learning

IFeedbackIntegration.getStats

Implementation of

onOutcomeProcessed()

onOutcomeProcessed(callback): () => void;

Defined in: packages/nexus-agents/src/learning/feedback-integration.ts:354

Subscribe to outcome processed events

Parameters

callback

OutcomeProcessedCallback

Returns

() => void

Implementation of

IFeedbackIntegration.onOutcomeProcessed

recordOutcome()

recordOutcome(params): void;

Defined in: packages/nexus-agents/src/learning/feedback-integration.ts:264

Record a generic task outcome

Parameters

params

RecordOutcomeParams

Returns

void

Implementation of

IFeedbackIntegration.recordOutcome

recordRoutingDecision()

recordRoutingDecision(
   decision, 
   traceId?, 
   options?
): string;

Defined in: packages/nexus-agents/src/learning/feedback-integration.ts:188

Record a routing decision; omitted task text skips preference training.

Parameters

decision

CompositeRoutingDecision

traceId?

string

options?

query?

string

Returns

string

Implementation of

IFeedbackIntegration.recordRoutingDecision

recordStepOutcome()

recordStepOutcome(
   routingDecisionId, 
   stepResult, 
   durationMs, 
   tokenUsage
): void;

Defined in: packages/nexus-agents/src/learning/feedback-integration.ts:246

Record a step outcome from workflow execution

Parameters

routingDecisionId

string

stepResult

StepResult

durationMs

number

tokenUsage

number

Returns

void

Implementation of

IFeedbackIntegration.recordStepOutcome

registerCompositeRouter()

registerCompositeRouter(router): void;

Defined in: packages/nexus-agents/src/learning/feedback-integration.ts:358

Parameters

router

ICompositeRouter

Returns

void

Implementation of

IFeedbackIntegration.registerCompositeRouter

reset()

reset(): void;

Defined in: packages/nexus-agents/src/learning/feedback-integration.ts:363

Reset all collected data

Returns

void

Implementation of

IFeedbackIntegration.reset

OutcomeFeedbackCollector

Defined in: packages/nexus-agents/src/learning/outcome-feedback.ts:63

Learning exports - Closed-loop feedback and routing improvement Split from index.ts for file size compliance (Issue #285)

Implements

IOutcomeFeedback

Constructors

Constructor

new OutcomeFeedbackCollector(config?): OutcomeFeedbackCollector;

Defined in: packages/nexus-agents/src/learning/outcome-feedback.ts:74

Parameters

config?

Partial<FeedbackCollectorConfig>

Returns

OutcomeFeedbackCollector

Methods

clearExpiredDecisions()

clearExpiredDecisions(): number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback.ts:268

Clear expired pending decisions.

Returns

number

Implementation of

IOutcomeFeedback.clearExpiredDecisions

computeReward()

computeReward(outcome): ComputedReward;

Defined in: packages/nexus-agents/src/learning/outcome-feedback.ts:189

Compute reward from an outcome.

Parameters

outcome

Returns

IOutcomeFeedback.computeReward

Implementation of

getPendingDecisions()

getPendingDecisions(): readonly FeedbackRoutingDecision[];

Defined in: packages/nexus-agents/src/learning/outcome-feedback.ts:264

Get pending decisions (waiting for outcomes).

Returns

readonly FeedbackRoutingDecision[]

Implementation of

IOutcomeFeedback.getPendingDecisions

getStats()

getStats(): FeedbackLoopStats;

Defined in: packages/nexus-agents/src/learning/outcome-feedback.ts:246

Get feedback loop statistics.

Returns

IOutcomeFeedback.getStats

Implementation of

onOutcomeProcessed()

onOutcomeProcessed(callback): () => void;

Defined in: packages/nexus-agents/src/learning/outcome-feedback.ts:105

Subscribe to outcome processed events.

Parameters

callback

OutcomeProcessedCallback

Returns

() => void

processOutcome()

processOutcome(traceId, partialOutcome): void;

Defined in: packages/nexus-agents/src/learning/outcome-feedback.ts:173

Process outcome for a trace ID (finds matching decision and computes reward).

Parameters

traceId

string

partialOutcome

Omit<TaskOutcome, "routingDecisionId">

Returns

void

Implementation of

IOutcomeFeedback.processOutcome

recordOutcome()

recordOutcome(outcome): void;

Defined in: packages/nexus-agents/src/learning/outcome-feedback.ts:152

Record an outcome for a routing decision.

Parameters

outcome

IOutcomeFeedback.recordOutcome

Returns

void

Implementation of

recordRoutingDecision()

recordRoutingDecision(decision): void;

Defined in: packages/nexus-agents/src/learning/outcome-feedback.ts:112

Record a routing decision for tracking.

Parameters

decision

IOutcomeFeedback.recordRoutingDecision

Returns

void

Implementation of

registerLinUCBBandit()

registerLinUCBBandit(bandit): void;

Defined in: packages/nexus-agents/src/learning/outcome-feedback.ts:89

Parameters

bandit

LinUCBBandit

Returns

void

registerPreferenceRouter()

registerPreferenceRouter(router): void;

Defined in: packages/nexus-agents/src/learning/outcome-feedback.ts:97

Parameters

router

PreferenceRouter

Returns

void

reset()

reset(): void;

Defined in: packages/nexus-agents/src/learning/outcome-feedback.ts:300

Reset all state.

Returns

void

Implementation of

IOutcomeFeedback.reset

OutcomeStorageError

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:29

Error class for outcome storage operations.

Extends

NexusError

Constructors

Constructor

new OutcomeStorageError(message, options?): OutcomeStorageError;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:30

Parameters

message

string

options?

