@clossys/influencer

Influencer operates a declared organization or product presence and improves independently verified qualified audience response.

npm install @clossys/influencer

This package is published to the public npm registry, https://registry.npmjs.org. Installing it needs no authentication: no npm token, no .npmrc registry override, and no GitHub credential of any kind.

Qualified response yield per thousand

Independent consumer evidence shows the position's owned metric meets its setpoint over the declared review cadence. The owned metric is qualified response yield per thousand, computed by assessQualifiedResponseYieldPerThousand(). An empty eligible-exposure set is indeterminate, never a perfect yield. checkResponseYield remains the kebab-metric gate; it is not this assessment. This package does not measure consumer evidence and does not close the loop. A green run of this package's tests is not a close.

import { assessQualifiedResponseYieldPerThousand } from "@clossys/influencer";

const report = assessQualifiedResponseYieldPerThousand(input);

influencer-rate-check assessment.json

The command prints JSON and exits 0 for satisfied, 1 for violated, and 2 for indeterminate, unreadable, or invalid input.

This package declares that command as its first-day assessment surface in its own manifest:

"foundry": { "assessment": { "bin": "influencer-rate-check", "invocation": "single-json-input" } }

Onboarding discovers that declaration from the installed manifest and never infers a surface. influencer-check remains the two-argument CLI and is not the assessment surface. Influencer is not a required first-day role; Advisor remains the only required first-day assessment.

The durable job question is:

Is this governed outbound presence producing qualified audience responses at the declared rate?

The package does not manufacture a persona, impersonate a person, generate or approve content, hold credentials, or define what a qualified response means. PresenceSubjectKind contains only organization and product, and every installation must set impersonationProhibited: true.

Metric and loop

Influencer's canonical primary loop mode is optimize and its secondary loop mode is fulfill. Optimize owns the content/channel/cadence learning cycle; fulfill is the internal action mechanism that makes one bounded presence intent real.

qualified response yield per thousand
= 1,000 × independently observed qualified responses
  / independently observed eligible exposures

The unit is qualified responses per 1,000 exposures and the desired direction is up. Each response must have a unique event id, match a consumer-declared qualified action kind, join exactly to its experiment/content/publication, and occur inside the declared measurement window. A readable response source that found no events is a measured zero. No due window, insufficient exposure, an unreadable source, or unjoinable evidence is indeterminate—not a clean result.

The five universal stages are concrete here:

1. Sense: read the installed strategy/audience revision, approved content candidates, channel readiness, cadence history, publication join keys, and independent exposure/response evidence.
2. Judge: validate the current intent-bound authority and zero-spend guardrail, then apply the consumer's decision policy to a bounded audience/channel/content/cadence hypothesis.
3. Act: atomically claim and execute configure-presence, publish, or reply through the injected PresenceActuator; replies require an opaque inbound-admission evidence id.
4. Verify: durably complete the action, re-read external outcomes, and compute response yield against the installed setpoint.
5. Learn or escalate: persist a superseding experiment, close, pause, or hand off. The package may propose upstream changes but never rewrites strategy, copy, design, or authority by itself.

Consumer binding

validatePresenceInstallation() requires the values that make the role an installed position rather than an empty mission document:

• business metric node and causal hypothesis;
• organization/product subject, audience, strategy revision, channels, and channel-readiness evidence;
• qualified action vocabulary, setpoint, sample floor, attribution window, evaluation cadence, and exposure/response evidence sources;
• allowed actions and current authority source;
• content-policy, cadence, and anti-impersonation guardrails;
• durable experiment, action, and outcome stores; and
• responsible owner and escalation route.

paidSpendCeiling is the literal value 0 in both PresenceInstallation and PresenceAuthorityEvidence. V1 cannot express, validate, or execute paid distribution. Paid media requires later requalification; no provider adapter or credential resolver ships here.

Action runtime

import {
  createInfluencer,
  type PresenceActionLedger,
  type PublishIntent,
} from "@clossys/influencer";

declare const ledger: PresenceActionLedger;

const influencer = createInfluencer({
  ledger,
  authority: async () => ({ state: "authorized" }),
  actuator: {
    async execute(intent) {
      // The consumer owns this provider integration and its credentials.
      return {
        provider: "configured-provider",
        remoteActionId: intent.id,
        observedAt: "2026-08-23T10:00:01.000Z",
      };
    },
  },
});

const intent: PublishIntent = {
  id: "intent-one",
  installationId: "presence-one",
  subjectId: "product-one",
  experimentId: "experiment-one",
  channelId: "channel-one",
  actionKind: "publish",
  requestedAt: "2026-08-23T10:00:00.000Z",
  authority: {
    id: "authority-one",
    intentId: "intent-one",
    subjectId: "product-one",
    actorId: "agent-one",
    humanOwnerId: "owner-one",
    allowedActions: ["publish"],
    channelIds: ["channel-one"],
    issuedAt: "2026-08-23T09:00:00.000Z",
    expiresAt: "2026-08-23T11:00:00.000Z",
    paidSpendCeiling: 0,
  },
  contentId: "approved-content-one",
  publicationId: "publication-one",
};

const result = await influencer.act(intent);

Claiming happens before the current authority read. A denial is durably completed as skipped. An explicit unverifiable result, a thrown authority read, or a malformed decision never invokes the actuator and leaves the lease uncompleted for host-governed reclamation. applied proves only that the injected actuator returned a valid observed receipt; it is not audience response evidence.

