@ai-sdlc/orchestrator 0.9.0 → 0.13.0

package/dist/signal-ingestion/d1.d.ts

@@ -0,0 +1,252 @@
+/**
+ * RFC-0030 Phase 5 — D1 formula reformulation + RFC-0008 PPA integration.
+ *
+ * Reformulates the existing PPA D1 (Demand Pressure) input to consume
+ * cluster-level demand from the signal-ingestion pipeline per RFC-0030 §10:
+ *
+ * ```
+ * D1(cluster) = Σ over signals in cluster:
+ *     signal.baseWeight              # 1.0 Tier 1; 0.3 Tier 2 above threshold; 0 below
+ *     × signal.tierMultiplier        # configurable per deployment (§6.1)
+ *     × signal.icpResonance          # configurable per deployment (§6.2)
+ *     × signal.recencyDecay          # exp(-age_days × ln(2) / half_life_days)
+ *     × cluster.saResonance          # filter per §9 (applied via the SA bucket multiplier)
+ *
+ * D1 is then normalized across all active clusters to [0, 1] and fed into the
+ * existing PPA D formula (PPA v1.1 §3.1).
+ * ```
+ *
+ * **Non-replacement contract (AC #2 + AC #4)**: human-authored backlog items
+ * continue to feed D1 alongside signal-pipeline-derived demand. The pipeline
+ * adds a parallel input path; the existing path is preserved. When the
+ * pipeline is disabled (`config.enabled === false`, the default), this module
+ * contributes zero signal-pipeline demand — the composed D1 reduces to the
+ * backlog-item input unchanged. When both inputs are active, weights from
+ * `config.d1Composition` blend them (default 50/50).
+ *
+ * **RFC-0008 PPA Triad integration (AC #3)**: the resulting per-item composed
+ * D1 score is consumed by the `enrichDemandSignalFromClusters()` helper which
+ * overlays it onto a `PriorityInput.demandSignal` field — the same field the
+ * admission composite (`admission-composite.ts`) reads via `mapIssueToPriorityInput`
+ * for the Sα₁ × Dπ_adjusted × Eρ₅ admission gate per RFC-0008 §A.6.
+ *
+ * **Cluster-to-item routing**: the pipeline produces cluster-level demand;
+ * admission scoring is per-item (per backlog task / GitHub issue). The
+ * composition treats the pipeline-derived signal as a *deployment-level
+ * demand floor* that lifts the demand signal for items whose subject
+ * matches an active cluster theme (cluster-to-item matching is left to a
+ * caller-supplied lookup in v1; v2 will wire to RFC-0024 capture matching).
+ *
+ * @module signal-ingestion/d1
+ */
+import type { SignalIngestionConfig } from './config.js';
+import type { QuarantineStore, SignificanceAssessedCluster } from './significance.js';
+/**
+ * Options for `computeClusterD1()` / `aggregateD1FromClusters()` per
+ * AISDLC-433. The optional `quarantineStore` lets the D1 path consult the
+ * z-score detector's quarantine state and EXCLUDE quarantined signals from
+ * cluster scoring (AC #6). Back-compat: when omitted, no signals are treated
+ * as quarantined — pre-AISDLC-433 callers continue to work unchanged.
+ */
+export interface ComputeClusterD1Options {
+    quarantineStore?: QuarantineStore;
+    /** Clock override for quarantine expiry checks. Defaults to `new Date()`. */
+    asOf?: Date;
+}
+/**
+ * Per-cluster D1 score breakdown — produced by `computeClusterD1()`.
+ *
+ * `rawScore` is the §10 formula output BEFORE normalisation (sum across
+ * cluster members × the SA bucket multiplier). `normalizedScore` is filled
+ * in by `aggregateD1FromClusters()` after the full population is known.
+ */
+export interface ClusterD1Score {
+    clusterId: string;
+    /** Sum of per-signal weights × cluster SA multiplier; pre-normalisation. */
+    rawScore: number;
+    /**
+     * Filled by `aggregateD1FromClusters()` — `rawScore / max(rawScore across
+     * eligible clusters)`. Range [0, 1]. `0` when the cluster is ineligible
+     * (significance gate failed or SA bucket excludes it).