Partial<Omit<{ cause?: Error; code: ErrorCode; context?: Record<string, unknown>; }, "code">>

Returns

OutcomeStorageError

Overrides

NexusError.constructor

Properties

cause

readonly cause: Error | undefined;

Defined in: packages/nexus-agents/src/core/errors.ts:105

Inherited from

NexusError.cause

code

readonly code: ErrorCode;

Defined in: packages/nexus-agents/src/core/errors.ts:103

Inherited from

NexusError.code

context

readonly context: Record<string, unknown> | undefined;

Defined in: packages/nexus-agents/src/core/errors.ts:104

Inherited from

NexusError.context

message

message: string;

Defined in: node_modules/.pnpm/typescript@6.0.3/node_modules/typescript/lib/lib.es5.d.ts:1075

Inherited from

NexusError.message

name

name: string;

Defined in: node_modules/.pnpm/typescript@6.0.3/node_modules/typescript/lib/lib.es5.d.ts:1074

Inherited from

NexusError.name

stack?

stack?: string;

Defined in: node_modules/.pnpm/typescript@6.0.3/node_modules/typescript/lib/lib.es5.d.ts:1076

Inherited from

NexusError.stack

stackTraceLimit

static stackTraceLimit: number;

Defined in: node_modules/.pnpm/@types+node@24.19.2/node_modules/@types/node/globals.d.ts:68

The Error.stackTraceLimit property specifies the number of stack frames collected by a stack trace (whether generated by new Error().stack or Error.captureStackTrace(obj)).

The default value is 10 but may be set to any valid JavaScript number. Changes will affect any stack trace captured after the value has been changed.

If set to a non-number value, or set to a negative number, stack traces will not capture any frames.

Inherited from

NexusError.stackTraceLimit

Methods

toJSON()

toJSON(): SerializedError;

Defined in: packages/nexus-agents/src/core/errors.ts:121

Serializes the error to a JSON-safe object.

Returns

SerializedError

Inherited from

NexusError.toJSON

captureStackTrace()

static captureStackTrace(targetObject, constructorOpt?): void;

Defined in: node_modules/.pnpm/@types+node@24.19.2/node_modules/@types/node/globals.d.ts:52

Creates a .stack property on targetObject, which when accessed returns a string representing the location in the code at which Error.captureStackTrace() was called.

const myObject = {};
Error.captureStackTrace(myObject);
myObject.stack;  // Similar to `new Error().stack`

The first line of the trace will be prefixed with ${myObject.name}: ${myObject.message}.

The optional constructorOpt argument accepts a function. If given, all frames above constructorOpt, including constructorOpt, will be omitted from the generated stack trace.

The constructorOpt argument is useful for hiding implementation details of error generation from the user. For instance:

function a() {
  b();
}

function b() {
  c();
}

function c() {
  // Create an error without stack trace to avoid calculating the stack trace twice.
  const { stackTraceLimit } = Error;
  Error.stackTraceLimit = 0;
  const error = new Error();
  Error.stackTraceLimit = stackTraceLimit;

  // Capture the stack trace above function b
  Error.captureStackTrace(error, b); // Neither function c, nor b is included in the stack trace
  throw error;
}

a();

Parameters

targetObject

object

constructorOpt?

Function

Returns

void

Inherited from

NexusError.captureStackTrace

prepareStackTrace()

static prepareStackTrace(err, stackTraces): any;

Defined in: node_modules/.pnpm/@types+node@24.19.2/node_modules/@types/node/globals.d.ts:56

Parameters

err

Error

stackTraces

CallSite[]

Returns

any

See

https://v8.dev/docs/stack-trace-api#customizing-stack-traces

Inherited from

NexusError.prepareStackTrace

PersistentStrategyDistiller

Defined in: packages/nexus-agents/src/learning/strategy-distiller-persistence.ts:133

StrategyDistiller that persists distilled rules to a JSON file.

Construction: hydrates from rules.json via Zod validation
distill(): calls super.distill() then atomically saves snapshot
Corruption: warn + start fresh (no partial loads)

Extends

StrategyDistiller

Constructors

Constructor

new PersistentStrategyDistiller(
   outcomeStore, 
   persistConfig?, 
   logger?, 
   distillerConfig?
): PersistentStrategyDistiller;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-persistence.ts:139

Parameters

outcomeStore

OutcomeStore

persistConfig?

PersistentDistillerConfig

logger?

ILogger

distillerConfig?

Partial<DistillerConfig>

Returns

PersistentStrategyDistiller

Overrides

StrategyDistiller.constructor

Properties

config

protected readonly config: DistillerConfig;

Defined in: packages/nexus-agents/src/learning/strategy-distiller.ts:223

Protected so PersistentStrategyDistiller hydrates legacy rules under the live thresholds.

Inherited from

StrategyDistiller.config

Methods

checkPersistedTrigger()

checkPersistedTrigger(): boolean;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-persistence.ts:159

An existing but unreadable rules.json is not overwritten by the first-route trigger: it may hold rules written by a newer build, or be recoverable by hand. It is logged and skipped; the in-process counter still distills.

Returns

boolean

Overrides

StrategyDistiller.checkPersistedTrigger

distill()

distill(): void;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-persistence.ts:168

Override distill to persist rules after each run.

Returns

void

Overrides

StrategyDistiller.distill

getRules()

getRules(status?): readonly DistilledRule[];

Defined in: packages/nexus-agents/src/learning/strategy-distiller.ts:318

Get rules filtered by status.

Parameters

status?

RuleStatus

Returns

readonly DistilledRule[]

Inherited from

StrategyDistiller.getRules

getStats()

getStats(): DistillerStats;

Defined in: packages/nexus-agents/src/learning/strategy-distiller.ts:325

Get distiller statistics.

Returns

DistillerStats

Inherited from

StrategyDistiller.getStats

loadRules()

protected loadRules(rules): void;

Defined in: packages/nexus-agents/src/learning/strategy-distiller.ts:348

Load pre-existing rules (e.g., from disk). Used by PersistentStrategyDistiller.

Parameters

rules

readonly DistilledRule[]

Returns

void

Inherited from

StrategyDistiller.loadRules

onOutcome()

onOutcome(): void;

Defined in: packages/nexus-agents/src/learning/strategy-distiller.ts:243

Called for each processed outcome. Triggers distillation at threshold. This counter is per-process; checkPersistedTrigger covers the cross-process case (#6512).

Returns

void

Inherited from

StrategyDistiller.onOutcome

restoreDistillState()

protected restoreDistillState(lastDistillAt, eligibleOutcomes): void;

Defined in: packages/nexus-agents/src/learning/strategy-distiller.ts:358

Restore last-distill state from a persisted snapshot (#6512), so the first-route trigger counts only outcomes newer than that snapshot.

