: > Decided is how many decisions contributed. Rejections are absent — they never type: integer p50_seconds: type: number p90_seconds: type: number title: > DecisionDecideLatency is how long decisions waited between being proposed and type: object types.DecisionDeciderBreakdown: properties: answered: description: > proposal nobody has answered has no decider yet, so including it would deflate every rate below by however long the queue happens to be. format: int64 title: > Answered is the denominator: decisions that reached a decider at all. A type: integer by_decider_kind: additionalProperties: format: int64 type: integer title: > ByDeciderKind: human, llm_agent, deterministic_policy, safe_default, system. type: object human_override: description: > rejected, decided by a person. It is the disagreement rate, and it is reported beside its denominator rather than as a bare percentage because ten refusals out of twelve and ten out of a thousand are opposite findings. format: int64 title: > HumanOverride is a human refusing something the system proposed — status type: integer human_override_rate: type: number safe_default: description: > the clearest signal in this whole readout that a queue is being ignored rather than governed. format: int64 title: > SafeDefault is how often nobody answered and the safe default applied. It is type: integer safe_default_rate: type: number type: object types.DecisionExecutionCoverage: properties: actions: format: int64 type: integer by_status: additionalProperties: format: int64 type: integer title: | ByStatus counts every action by its single attempt's outcome. type: object decisions_with_actions: description: > all. The rest are real decisions with nothing to execute — a strategy commitment, an experiment settlement — and counting them as unexecuted would invent a failure. format: int64 title: > DecisionsWithActions is how many decisions authorized an external call at type: integer decisions_without_actions: format: int64 type: integer unknown: description: > whether the rest of this section can be trusted: an action nothing confirmed may or may not be live, and a record with many of them is one whose execution truth is partly unknown. format: int64 title: > Unknown is called out separately because it is the number that decides type: integer type: object types.DecisionMeasurementCoverage: properties: by_measurability: additionalProperties: format: int64 type: integer description: > that never will: measurable, unmeasurable, unmeasurable\_invalid\_contract. title: > ByMeasurability separates the declarations that can be scored from the ones type: object by_status: additionalProperties: format: int64 type: integer title: > ByStatus is where every effect currently sits, including the parked ones. type: object censored: format: int64 type: integer decisions_with_declared_effect: description: > expected to change. The complement is not a failure — most decisions have no business effect anything could measure — but it bounds everything the outcome strata can claim about the record as a whole. format: int64 title: > DecisionsWithDeclaredEffect is how many decisions said in advance what they type: integer decisions_without_declared_effect: format: int64 type: integer effects: format: int64 type: integer measurement_unavailable: format: int64 type: integer settled: description: > numbers a reader needs to size the outcome split honestly: what produced an answer, what was overtaken before it could, and what is waiting on a reader that does not exist yet. The size of Unavailable is what will justify building the synced-dataset measurement runner. format: int64 title: > Settled, Censored and Unavailable are pulled out because they are the three type: integer type: object types.DecisionOutcomeStratum: description: > stratum names the evidence kind that settled it, because the same counts mean different things under different evidence and the label is what stops a reader carrying the stronger claim across. properties: decision_class: description: | other is empty. title: > DecisionClass or Source, depending on which cut this stratum belongs to; the type: string evidence_kind: type: string inconclusive: format: int64 type: integer met: format: int64 type: integer met_label: description: > SAY about those two counts — "confirmed"/"not confirmed" for deterministic, "worked"/"did not work" only for experimental, "moved as expected"/"did not move as expected" for observational. They ride on the stratum so no consumer has to re-derive the vocabulary, and so a surface that ignores the evidence kind is visibly ignoring it. title: > MetLabel and NotMetLabel are what this stratum's evidence entitles a reader to type: string not_met: format: int64 type: integer not_met_label: type: string settled: format: int64 title: | Settled is this stratum's denominator: Met + NotMet + Inconclusive. type: integer source: type: string title: | DecisionOutcomeStratum is one comparable group of SETTLED effects. Every type: object types.DecisionTotals: properties: by_source: additionalProperties: format: int64 type: integer type: object by_status: additionalProperties: format: int64 type: integer description: | than any single number derived from it. title: > ByStatus and BySource are counts, not rates — a status distribution says more type: object decisions: description: | here is a share of. format: int64 title: > Decisions is every decision in the window, the denominator everything else type: integer legacy_unclassified: description: > is inside Decisions and excluded from every outcome stratum, and it is named here so the two numbers can be reconciled rather than leaving a reader to wonder where the difference went. format: int64 title: > LegacyUnclassified is the backfilled history that predates intent capture. It type: integer type: object types.DecisionScorecardWindow: properties: since: format: date-time type: string until: format: date-time type: string type: object responses: APIError: content: application/json: schema: externalDocs: url: https://pkg.go.dev/encore.dev/beta/errs#Error properties: code: description: Error code example: not_found externalDocs: url: https://pkg.go.dev/encore.dev/beta/errs#ErrCode type: string details: description: Error details type: object message: description: Error message type: string title: APIError type: object description: Error response securitySchemes: bearerAuth: type: http scheme: bearer description: An Erdo API key (erdo_api_...) or scoped token (erdo_token_...). ````