+     */
+    normalizedScore: number;
+    /** Whether the cluster contributes to D1 at all. Mirrors `eligibleForD1`. */
+    eligible: boolean;
+    /** Cluster size (member count) — useful for downstream auditing. */
+    signalCount: number;
+}
+/**
+ * Result of aggregating per-cluster D1 scores into a deployment-level
+ * pipeline-derived demand input.
+ */
+export interface AggregatedD1Result {
+    /** Per-cluster scores, with normalised values populated. */
+    clusters: ClusterD1Score[];
+    /**
+     * Maximum raw score across eligible clusters (the normalisation denominator).
+     * `0` when no clusters are eligible.
+     */
+    maxRawScore: number;
+    /**
+     * Population-level summary — the mean normalised score across eligible
+     * clusters. Range [0, 1]. Useful as a deployment-level "ambient pipeline
+     * demand" indicator when no per-item cluster match exists.
+     */
+    meanNormalizedScore: number;
+}
+/**
+ * Inputs to `composeD1Inputs()` per RFC-0030 §10 non-replacement contract.
+ */
+export interface ComposeD1InputsArgs {
+    /**
+     * Pipeline-derived per-item demand score (post-`enrichDemandSignalFromClusters`
+     * matching). Range expected [0, 1]. `undefined` when no cluster match exists
+     * — composition then falls back to `backlogItemD1` alone.
+     */
+    signalPipelineD1?: number;
+    /**
+     * Human-authored backlog-item demand score from the existing PPA path
+     * (`mapIssueToPriorityInput().demandSignal`). Range [0, 1]. `undefined`
+     * when the backlog has no demand for the item (rare; treated as 0).
+     */
+    backlogItemD1?: number;
+    /**
+     * Signal-ingestion config — `d1Composition` weights + `enabled` flag drive
+     * the blend. Defaults to `DEFAULT_SIGNAL_INGESTION_CONFIG` (pipeline disabled).
+     */
+    config?: SignalIngestionConfig;
+}
+/** Result of composing the two D1 input streams. */
+export interface ComposedD1Result {
+    /** The final blended D1 input. Range [0, 1]. */
+    composedScore: number;
+    /** Normalised weight actually applied to the pipeline input. */
+    signalPipelineWeightApplied: number;
+    /** Normalised weight actually applied to the backlog input. */
+    backlogItemWeightApplied: number;
+    /**
+     * Audit trail — what each branch contributed before weighting:
+     *   - `signalPipelineContribution` = `signalPipelineD1 × signalPipelineWeightApplied`
+     *   - `backlogItemContribution` = `backlogItemD1 × backlogItemWeightApplied`
+     */
+    signalPipelineContribution: number;
+    backlogItemContribution: number;
+    /**
+     * Backward-compat indicator — `true` when the pipeline was disabled OR no
+     * pipeline-derived signal was provided, so the result is the backlog input
+     * unchanged. AC #4 audit hook.
+     */
+    pipelineBypass: boolean;
+}
+/**
+ * Apply RFC-0030 §10 D1 formula to a single significance-assessed cluster.
+ *
+ * `rawScore` is the sum of per-member weights × the cluster's SA bucket
+ * multiplier (§10 line 5 — the SA filter applied at cluster level).
+ *
+ * When the cluster is ineligible (`assessment.eligibleForD1 === false` —
+ * Tier 2 threshold failed OR SA bucket is `out-of-scope` / `pending`), the
+ * raw score is forced to 0 per §8 / §9 contract.
+ */
+export declare function computeClusterD1(assessment: SignificanceAssessedCluster, config?: SignalIngestionConfig, options?: ComputeClusterD1Options): ClusterD1Score;
+/**
+ * Aggregate per-cluster D1 scores into a deployment-level result with
+ * normalised scores per §10 final line ("D1 is normalized across all active
+ * clusters to [0, 1]").
+ *
+ * Normalisation divides each `rawScore` by the population max — keeps the
+ * eligible cluster with the strongest demand at 1.0 and orders the rest
+ * proportionally. When no clusters are eligible, all `normalizedScore` are 0
+ * and `meanNormalizedScore` is 0 — D1 cleanly degrades to "no pipeline-derived
+ * demand" without surfacing NaNs.
+ *
+ * `meanNormalizedScore` is the mean across **eligible** clusters only —
+ * ineligible clusters (rawScore = 0) don't pull the population mean down.