Parameters

lastDistillAt

number

eligibleOutcomes

number | undefined

Returns

void

Inherited from

StrategyDistiller.restoreDistillState

SQLiteOutcomeStorage

Defined in: packages/nexus-agents/src/learning/outcome-storage.ts:84

SQLite-based outcome storage implementation.

Implements

IOutcomeStorage

Constructors

Constructor

new SQLiteOutcomeStorage(config): SQLiteOutcomeStorage;

Defined in: packages/nexus-agents/src/learning/outcome-storage.ts:91

Parameters

config

OutcomeStorageConfig

Returns

SQLiteOutcomeStorage

Methods

close()

close(): void;

Defined in: packages/nexus-agents/src/learning/outcome-storage.ts:363

Close the database connection.

Returns

void

getCounts()

getCounts(): Promise<Result<{
  decisions: number;
  outcomes: number;
  rewards: number;
}, OutcomeStorageError>>;

Defined in: packages/nexus-agents/src/learning/outcome-storage.ts:341

Get total record counts.

Returns

Promise<Result<{ decisions: number; outcomes: number; rewards: number; }, OutcomeStorageError>>

Implementation of

IOutcomeStorage.getCounts

getDecision()

getDecision(id): Promise<Result<StoredRoutingDecision | null, OutcomeStorageError>>;

Defined in: packages/nexus-agents/src/learning/outcome-storage.ts:252

Get routing decision by ID.

Parameters

id

string

Returns

Promise<Result<StoredRoutingDecision | null, OutcomeStorageError>>

Implementation of

IOutcomeStorage.getDecision

getDecisionsByRequestId()

getDecisionsByRequestId(requestId): Promise<Result<StoredRoutingDecision[], OutcomeStorageError>>;

Defined in: packages/nexus-agents/src/learning/outcome-storage.ts:304

Get decisions by request ID (for audit trail integration).

Parameters

requestId

string

Returns

Promise<Result<StoredRoutingDecision[], OutcomeStorageError>>

Implementation of

IOutcomeStorage.getDecisionsByRequestId

getModelStats()

getModelStats(): Promise<Result<StoredModelStats[], OutcomeStorageError>>;

Defined in: packages/nexus-agents/src/learning/outcome-storage.ts:278

Get aggregated statistics per model.

Returns

Promise<Result<StoredModelStats[], OutcomeStorageError>>

Implementation of

IOutcomeStorage.getModelStats

getOutcome()

getOutcome(decisionId): Promise<Result<StoredTaskOutcome | null, OutcomeStorageError>>;

Defined in: packages/nexus-agents/src/learning/outcome-storage.ts:265

Get outcome for a routing decision.

Parameters

decisionId

string

Returns

Promise<Result<StoredTaskOutcome | null, OutcomeStorageError>>

Implementation of

IOutcomeStorage.getOutcome

getRecentDecisions()

getRecentDecisions(model, limit): Promise<Result<StoredRoutingDecision[], OutcomeStorageError>>;

Defined in: packages/nexus-agents/src/learning/outcome-storage.ts:289

Get recent decisions for a model.

Parameters

model

"claude" | "gemini" | "codex" | "opencode"

limit

number

Returns

Promise<Result<StoredRoutingDecision[], OutcomeStorageError>>

Implementation of

IOutcomeStorage.getRecentDecisions

initialize()

initialize(): Promise<Result<void, OutcomeStorageError>>;

Defined in: packages/nexus-agents/src/learning/outcome-storage.ts:115

Initialize the storage backend.

Returns

Promise<Result<void, OutcomeStorageError>>

initializeWithDatabase()

initializeWithDatabase(database): void;

Defined in: packages/nexus-agents/src/learning/outcome-storage.ts:107

Initialize with an existing database instance (for testing).

Parameters

database

ISQLiteDatabase

Returns

void

prune()

prune(olderThan): Promise<Result<number, OutcomeStorageError>>;

Defined in: packages/nexus-agents/src/learning/outcome-storage.ts:318

Prune old records.

Parameters

olderThan

Date

Returns

Promise<Result<number, OutcomeStorageError>>

Implementation of

IOutcomeStorage.prune

storeDecision()

storeDecision(decision): Promise<Result<void, OutcomeStorageError>>;

Defined in: packages/nexus-agents/src/learning/outcome-storage.ts:160

Store a routing decision.

Parameters

decision

StoredRoutingDecision

Returns

Promise<Result<void, OutcomeStorageError>>

Implementation of

IOutcomeStorage.storeDecision

storeOutcome()

storeOutcome(outcome): Promise<Result<void, OutcomeStorageError>>;

Defined in: packages/nexus-agents/src/learning/outcome-storage.ts:194

Store a task outcome.

Parameters

outcome

StoredTaskOutcome

Returns

Promise<Result<void, OutcomeStorageError>>

Implementation of

IOutcomeStorage.storeOutcome

storeReward()

storeReward(reward): Promise<Result<void, OutcomeStorageError>>;

Defined in: packages/nexus-agents/src/learning/outcome-storage.ts:223

Store a computed reward.

Parameters

reward

StoredReward

Returns

Promise<Result<void, OutcomeStorageError>>

Implementation of

IOutcomeStorage.storeReward

StrategyDistiller

Defined in: packages/nexus-agents/src/learning/strategy-distiller.ts:221

Distills outcome patterns into routing rules.

Subscribe to OutcomeFeedbackCollector.onOutcomeProcessed() and call onOutcome() for each processed outcome. Distillation triggers automatically every triggerThreshold outcomes.

Extended by

PersistentStrategyDistiller

Constructors

Constructor

new StrategyDistiller(
   outcomeStore, 
   logger?, 
   config?
): StrategyDistiller;

Defined in: packages/nexus-agents/src/learning/strategy-distiller.ts:232

Parameters

outcomeStore

OutcomeStore

logger?

ILogger

config?

Partial<DistillerConfig>

Returns

StrategyDistiller

Properties

config

protected readonly config: DistillerConfig;

Defined in: packages/nexus-agents/src/learning/strategy-distiller.ts:223

Protected so PersistentStrategyDistiller hydrates legacy rules under the live thresholds.

Methods

checkPersistedTrigger()

checkPersistedTrigger(): boolean;

Defined in: packages/nexus-agents/src/learning/strategy-distiller.ts:268

First-route trigger against the persisted store (#6512).