Response-yield gate

influencer-check response-yield response-evidence.json

The input is one JSON ResponseYieldInput: evaluatedAt, a positive setpointPerThousand, positive minimumExposureCount, non-empty qualifiedActionKinds, and non-empty records. Every record carries exact experiment/content/publication join keys, full intent-bound PresenceAuthorityEvidence, its window, and ternary exposures and responses reads.

Exit codes are 0 satisfied, 1 measured below setpoint, and 2 indeterminate or unable to evaluate. The pure checkResponseYield() produces the same judgment without file I/O; validateResponseYieldInput() keeps bad evidence separate from a measured violation.

Boundaries

• Strategist owns audience, positioning, governed claims, and the business causal hypothesis. Influencer references their revisions.
• Writer and Designer own language and visual quality. Influencer selects approved content; it does not approve its own work.
• Publisher resolves, renders, and records what shipped. Influencer selects channel/cadence experiments and consumes publication join evidence; it ships no renderer or publication ledger.
• Messenger owns directed finished-message transport. Public algorithmic exposure is not recipient delivery.
• Butler owns person-request admission and current wants. A reply must cite admitted inbound evidence rather than bypass that boundary.
• Observer remains an independent measurer. Influencer accepts host-owned evidence but cannot call its own action receipt an independent outcome.
• Authorization, credential custody, topology, and provider provisioning remain outside this package and enter through explicit consumer ports and evidence.

API

| Export | Kind | Purpose | | --- | --- | --- | | InfluencerActionError | class | Normalized actuator failure | | InfluencerValidationError | class | Structured invalid-input error | | createInfluencer | function | Build claim, authority, action, and completion orchestration | | checkResponseYield | function | Compute the kebab metric and ternary judgment | | assessQualifiedResponseYieldPerThousand | function | Charter close metric from independent observations | | validateResponseYieldInput | function | Validate metric evidence | | QualifiedResponseYieldAssessment | type | Charter close report | | QualifiedResponseYieldFinding | type | Charter close finding | | QualifiedResponseYieldState | type | Satisfied, violated, or indeterminate | | PRESENCE_ACTION_KINDS | const | Closed v1 action vocabulary | | PRESENCE_SUBJECT_KINDS | const | Organization/product subject vocabulary | | assertValidPresenceActionIntent | function | Throw for invalid action authority or payload | | validatePresenceActionIntent | function | Validate one action intent | | validatePresenceExperiment | function | Validate one versioned experiment | | validatePresenceInstallation | function | Validate the complete consumer binding | | AudienceResponseEvent | type | Independently observed joined audience response | | CompletedPresenceActionResult | type | Durable applied, failed, or skipped result | | ConfigurePresenceIntent | type | Desired-presence revision action | | CountObservation | type | Ternary exposure evidence | | Influencer | type | Action runtime interface | | InfluencerConfig | type | Required authority, actuator, and ledger ports | | PresenceActionClaim | type | Opaque action lease | | PresenceActionDuplicate | type | Duplicate claim result | | PresenceActionFailure | type | Normalized action failure | | PresenceActionIntent | type | Closed v1 action-intent union | | PresenceActionKind | type | Configure, publish, or reply action | | PresenceActionLedger | type | Atomic claim and durable completion port | | PresenceActionReceipt | type | Injected actuator's observed receipt | | PresenceActionResult | type | Applied, failed, skipped, duplicate, or unverifiable result | | PresenceActuator | type | Consumer-owned provider action port | | PresenceAuthorityDecision | type | Authorized, denied, or unverifiable judgment | | PresenceAuthorityEvidence | type | Intent-bound zero-spend delegated authority | | PresenceAuthorityPolicy | type | Required current authority reader | | PresenceChannelBinding | type | Consumer channel/account/readiness binding | | PresenceExperiment | type | Versioned audience/channel/content/cadence hypothesis | | PresenceInstallation | type | Complete consumer-specific position binding | | PresenceSubjectKind | type | Organization or product only | | PublishIntent | type | Approved-content publication action | | ReplyIntent | type | Admitted-inbound reply action | | ResponseObservation | type | Ternary independent response evidence | | ResponseYieldInput | type | Gate input and metric setpoint | | ResponseYieldRecord | type | Governed publication evidence window | | ResponseYieldResult | type | Metric components and ternary result | | ValidationFinding | type | Structured validation finding |

Node 20 or newer. Zero runtime dependencies. MIT licensed.