+ */
+export declare function aggregateD1FromClusters(assessments: SignificanceAssessedCluster[], config?: SignalIngestionConfig, options?: ComputeClusterD1Options): AggregatedD1Result;
+/**
+ * Blend signal-pipeline-derived demand with the existing backlog-item demand
+ * input per RFC-0030 §10 non-replacement contract.
+ *
+ * **Backward compatibility (AC #4)**: when `config.enabled === false` the
+ * function bypasses the pipeline entirely and returns `composedScore =
+ * backlogItemD1` (or 0 when absent). `pipelineBypass = true` so callers /
+ * tests can audit the bypass branch.
+ *
+ * **Weight balancing (AC #5)**: when both inputs are present and the pipeline
+ * is enabled, weights from `config.d1Composition` are normalised to sum to 1
+ * and applied as a linear blend. Negative weights are clamped to 0 by the
+ * config validator; if both weights are 0 (operator misconfiguration), the
+ * function falls back to the unweighted average to avoid emitting NaN.
+ *
+ * **Single-input cases**: when only one input is present, the other is
+ * treated as 0; weights still apply so an operator who configured
+ * `{signalPipelineWeight: 0.8, backlogItemWeight: 0.2}` and supplies only
+ * `backlogItemD1` gets `0.2 × backlogItemD1` (which is documented behaviour —
+ * the missing input is genuinely missing demand). If you want "use this one
+ * unweighted when the other is absent" semantics, omit the absent field AND
+ * set the other input's weight to 1.0.
+ */
+export declare function composeD1Inputs(args: ComposeD1InputsArgs): ComposedD1Result;
+/**
+ * Pluggable matcher: resolve a backlog work item to the most relevant cluster
+ * D1 score (if any). Callers supply this so the cluster-to-item match policy
+ * can be swapped per deployment without rewriting the core composition.
+ *
+ * v1 contract: the matcher receives a free-form `itemKey` (typically the
+ * task title or a normalised slug); returns the best-matching `ClusterD1Score`
+ * or `undefined` when no cluster is a meaningful match. The matcher is
+ * intentionally NOT given write access to the score — it only chooses which
+ * cluster the item maps onto. RFC-0024 capture matching wires this in v2.
+ */
+export interface ClusterMatcher {
+    (itemKey: string, aggregated: AggregatedD1Result): ClusterD1Score | undefined;
+}
+/**
+ * RFC-0008 PPA Triad integration helper — applies the composed D1 to a
+ * `PriorityInput` so the admission composite consumes pipeline-derived
+ * demand through the same `demandSignal` field it already reads.
+ *
+ * The returned object is a NEW `PriorityInput` (input is not mutated). When
+ * the pipeline is disabled OR no cluster match exists, the returned object
+ * carries the input's original `demandSignal` unchanged — backward-compatible
+ * (AC #4).
+ *
+ * Composition algorithm:
+ *   1. Resolve `signalPipelineD1` via `matcher(itemKey, aggregated)` →
+ *      `match?.normalizedScore` (range [0, 1]); when matcher returns
+ *      `undefined`, no pipeline contribution.
+ *   2. Read `backlogItemD1` from `priorityInput.demandSignal` (already the
+ *      output of `mapIssueToPriorityInput()` per existing path).
+ *   3. Compose via `composeD1Inputs()` with the configured weights.
+ *   4. Override `demandSignal` on a shallow clone of the input.
+ *
+ * The composed result and its audit fields are also returned so callers can
+ * surface the pipeline contribution in pillar-breakdown logs.
+ */
+export interface EnrichDemandSignalArgs<T extends {
+    demandSignal?: number;
+}> {
+    /** The PriorityInput-shaped object to enrich. NOT mutated. */
+    priorityInput: T;
+    /** A short key the matcher uses to find a cluster (title, slug, or item ID). */
+    itemKey: string;
+    /** The aggregation result from `aggregateD1FromClusters()`. */
+    aggregated: AggregatedD1Result;
+    /** Cluster-to-item matcher; see `ClusterMatcher` docstring. */
+    matcher: ClusterMatcher;
+    /** Signal ingestion config. Defaults to `DEFAULT_SIGNAL_INGESTION_CONFIG`. */
+    config?: SignalIngestionConfig;
+}
+export interface EnrichDemandSignalResult<T> {
+    /** The enriched PriorityInput (shallow clone with `demandSignal` overlaid). */
+    enriched: T;
+    /** Full composition result for audit / breakdown surfacing. */
+    composition: ComposedD1Result;
+    /** Which cluster matched, when any. `undefined` when matcher returned none. */
+    matchedCluster: ClusterD1Score | undefined;
+}
+export declare function enrichDemandSignalFromClusters<T extends {
+    demandSignal?: number;
+}>(args: EnrichDemandSignalArgs<T>): EnrichDemandSignalResult<T>;
+//# sourceMappingURL=d1.d.ts.map

package/dist/signal-ingestion/d1.js

@@ -0,0 +1,235 @@
+/**
+ * RFC-0030 Phase 5 — D1 formula reformulation + RFC-0008 PPA integration.