The in-process counter above only fires after triggerThreshold outcomes in ONE process, which short-lived CLI processes never reach. This checks the store instead: distill when it holds at least triggerThreshold eligible outcomes newer than the last distill (a hydrated snapshot’s savedAt; no snapshot means all of them count). With no snapshot and zero eligible outcomes it distills anyway, so the empty state is recorded as “0 eligible, 0 rules, at ” rather than left as “never”.

Before deciding, rules past ruleExpiryMs are expired, so a hydrated rule that is out of date is never applied.

Runs at most once per instance; DistilledRuleStage calls it when it first runs, so constructing a distiller stays cheap. Returns whether it distilled.

Returns

boolean

distill()

distill(): void;

Defined in: packages/nexus-agents/src/learning/strategy-distiller.ts:286

Run distillation on the eligible outcomes in the OutcomeStore (#6512).

Returns

void

getRules()

getRules(status?): readonly DistilledRule[];

Defined in: packages/nexus-agents/src/learning/strategy-distiller.ts:318

Get rules filtered by status.

Parameters

status?

RuleStatus

Returns

readonly DistilledRule[]

getStats()

getStats(): DistillerStats;

Defined in: packages/nexus-agents/src/learning/strategy-distiller.ts:325

Get distiller statistics.

Returns

DistillerStats

loadRules()

protected loadRules(rules): void;

Defined in: packages/nexus-agents/src/learning/strategy-distiller.ts:348

Load pre-existing rules (e.g., from disk). Used by PersistentStrategyDistiller.

Parameters

rules

readonly DistilledRule[]

Returns

void

onOutcome()

onOutcome(): void;

Defined in: packages/nexus-agents/src/learning/strategy-distiller.ts:243

Called for each processed outcome. Triggers distillation at threshold. This counter is per-process; checkPersistedTrigger covers the cross-process case (#6512).

Returns

void

restoreDistillState()

protected restoreDistillState(lastDistillAt, eligibleOutcomes): void;

Defined in: packages/nexus-agents/src/learning/strategy-distiller.ts:358

Restore last-distill state from a persisted snapshot (#6512), so the first-route trigger counts only outcomes newer than that snapshot.

Parameters

lastDistillAt

number

eligibleOutcomes

number | undefined

Returns

void

Interfaces

ComparisonResult

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:57

Result of a two-sample comparison test.

Properties

alpha

readonly alpha: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:63

Alpha level used for significance

difference

readonly difference: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:65

Difference in success rates (group1 - group2)

differenceCI

readonly differenceCI: ConfidenceInterval;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:67

Confidence interval for the difference

effectSize

readonly effectSize: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:69

Effect size (Cohen’s h for proportions)

n1

readonly n1: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:71

Sample sizes

n2

readonly n2: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:72

pValue

readonly pValue: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:59

P-value from the test

significant

readonly significant: boolean;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:61

Whether result is significant at alpha level

ComputedReward

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:204

Computed reward for bandit update.

Properties

components

readonly components: {
  baseReward: number;
  efficiencyBonus: number;
  qualityBonus: number;
  retryPenalty: number;
  speedBonus: number;
};

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:208

Components that contributed to the reward

baseReward

readonly baseReward: number;

efficiencyBonus

readonly efficiencyBonus: number;

qualityBonus

readonly qualityBonus: number;

retryPenalty

readonly retryPenalty: number;

speedBonus

readonly speedBonus: number;

explanation

readonly explanation: string;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:216

Explanation of reward computation

reward

readonly reward: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:206

The reward value (0-1)

ConfidenceInterval

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:14

Confidence interval result with bounds and metadata.

Properties

confidence

readonly confidence: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:22

Confidence level (e.g., 0.95 for 95% CI)

estimate

readonly estimate: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:20

Point estimate (center of interval)

lower

readonly lower: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:16

Lower bound of the interval

measured

readonly measured: boolean;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:51

Whether the interval was computed from data (#5760 item 3).

false means the sample was empty and lower/upper/estimate carry no information — they are whatever the empty branch happened to return, not a result. Read this before reading the bounds.

The bounds alone cannot say it. meanConfidenceInterval([]) returned lower === upper === 0 — the strongest possible precision claim, over no data. proportionConfidenceInterval(0, 0) returned [0, 1], which is honest by luck rather than by construction: a proportion’s domain happens to be bounded, so its uninformative interval is expressible, and a mean’s is not. calculateDifferenceCI divided by (total1 || 1) and produced a finite interval from nothing at all.

Named for the vocabulary this repo already uses for exactly this — tokensMeasured, policyEvaluated, gapsMeasured, routerTypeMeasured. REQUIRED rather than optional so the compiler names every producer; optional would let a new one stay silent, which is the state this removes.

Ratified 5 of 6 approvers at supermajority (#5760). Infinity and NaN bounds were both rejected for the same reason: each serialises to null in JSON, turning a loud in-process signal into a silent persisted one.

n

readonly n: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:24

Sample size used

standardError

readonly standardError: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:26

Standard error

upper

readonly upper: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:18

Upper bound of the interval

DistilledRule

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:30

A distilled routing rule extracted from outcome patterns.

Rules are fingerprinted by patternType:cli:category to prevent duplicates and cap total rules at a bounded maximum.

Properties

action

readonly action: StrategyAction;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:40

What routing action to take

cli

readonly cli: RoutingArmId;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:36

Which routing arm (CLI slot or API/endpoint arm) this rule applies to (#6555)

confidence

readonly confidence: number;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:50

Confidence 0-1 = support × effect (#5004 finding 3).