+ *
+ * Reformulates the existing PPA D1 (Demand Pressure) input to consume
+ * cluster-level demand from the signal-ingestion pipeline per RFC-0030 §10:
+ *
+ * ```
+ * D1(cluster) = Σ over signals in cluster:
+ *     signal.baseWeight              # 1.0 Tier 1; 0.3 Tier 2 above threshold; 0 below
+ *     × signal.tierMultiplier        # configurable per deployment (§6.1)
+ *     × signal.icpResonance          # configurable per deployment (§6.2)
+ *     × signal.recencyDecay          # exp(-age_days × ln(2) / half_life_days)
+ *     × cluster.saResonance          # filter per §9 (applied via the SA bucket multiplier)
+ *
+ * D1 is then normalized across all active clusters to [0, 1] and fed into the
+ * existing PPA D formula (PPA v1.1 §3.1).
+ * ```
+ *
+ * **Non-replacement contract (AC #2 + AC #4)**: human-authored backlog items
+ * continue to feed D1 alongside signal-pipeline-derived demand. The pipeline
+ * adds a parallel input path; the existing path is preserved. When the
+ * pipeline is disabled (`config.enabled === false`, the default), this module
+ * contributes zero signal-pipeline demand — the composed D1 reduces to the
+ * backlog-item input unchanged. When both inputs are active, weights from
+ * `config.d1Composition` blend them (default 50/50).
+ *
+ * **RFC-0008 PPA Triad integration (AC #3)**: the resulting per-item composed
+ * D1 score is consumed by the `enrichDemandSignalFromClusters()` helper which
+ * overlays it onto a `PriorityInput.demandSignal` field — the same field the
+ * admission composite (`admission-composite.ts`) reads via `mapIssueToPriorityInput`
+ * for the Sα₁ × Dπ_adjusted × Eρ₅ admission gate per RFC-0008 §A.6.
+ *
+ * **Cluster-to-item routing**: the pipeline produces cluster-level demand;
+ * admission scoring is per-item (per backlog task / GitHub issue). The
+ * composition treats the pipeline-derived signal as a *deployment-level
+ * demand floor* that lifts the demand signal for items whose subject
+ * matches an active cluster theme (cluster-to-item matching is left to a
+ * caller-supplied lookup in v1; v2 will wire to RFC-0024 capture matching).
+ *
+ * @module signal-ingestion/d1
+ */
+import { DEFAULT_SIGNAL_INGESTION_CONFIG } from './config.js';
+import { isSignalQuarantined } from './significance.js';
+// ── §10 formula — per cluster ───────────────────────────────────────────────
+/**
+ * Compute the per-signal weight from a cluster member's `ClusteredSignalInput`
+ * + config-derived multipliers. Mirrors `classifier.computeSignalWeight()` but
+ * operates on the Phase-3 cluster-member shape (which dropped the
+ * `tierMultiplier` / `icpResonanceWeight` convenience fields the classifier
+ * carried for Phase-4 use).
+ *
+ * Order: `baseWeight × tierMultiplier × icpResonanceWeight × recencyDecay`
+ * exactly per §10 lines 1-4 of the formula. The cluster-level
+ * `saResonance` factor (§10 line 5) is applied OUTSIDE this function, in
+ * `computeClusterD1()`, since it's a cluster-wide property.
+ */
+function computeClusterMemberWeight(member, config) {
+    // Tier 1 sources base weight = 1.0; Tier 2 = 0.3 (matches classifier convention
+    // — Phase 4 significance gate independently zeros out below-threshold clusters
+    // via SA_WEIGHT_MULTIPLIERS, so we don't double-discount here).