This is the value DistilledRuleStage.computeDelta multiplies the base delta by, so it must answer “how much should routing move” — not “how many samples did we see”. Before #5004 it was the sigmoid over observations alone, and a rule at 62.5% failure penalised exactly as hard as one at 100% given the same traffic.

createdAt

readonly createdAt: number;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:67

Epoch ms when rule was first created

effect

readonly effect: number;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:59

Effect size 0-1: how far metric sits past its detector threshold, normalised over the remaining headroom — see effectFor. Persisted at distill time because the threshold is config; recomputing it on load would silently rescale old rules when the threshold changes.

id

readonly id: string;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:32

Fingerprint: ${patternType}:${cli}:${category}

metric

readonly metric: number;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:63

The metric value (failure rate, success rate, or p90/median ratio)

observationCount

readonly observationCount: number;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:61

Number of observations that informed this rule

patternType

readonly patternType: PatternType;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:34

What kind of pattern triggered this rule

status

readonly status: RuleStatus;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:65

Current lifecycle status

support

readonly support: number;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:52

Sample support 0-1: sigmoidConfidence(observationCount) (center=30).

updatedAt

readonly updatedAt: number;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:69

Epoch ms when rule was last updated

DistillerConfig

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:73

Configuration for the strategy distiller.

Properties

failureRateThreshold

readonly failureRateThreshold: number;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:81

Failure rate above which a failure pattern is detected (default: 0.6)

latencyRatioThreshold

readonly latencyRatioThreshold: number;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:85

p90/median ratio above which a latency spike is detected (default: 2.0)

maxRules

readonly maxRules: number;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:87

Maximum number of rules to store (default: 90)

minObservationsForActive

readonly minObservationsForActive: number;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:79

Minimum observations before activating a rule (default: 5)

minObservationsForDraft

readonly minObservationsForDraft: number;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:77

Minimum observations before creating a draft rule (default: 3)

ruleExpiryMs

readonly ruleExpiryMs: number;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:89

Rule expiry time in ms (default: 24h)

successRateThreshold

readonly successRateThreshold: number;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:83

Success rate above which a success pattern is detected (default: 0.8)

triggerThreshold

readonly triggerThreshold: number;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:75

Distill every N outcomes (default: 50)

DistillerStats

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:111

Statistics returned by StrategyDistiller.getStats().

Properties

eligibleOutcomesAtLastDistill?

readonly eligibleOutcomesAtLastDistill?: number;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:125

Eligible outcomes the last distillation trained on (#6512). Undefined when no distillation has run or been restored from a snapshot; 0 is a real measurement (a distill ran and nothing was eligible).

lastDistillAt

readonly lastDistillAt: number | undefined;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:117

Epoch ms of last distillation run

outcomesSinceLastDistill

readonly outcomesSinceLastDistill: number;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:119

Number of outcomes processed since last distillation

ruleCountByStatus

readonly ruleCountByStatus: Readonly<Record<RuleStatus, number>>;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:113

Number of rules in each status

totalRules

readonly totalRules: number;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:115

Total number of rules

DistributionStats

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:78

Descriptive statistics for a distribution.

Properties

max

readonly max: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:84

mean

readonly mean: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:79

median

readonly median: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:80

min

readonly min: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:83

n

readonly n: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:85

percentiles

readonly percentiles: {
  p25: number;
  p5: number;
  p50: number;
  p75: number;
  p95: number;
};

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:87

Percentiles: p5, p25, p50, p75, p95

p25

readonly p25: number;

p5

readonly p5: number;

p50

readonly p50: number;

p75

readonly p75: number;

p95

readonly p95: number;

stdDev

readonly stdDev: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:81

variance

readonly variance: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:82

ExperimentDefinition

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:37

Experiment definition.

Properties

description

readonly description: string;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:43

Description of the experiment’s hypothesis

endedAt

readonly endedAt: string | null;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:51

End timestamp (ISO 8601)

id

readonly id: string;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:39

Unique experiment identifier

minimumDetectableEffect

readonly minimumDetectableEffect: number;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:57

Minimum detectable effect size

minSampleSize

readonly minSampleSize: number;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:53

Minimum sample size per variant

name

readonly name: string;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:41

Human-readable name

primaryMetric

readonly primaryMetric: "successRate" | "avgReward" | "avgLatency";

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:55

Primary metric to optimize

startedAt

readonly startedAt: string | null;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:49

Start timestamp (ISO 8601)

status

readonly status: ExperimentStatus;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:45

Current status

variants

readonly variants: readonly ExperimentVariant[];

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:47

Experiment variants

ExperimentExport

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:187

Export format for experiment data.

Properties

experiments

readonly experiments: readonly ExperimentDefinition[];

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:191

All experiments

exportedAt

readonly exportedAt: string;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:189

Export timestamp

outcomes

readonly outcomes: readonly ExperimentOutcome[];

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:193

All outcomes

summaries

readonly summaries: readonly ExperimentSummary[];

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:195

Summaries for completed experiments

ExperimentOutcome

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:65

Recorded outcome for an experiment.

Properties

experimentId

readonly experimentId: string;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:67

Experiment ID

latencyMs

readonly latencyMs: number;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:77

Latency in milliseconds

metadata?

readonly metadata?: Record<string, unknown>;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:81

Additional metadata

reward

readonly reward: number;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:75

Reward value

success

readonly success: boolean;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:73

Whether the task succeeded

timestamp

readonly timestamp: string;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:79

Timestamp (ISO 8601)

traceId

readonly traceId: string;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:71

Routing decision trace ID

variantId

readonly variantId: string;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:69

Assigned variant ID

ExperimentResult

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:158

A/B test experiment result.

Properties

comparison

readonly comparison: ComparisonResult;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:166

Comparison between groups

control

readonly control: VariantResultSummary;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:162

Control group statistics

experimentId

readonly experimentId: string;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:160

Experiment identifier

hasMinimumSampleSize

readonly hasMinimumSampleSize: boolean;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:177

Whether experiment has enough data for valid conclusions

recommendedSampleSize

readonly recommendedSampleSize: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:179

Minimum recommended sample size per group

recommendedSampleSizeMeasured

readonly recommendedSampleSizeMeasured: boolean;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:195

Whether recommendedSampleSize was computed from a baseline the control actually measured.

false means the control has no observations, so the 0 baseline calculateMinSampleSize was handed is a default, not a rate. The number it returns is real arithmetic over a fabricated input and comes out several times too small — in the direction that tells an operator to stop collecting early, which is exactly the reader this field exists to warn.