+    const adapterTier = member.adapterTier ?? 1;
+    const baseWeight = adapterTier === 1 ? 1.0 : 0.3;
+    const tierMultiplier = config.tierMultipliers[member.customerTier];
+    const icpWeight = config.icpResonanceWeights[member.icpResonance];
+    return baseWeight * tierMultiplier * icpWeight * member.recencyDecay;
+}
+/**
+ * Apply RFC-0030 §10 D1 formula to a single significance-assessed cluster.
+ *
+ * `rawScore` is the sum of per-member weights × the cluster's SA bucket
+ * multiplier (§10 line 5 — the SA filter applied at cluster level).
+ *
+ * When the cluster is ineligible (`assessment.eligibleForD1 === false` —
+ * Tier 2 threshold failed OR SA bucket is `out-of-scope` / `pending`), the
+ * raw score is forced to 0 per §8 / §9 contract.
+ */
+export function computeClusterD1(assessment, config = DEFAULT_SIGNAL_INGESTION_CONFIG, options = {}) {
+    if (!assessment.eligibleForD1) {
+        return {
+            clusterId: assessment.cluster.clusterId,
+            rawScore: 0,
+            normalizedScore: 0,
+            eligible: false,
+            signalCount: assessment.cluster.signalCount,
+        };
+    }
+    // AC #6: signals currently in quarantine (flooding-flagged with active
+    // expiresAt) are excluded from the D1(cluster) formula. The exclusion is
+    // member-level — a cluster mixing quarantined + clean signals still
+    // contributes the clean members' weight; clusters whose entire membership
+    // is quarantined effectively score 0 even though they're "eligible" per
+    // significance + SA.
+    const asOf = options.asOf ?? new Date();
+    const memberSum = assessment.cluster.members.reduce((acc, member) => {
+        if (isSignalQuarantined(member.signal, options.quarantineStore, asOf))
+            return acc;
+        return acc + computeClusterMemberWeight(member, config);
+    }, 0);
+    // `assessment.d1WeightMultiplier` is the precomputed combined multiplier
+    // from Phase 4: `significanceMultiplier × SA_WEIGHT_MULTIPLIERS[bucket]`.
+    // For any eligible cluster, significanceMultiplier === 1.0 (qualified) and
+    // the bucket is one of `full` / `discounted` / `low-sa-review`, so this
+    // value is simply the §10 line-5 SA factor (range 0.3..1.0). Multiplying
+    // through it preserves the spec algebra `members × cluster.saResonance`
+    // while letting the Phase 4 single source of truth drive the math.
+    const rawScore = memberSum * assessment.d1WeightMultiplier;
+    return {
+        clusterId: assessment.cluster.clusterId,
+        rawScore,
+        normalizedScore: 0, // filled by aggregateD1FromClusters
+        eligible: true,
+        signalCount: assessment.cluster.signalCount,
+    };
+}
+/**
+ * Aggregate per-cluster D1 scores into a deployment-level result with
+ * normalised scores per §10 final line ("D1 is normalized across all active
+ * clusters to [0, 1]").
+ *
+ * Normalisation divides each `rawScore` by the population max — keeps the
+ * eligible cluster with the strongest demand at 1.0 and orders the rest
+ * proportionally. When no clusters are eligible, all `normalizedScore` are 0
+ * and `meanNormalizedScore` is 0 — D1 cleanly degrades to "no pipeline-derived
+ * demand" without surfacing NaNs.
+ *
+ * `meanNormalizedScore` is the mean across **eligible** clusters only —
+ * ineligible clusters (rawScore = 0) don't pull the population mean down.
+ */
+export function aggregateD1FromClusters(assessments, config = DEFAULT_SIGNAL_INGESTION_CONFIG, options = {}) {
+    const perCluster = assessments.map((a) => computeClusterD1(a, config, options));
+    const eligibleScores = perCluster.filter((c) => c.eligible);
+    const maxRawScore = eligibleScores.reduce((max, c) => Math.max(max, c.rawScore), 0);
+    // Populate normalisedScore in-place. When maxRawScore is 0 (no eligible
+    // clusters OR all eligible clusters had zero weight — possible if every
+    // member has icpResonance × tierMultiplier × recencyDecay = 0), normalised
+    // scores stay at 0 by short-circuiting.
+    for (const c of perCluster) {
+        c.normalizedScore = maxRawScore > 0 && c.eligible ? c.rawScore / maxRawScore : 0;
+    }
+    const meanNormalizedScore = eligibleScores.length > 0
+        ? eligibleScores.reduce((sum, c) => sum + c.normalizedScore, 0) / eligibleScores.length
+        : 0;
+    return {
+        clusters: perCluster,
+        maxRawScore,
+        meanNormalizedScore,
+    };
+}
+// ── Non-replacement composition (§10.4 spirit; AC #2, #4, #5) ───────────────
+/**
+ * Blend signal-pipeline-derived demand with the existing backlog-item demand
+ * input per RFC-0030 §10 non-replacement contract.
+ *
+ * **Backward compatibility (AC #4)**: when `config.enabled === false` the
+ * function bypasses the pipeline entirely and returns `composedScore =
+ * backlogItemD1` (or 0 when absent). `pipelineBypass = true` so callers /
+ * tests can audit the bypass branch.
+ *
+ * **Weight balancing (AC #5)**: when both inputs are present and the pipeline
+ * is enabled, weights from `config.d1Composition` are normalised to sum to 1
+ * and applied as a linear blend. Negative weights are clamped to 0 by the
+ * config validator; if both weights are 0 (operator misconfiguration), the
+ * function falls back to the unweighted average to avoid emitting NaN.
+ *
+ * **Single-input cases**: when only one input is present, the other is
+ * treated as 0; weights still apply so an operator who configured
+ * `{signalPipelineWeight: 0.8, backlogItemWeight: 0.2}` and supplies only
+ * `backlogItemD1` gets `0.2 × backlogItemD1` (which is documented behaviour —
+ * the missing input is genuinely missing demand). If you want "use this one
+ * unweighted when the other is absent" semantics, omit the absent field AND
+ * set the other input's weight to 1.0.
+ */
+export function composeD1Inputs(args) {
+    const config = args.config ?? DEFAULT_SIGNAL_INGESTION_CONFIG;
+    const backlog = args.backlogItemD1 ?? 0;
+    const pipeline = args.signalPipelineD1 ?? 0;
+    // Backward-compat fast path: pipeline disabled OR no pipeline input.
+    if (!config.enabled || args.signalPipelineD1 === undefined) {
+        return {
+            composedScore: clamp01(backlog),
+            signalPipelineWeightApplied: 0,
+            backlogItemWeightApplied: 1,
+            signalPipelineContribution: 0,
+            backlogItemContribution: clamp01(backlog),
+            pipelineBypass: true,
+        };
+    }
+    const { signalPipelineWeightApplied, backlogItemWeightApplied } = normaliseWeights(config.d1Composition);
+    const pipelineContribution = pipeline * signalPipelineWeightApplied;
+    const backlogContribution = backlog * backlogItemWeightApplied;
+    const composedScore = clamp01(pipelineContribution + backlogContribution);
+    return {
+        composedScore,
+        signalPipelineWeightApplied,
+        backlogItemWeightApplied,
+        signalPipelineContribution: pipelineContribution,
+        backlogItemContribution: backlogContribution,
+        pipelineBypass: false,
+    };
+}
+/**
+ * Normalise the configured weights so they sum to 1. When both are 0
+ * (degenerate operator config), fall back to a 50/50 split so the function
+ * never returns NaN downstream.
+ */
+function normaliseWeights(weights) {
+    const sum = weights.signalPipelineWeight + weights.backlogItemWeight;
+    if (sum <= 0) {
+        return { signalPipelineWeightApplied: 0.5, backlogItemWeightApplied: 0.5 };
+    }
+    return {
+        signalPipelineWeightApplied: weights.signalPipelineWeight / sum,
+        backlogItemWeightApplied: weights.backlogItemWeight / sum,
+    };
+}
+function clamp01(n) {
+    if (!Number.isFinite(n) || n <= 0)
+        return 0;
+    if (n >= 1)
+        return 1;
+    return n;
+}
+export function enrichDemandSignalFromClusters(args) {
+    const config = args.config ?? DEFAULT_SIGNAL_INGESTION_CONFIG;
+    const matchedCluster = args.matcher(args.itemKey, args.aggregated);
+    const signalPipelineD1 = matchedCluster?.normalizedScore;
+    const backlogItemD1 = args.priorityInput.demandSignal;
+    const composition = composeD1Inputs({ signalPipelineD1, backlogItemD1, config });
+    const enriched = {
+        ...args.priorityInput,
+        demandSignal: composition.composedScore,
+    };
+    return { enriched, composition, matchedCluster };
+}
+//# sourceMappingURL=d1.js.map

package/dist/signal-ingestion/errors.d.ts

@@ -0,0 +1,73 @@
+import type { SignalSourceName } from './types.js';
+export declare class UnknownSignalSource extends Error {
+    readonly source: string;
+    constructor(source: string);
+}
+export declare class SignalSourceUnavailable extends Error {
+    readonly source: SignalSourceName;
+    constructor(source: SignalSourceName, message?: string);
+}
+export declare class AdapterCredentialInvalid extends Error {
+    readonly source: SignalSourceName;
+    constructor(source: SignalSourceName, message?: string);
+}
+/**
+ * RFC-0030 OQ-13.1 v0.3 re-walkthrough: distinct from `AdapterCredentialInvalid`.