Note the gate is control.n > 0, NOT control.successRate > 0: a control of 0/50 is a measured baseline of 0.0 and a legitimate input here. That is a different question from whether relativeImprovement is non-null: the ratio over a zero control rate is undefined (#5857).

relativeImprovement

readonly relativeImprovement: number | null;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:175

Relative improvement (treatment vs control), or null when control has no samples or its measured success rate is zero (the ratio is undefined). Inspect control.n: zero means no samples; a positive count with null lift means a measured zero baseline, including unbounded lift when the treatment rate is positive. A measured zero lift remains numeric 0. Render null as ‘-’ rather than as 0% improvement.

treatment

readonly treatment: VariantResultSummary;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:164

Treatment group statistics

ExperimentSummary

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:111

Experiment summary with all variants and comparison.

Properties

experiment

readonly experiment: ExperimentDefinition;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:113

Experiment definition

hasMinimumSampleSize

readonly hasMinimumSampleSize: boolean;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:119

Whether experiment has reached minimum sample size

recommendation

readonly recommendation: "continue" | "stop_winner" | "stop_inconclusive";

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:121

Recommended action based on results

result

readonly result: ExperimentResult | null;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:117

Statistical comparison result

variantStats

readonly variantStats: readonly VariantStats[];

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:115

Statistics per variant

ExperimentVariant

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:21

Experiment variant configuration.

Properties

description

readonly description: string;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:27

Description of what this variant does

id

readonly id: string;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:23

Variant identifier

isControl

readonly isControl: boolean;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:31

Whether this is the control variant

name

readonly name: string;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:25

Human-readable name

trafficPercent

readonly trafficPercent: number;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:29

Traffic allocation percentage (0-100)

FeedbackCollectorConfig

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:248

Configuration for the feedback collector.

Properties

efficiencyWeight

readonly efficiencyWeight: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:260

Weight for efficiency in reward computation

enableAutoReward

readonly enableAutoReward: boolean;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:254

Enable automatic reward computation

maxHistorySize

readonly maxHistorySize: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:268

Maximum history entries (outcomes + decisions) to retain

maxPendingDecisions

readonly maxPendingDecisions: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:250

Maximum pending decisions to track

pendingTimeoutMs

readonly pendingTimeoutMs: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:252

Timeout for pending decisions (ms)

qualityWeight

readonly qualityWeight: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:256

Weight for quality in reward computation

retryPenalty

readonly retryPenalty: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:262

Penalty per retry in reward computation

speedWeight

readonly speedWeight: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:258

Weight for speed in reward computation

targetDurationMs

readonly targetDurationMs: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:264

Target duration for speed bonus (ms)

targetTokenUsage

readonly targetTokenUsage: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:266

Target token usage for efficiency bonus

FeedbackIntegrationConfig

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:53

Configuration for feedback integration.

Properties

decisionTtlMs?

readonly decisionTtlMs?: number;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:61

TTL for decision entries in milliseconds (default: 3600000 = 1 hour)

enableAutoFeedback

readonly enableAutoFeedback: boolean;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:55

Enable automatic feedback to routers (default: true)

enablePersistence?

readonly enablePersistence?: boolean;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:69

Enable persistent storage via SQLite (default: false). Requires outcomeStorage to be provided. (Source: Issue #560 - Wire SQLiteOutcomeStorage to feedback loop)

logger?

readonly logger?: ILogger;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:63

Logger instance

outcomeStorage?

readonly outcomeStorage?: IOutcomeStorage;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:75

SQLite outcome storage instance for cross-session learning. Only used when enablePersistence is true. (Source: Issue #560 - Wire SQLiteOutcomeStorage to feedback loop)

partialQualityThreshold

readonly partialQualityThreshold: number;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:59

Quality score threshold for partial success (default: 0.4)

successQualityThreshold

readonly successQualityThreshold: number;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:57

Quality score threshold for success (default: 0.7)

FeedbackLoopStats

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:222

Feedback loop statistics.

Properties

avgQualityScore

readonly avgQualityScore: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:232

Average quality score

avgReward

readonly avgReward: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:234

Average reward computed

decisionsByRouter

readonly decisionsByRouter: Record<RouterType, number>;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:240

Decisions by router type. The unattributed bucket includes explicit unattributed labels and legacy rows with false or absent measurement evidence. Empty decision history gives zero in every bucket.

lastUpdatedAt

readonly lastUpdatedAt: string;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:242

Last update timestamp

outcomesByClass

readonly outcomesByClass: Record<OutcomeClass, number>;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:230

Outcomes by classification

pendingOutcomes

readonly pendingOutcomes: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:228

Decisions pending outcome

totalDecisions

readonly totalDecisions: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:224

Total routing decisions recorded

totalOutcomes

readonly totalOutcomes: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:226

Total outcomes recorded

FeedbackRoutingDecision

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:88

Recorded routing decision for feedback tracking.

Properties

armIndex?

readonly armIndex?: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:112

Arm index for LinUCB bandit

banditContext?

readonly banditContext?: BanditContext;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:114

Bandit context (for LinUCB decisions)

confidence?

readonly confidence?: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:120

Confidence score

domain?

readonly domain?: string;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:124

Task domain classification

id

readonly id: string;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:90

Unique decision ID

query

readonly query: string;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:94

Original routed task text; empty means unavailable and skips preference training.

queryFeatures?

readonly queryFeatures?: QueryFeatures;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:116

Query features (for preference routing)

routerType

readonly routerType: RouterType;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:96

Type of router used

routerTypeMeasured?

readonly routerTypeMeasured?: boolean;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:106

Persisted evidence that a scoring stage attributed this decision (#5915). New fallback decisions use unattributed and false. This field remains necessary for legacy labels: stored topsis may have been a fallback, and its label is preserved because historical attribution is unknowable.

true means measured; false or absent means unmeasured. Never infer measurement from a legacy label when this evidence is absent.

selectedModel

readonly selectedModel: string;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:108

Selected model/adapter name

selectedTier?

readonly selectedTier?: "strong" | "weak";

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:110

Selected model tier (strong/weak for preference routing)

timestamp

readonly timestamp: string;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:92

Timestamp of decision

traceId

readonly traceId: string;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:122

Trace ID to correlate with SwarmObserver events

ucbScore?

readonly ucbScore?: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:118

UCB score (for LinUCB)

IAbTestTracker

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:127

A/B test tracker interface.

Methods

assignVariant()

assignVariant(experimentId, traceId): ExperimentVariant | null;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:153

Assign a variant for a given trace ID (deterministic assignment).