+ *
+ * Raised when an adapter probes its environment for the required credential
+ * env var (e.g. `SIGNAL_ZENDESK_PAT`) and the var is missing or empty.
+ * Downstream: emit `Decision: adapter-credential-not-configured` and surface
+ * an operator SETUP task ("create + set this env var"). Pipeline continues
+ * with the remaining valid adapters (AC #4).
+ */
+export declare class AdapterCredentialNotConfigured extends Error {
+    readonly source: SignalSourceName;
+    readonly envVarName: string;
+    constructor(source: SignalSourceName, envVarName: string, message?: string);
+}
+/**
+ * RFC-0030 OQ-13.1 v0.3 re-walkthrough: distinct from `AdapterCredentialInvalid`.
+ *
+ * Raised when an adapter has the credential env var present BUT the auth call
+ * against the upstream service is rejected (401 / 403). Downstream: emit
+ * `Decision: adapter-credential-rejected` and surface an operator ROTATION
+ * task ("rotate / re-generate this credential"). Pipeline continues with the
+ * remaining valid adapters (AC #4).
+ */
+export declare class AdapterCredentialRejected extends Error {
+    readonly source: SignalSourceName;
+    constructor(source: SignalSourceName, message?: string);
+}
+/**
+ * RFC-0030 OQ-13.1 v0.3 re-walkthrough: refuse OAuth-required adapters at
+ * registration. Phase 1 ships ENV-VAR-ONLY adapters; OAuth-required adapters
+ * (full Salesforce / HubSpot integrations, Zendesk-with-OAuth-scopes) wait
+ * for the future credential-management RFC. Adapters that declare
+ * `requiresOAuth = true` cannot register; the registry returns a Decision
+ * pointing the operator at the pending RFC.
+ */
+export declare class AdapterRequiresCredentialMgmtRfc extends Error {
+    readonly source: SignalSourceName;
+    constructor(source: SignalSourceName, message?: string);
+}
+export declare class ManualSignalIncomplete extends Error {
+    readonly sourceId?: string | undefined;
+    readonly source: "signal-source-manual";
+    constructor(sourceId?: string | undefined, message?: string);
+}
+/**
+ * RFC-0030 OQ-13.4 v0.3 re-walkthrough: per-operator rate-limit at signal
+ * entry. When a manual signal would exceed `manualEntry.dailyCapPerOperator`
+ * for the attested operator (in UTC-day buckets), the adapter raises this
+ * error. Downstream: emit `Decision: manual-signal-rate-limit-exceeded` and
+ * surface an operator BATCH-REVIEW escalation path.
+ */
+export declare class ManualSignalRateLimitExceeded extends Error {
+    readonly attestedBy: string;
+    readonly dailyCap: number;
+    readonly utcDate: string;
+    readonly sourceId?: string | undefined;
+    readonly source: "signal-source-manual";
+    constructor(attestedBy: string, dailyCap: number, utcDate: string, sourceId?: string | undefined, message?: string);
+}
+//# sourceMappingURL=errors.d.ts.map

package/dist/signal-ingestion/errors.js

@@ -0,0 +1,108 @@
+export class UnknownSignalSource extends Error {
+    source;
+    constructor(source) {
+        super(`Unknown signal source adapter: ${source}`);
+        this.source = source;
+        this.name = 'UnknownSignalSource';
+    }
+}
+export class SignalSourceUnavailable extends Error {
+    source;
+    constructor(source, message = `Signal source adapter unavailable: ${source}`) {
+        super(message);
+        this.source = source;
+        this.name = 'SignalSourceUnavailable';
+    }
+}
+export class AdapterCredentialInvalid extends Error {
+    source;
+    constructor(source, message = `Signal source adapter credentials invalid: ${source}`) {
+        super(message);
+        this.source = source;
+        this.name = 'AdapterCredentialInvalid';
+    }
+}
+/**
+ * RFC-0030 OQ-13.1 v0.3 re-walkthrough: distinct from `AdapterCredentialInvalid`.