Parameters

experimentId

string

traceId

string

Returns

ExperimentVariant | null

completeExperiment()

completeExperiment(experimentId): void;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:148

Complete an experiment.

Parameters

experimentId

string

Returns

void

createExperiment()

createExperiment(definition): ExperimentDefinition;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:131

Create a new experiment.

Parameters

definition

Omit<ExperimentDefinition, "status" | "startedAt" | "endedAt">

Returns

ExperimentDefinition

exportData()

exportData(): ExperimentExport;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:181

Export all experiment data.

Returns

ExperimentExport

getExperiment()

getExperiment(experimentId): ExperimentDefinition | null;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:176

Get experiment by ID.

Parameters

experimentId

string

Returns

ExperimentDefinition | null

getSummary()

getSummary(experimentId): ExperimentSummary | null;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:163

Get experiment summary with statistics.

Parameters

experimentId

string

Returns

ExperimentSummary | null

listExperiments()

listExperiments(filter?): readonly ExperimentDefinition[];

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:168

List all experiments.

Parameters

filter?

status?

ExperimentStatus

tags?

readonly string[]

Returns

readonly ExperimentDefinition[]

pauseExperiment()

pauseExperiment(experimentId): void;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:143

Pause a running experiment.

Parameters

experimentId

string

Returns

void

recordOutcome()

recordOutcome(outcome): void;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:158

Record an outcome for an experiment.

Parameters

outcome

ExperimentOutcome

Returns

void

startExperiment()

startExperiment(experimentId): void;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:138

Start an experiment (sets status to running).

Parameters

experimentId

string

Returns

void

IFeedbackIntegration

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:98

Interface for feedback integration.

Methods

evictStaleEntries()

evictStaleEntries(): number;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:130

Evict stale entries from decision map that exceed TTL

Returns

number

getDecisionMapSize()

getDecisionMapSize(): number;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:136

Get current size of decision map

Returns

number

getEvictedEntryCount()

getEvictedEntryCount(): number;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:133

Get total count of evicted entries since creation or last reset

Returns

number

getStats()

getStats(): FeedbackLoopStats;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:118

Get feedback statistics

Returns

OutcomeProcessedCallback

onOutcomeProcessed()

onOutcomeProcessed(callback): () => void;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:121

Subscribe to outcome processed events

Parameters

callback

Returns

() => void

recordOutcome()

recordOutcome(params): void;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:115

Record a generic task outcome

Parameters

params

RecordOutcomeParams

Returns

void

recordRoutingDecision()

recordRoutingDecision(
   decision, 
   traceId?, 
   options?
): string;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:100

Record a routing decision; omitted task text skips preference training.

Parameters

decision

CompositeRoutingDecision

traceId?

string

options?

query?

string

Returns

string

recordStepOutcome()

recordStepOutcome(
   routingDecisionId, 
   stepResult, 
   durationMs, 
   tokenUsage
): void;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:107

Record a step outcome from workflow execution

Parameters

routingDecisionId

string

stepResult

StepResult

durationMs

number

tokenUsage

number

Returns

void

registerCompositeRouter()

registerCompositeRouter(router): void;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:124

Parameters

router

ICompositeRouter

Returns

void

reset()

reset(): void;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:127

Reset all collected data

Returns

void

IOutcomeFeedback

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:306

Interface for outcome feedback collector.

Methods

clearExpiredDecisions()

clearExpiredDecisions(): number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:340

Clear expired pending decisions.

Returns

number

computeReward()

computeReward(outcome): ComputedReward;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:325

Compute reward from an outcome.

Parameters

outcome

Returns

getPendingDecisions()

getPendingDecisions(): readonly FeedbackRoutingDecision[];

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:335

Get pending decisions (waiting for outcomes).

Returns

readonly FeedbackRoutingDecision[]

getStats()

getStats(): FeedbackLoopStats;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:330

Get feedback loop statistics.

Returns

processOutcome()

processOutcome(traceId, outcome): void;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:320

Process outcome for a trace ID (finds matching decision and computes reward).

Parameters

traceId

string

outcome

Omit<TaskOutcome, "routingDecisionId">

Returns

void

recordOutcome()

recordOutcome(outcome): void;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:315

Record an outcome for a routing decision.

Parameters

outcome

Returns

void

recordRoutingDecision()

recordRoutingDecision(decision): void;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:310

Record a routing decision for tracking.

Parameters

decision

Returns

void

reset()

reset(): void;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:345

Reset all state.

Returns

void

IOutcomeStorage

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:194

Interface for outcome storage implementations.

Methods

getCounts()

getCounts(): Promise<Result<{
  decisions: number;
  outcomes: number;
  rewards: number;
}, OutcomeStorageError>>;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:250

Get total record counts.

Returns

Promise<Result<{ decisions: number; outcomes: number; rewards: number; }, OutcomeStorageError>>

getDecision()

getDecision(id): Promise<Result<StoredRoutingDecision | null, OutcomeStorageError>>;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:213

Get routing decision by ID.

Parameters

id

string

Returns

Promise<Result<StoredRoutingDecision | null, OutcomeStorageError>>

getDecisionsByRequestId()

getDecisionsByRequestId(requestId): Promise<Result<StoredRoutingDecision[], OutcomeStorageError>>;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:238

Get decisions by request ID (for audit trail integration).

Parameters

requestId

string

Returns

Promise<Result<StoredRoutingDecision[], OutcomeStorageError>>

getModelStats()

getModelStats(): Promise<Result<StoredModelStats[], OutcomeStorageError>>;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:225

Get aggregated statistics per model.

Returns

Promise<Result<StoredModelStats[], OutcomeStorageError>>

getOutcome()

getOutcome(routingDecisionId): Promise<Result<StoredTaskOutcome | null, OutcomeStorageError>>;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:218

Get outcome for a routing decision.

Parameters

routingDecisionId

string

Returns

Promise<Result<StoredTaskOutcome | null, OutcomeStorageError>>

getRecentDecisions()

getRecentDecisions(model, limit): Promise<Result<StoredRoutingDecision[], OutcomeStorageError>>;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:230

Get recent decisions for a model.