+ *
+ * Raised when an adapter probes its environment for the required credential
+ * env var (e.g. `SIGNAL_ZENDESK_PAT`) and the var is missing or empty.
+ * Downstream: emit `Decision: adapter-credential-not-configured` and surface
+ * an operator SETUP task ("create + set this env var"). Pipeline continues
+ * with the remaining valid adapters (AC #4).
+ */
+export class AdapterCredentialNotConfigured extends Error {
+    source;
+    envVarName;
+    constructor(source, envVarName, message = `Signal source adapter credential not configured: ${source} (missing env var ${envVarName})`) {
+        super(message);
+        this.source = source;
+        this.envVarName = envVarName;
+        this.name = 'AdapterCredentialNotConfigured';
+    }
+}
+/**
+ * RFC-0030 OQ-13.1 v0.3 re-walkthrough: distinct from `AdapterCredentialInvalid`.
+ *
+ * Raised when an adapter has the credential env var present BUT the auth call
+ * against the upstream service is rejected (401 / 403). Downstream: emit
+ * `Decision: adapter-credential-rejected` and surface an operator ROTATION
+ * task ("rotate / re-generate this credential"). Pipeline continues with the
+ * remaining valid adapters (AC #4).
+ */
+export class AdapterCredentialRejected extends Error {
+    source;
+    constructor(source, message = `Signal source adapter credentials rejected by upstream: ${source}`) {
+        super(message);
+        this.source = source;
+        this.name = 'AdapterCredentialRejected';
+    }
+}
+/**
+ * RFC-0030 OQ-13.1 v0.3 re-walkthrough: refuse OAuth-required adapters at
+ * registration. Phase 1 ships ENV-VAR-ONLY adapters; OAuth-required adapters
+ * (full Salesforce / HubSpot integrations, Zendesk-with-OAuth-scopes) wait
+ * for the future credential-management RFC. Adapters that declare
+ * `requiresOAuth = true` cannot register; the registry returns a Decision
+ * pointing the operator at the pending RFC.
+ */
+export class AdapterRequiresCredentialMgmtRfc extends Error {
+    source;
+    constructor(source, message = `Adapter ${source} requires OAuth credential management; deferred to future RFC`) {
+        super(message);
+        this.source = source;
+        this.name = 'AdapterRequiresCredentialMgmtRfc';
+    }
+}
+export class ManualSignalIncomplete extends Error {
+    sourceId;
+    source = 'signal-source-manual';
+    constructor(sourceId, message = 'manual signal missing required attestation fields') {
+        super(message);
+        this.sourceId = sourceId;
+        this.name = 'ManualSignalIncomplete';
+    }
+}
+/**
+ * RFC-0030 OQ-13.4 v0.3 re-walkthrough: per-operator rate-limit at signal
+ * entry. When a manual signal would exceed `manualEntry.dailyCapPerOperator`
+ * for the attested operator (in UTC-day buckets), the adapter raises this
+ * error. Downstream: emit `Decision: manual-signal-rate-limit-exceeded` and
+ * surface an operator BATCH-REVIEW escalation path.
+ */
+export class ManualSignalRateLimitExceeded extends Error {
+    attestedBy;
+    dailyCap;
+    utcDate;
+    sourceId;
+    source = 'signal-source-manual';
+    constructor(attestedBy, dailyCap, utcDate, sourceId, message = `Manual signal rate limit exceeded for ${attestedBy} on ${utcDate} (cap: ${dailyCap}/day)`) {
+        super(message);
+        this.attestedBy = attestedBy;
+        this.dailyCap = dailyCap;
+        this.utcDate = utcDate;
+        this.sourceId = sourceId;
+        this.name = 'ManualSignalRateLimitExceeded';
+    }
+}
+//# sourceMappingURL=errors.js.map