Parameters

model

"claude" | "gemini" | "codex" | "opencode"

limit

number

Returns

Promise<Result<StoredRoutingDecision[], OutcomeStorageError>>

prune()

prune(olderThan): Promise<Result<number, OutcomeStorageError>>;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:245

Prune old records.

Parameters

olderThan

Date

Returns

Promise<Result<number, OutcomeStorageError>>

storeDecision()

storeDecision(decision): Promise<Result<void, OutcomeStorageError>>;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:198

Store a routing decision.

Parameters

decision

StoredRoutingDecision

Returns

Promise<Result<void, OutcomeStorageError>>

storeOutcome()

storeOutcome(outcome): Promise<Result<void, OutcomeStorageError>>;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:203

Store a task outcome.

Parameters

outcome

StoredTaskOutcome

Returns

Promise<Result<void, OutcomeStorageError>>

storeReward()

storeReward(reward): Promise<Result<void, OutcomeStorageError>>;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:208

Store a computed reward.

Parameters

reward

StoredReward

Returns

Promise<Result<void, OutcomeStorageError>>

ISQLiteDatabase

Defined in: packages/nexus-agents/src/core/types/database-types.ts:57

Minimal interface for a SQLite database handle. Satisfied by node:sqlite’s DatabaseSync via openSqliteDatabase (#5388). ~30 helper signatures and ~12 test doubles are written against this, which is why swapping the engine underneath did not ripple outward.

Methods

close()

close(): void;

Defined in: packages/nexus-agents/src/core/types/database-types.ts:71

Close the database connection.

Returns

void

exec()

exec(sql): void;

Defined in: packages/nexus-agents/src/core/types/database-types.ts:61

Execute raw SQL statements (typically for DDL or multiple statements).

Parameters

sql

string

Returns

void

prepare()

prepare<T>(sql): ISQLiteStatement<T>;

Defined in: packages/nexus-agents/src/core/types/database-types.ts:66

Prepare a parameterized statement for execution.

Type Parameters

T

T = unknown

Parameters

sql

string

Returns

ISQLiteStatement<T>

ISQLiteStatement

Defined in: packages/nexus-agents/src/core/types/database-types.ts:18

Minimal interface for a prepared SQLite statement. Satisfied by node:sqlite’s StatementSync (#5388).

Type Parameters

T

T = unknown

Methods

all()

all(...params): T[];

Defined in: packages/nexus-agents/src/core/types/database-types.ts:34

Get all rows matching the statement.

Parameters

params

…unknown[]

Returns

T[]

get()

get(...params): T | undefined;

Defined in: packages/nexus-agents/src/core/types/database-types.ts:29

Get a single row matching the statement. Returns undefined if no match.

Parameters

params

…unknown[]

Returns

T | undefined

run()

run(...params): ISQLiteRunResult;

Defined in: packages/nexus-agents/src/core/types/database-types.ts:23

Execute the statement with the given parameters. Returns information about changes made.

Parameters

params

…unknown[]

Returns

ISQLiteRunResult

OutcomeStorageConfig

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:262

Configuration for SQLite outcome storage.

Properties

autoPruneInterval?

autoPruneInterval?: number;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:270

Auto-prune interval in ms (default: 3600000 = 1 hour)

dbPath

dbPath: string;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:264

Path to SQLite database file

logger?

logger?: ILogger;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:266

Optional logger instance

maxRecords?

maxRecords?: number;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:268

Maximum records to retain (default: 100000)

PerformanceMatrixEntry

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:201

Model performance matrix entry (model × task type).

Properties

avgLatencyMs

readonly avgLatencyMs: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:207

avgReward

readonly avgReward: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:206

model

readonly model: string;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:202

n

readonly n: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:204

successRate

readonly successRate: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:205

successRateCI

readonly successRateCI: ConfidenceInterval;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:208

taskType

readonly taskType: string;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:203

PersistentDistillerConfig

Defined in: packages/nexus-agents/src/learning/strategy-distiller-persistence.ts:115

Learning exports - Closed-loop feedback and routing improvement Split from index.ts for file size compliance (Issue #285)

Properties

dataDir?

readonly dataDir?: string;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-persistence.ts:119

Override the data directory (useful for testing).

filePath?

readonly filePath?: string;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-persistence.ts:117

Override the file path (useful for testing).

QualitySignals

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:55

Quality signals extracted from task execution.

Properties

coherenceScore?

readonly coherenceScore?: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:69

Response coherence score (0-1)

completionRatio

readonly completionRatio: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:65

Task completion percentage (0-1)

lintErrors?

readonly lintErrors?: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:59

Number of lint errors (for code tasks)

retryCount

readonly retryCount: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:63

Number of retries required

testsPass?

readonly testsPass?: boolean;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:57

Whether code tests passed (for code tasks)

userApproved?

readonly userApproved?: boolean;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:61

Explicit user approval/rejection

validStructure?

readonly validStructure?: boolean;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:67

Whether output was valid JSON/structured (for structured output tasks)

RecordOutcomeParams

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:27

Parameters for recording an outcome.

Properties

durationMs

readonly durationMs: number;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:35

Execution duration in milliseconds

errorMessage?

readonly errorMessage?: string;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:43

Error message for failed outcomes (sanitized before persistence)

qualityScore

readonly qualityScore: number;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:33

Quality score (0-1)

retryCount?

readonly retryCount?: number;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:39

Number of retries (default: 0)

routingDecisionId

readonly routingDecisionId: string;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:29

Routing decision ID

success

readonly success: boolean;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:31

Whether the task succeeded

tokenUsage

readonly tokenUsage: number;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:37

Token usage

traceId?

readonly traceId?: string;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:41

Trace ID for correlation

RegretAnalysis

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:99

Regret analysis result comparing actual vs optimal decisions.

Properties

avgRegret

readonly avgRegret: number | null;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:115

Average regret per decision, or null when unmeasured — see above.

cumulativeRegret

readonly cumulativeRegret: number | null;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:113

Total cumulative regret, or null when no decision was comparable (#5255).

These three were plain number and returned 0/0/1 for an empty decision set — which the dashboard rendered as “Cumulative Regret: 0.00” and “Optimal Decision Rate: 100.0%” with a full progress bar. A perfect routing record asserted over nothing is an absent measurement wearing a good score, and the primary consumer of these numbers is the learning loop itself, which has no way to consult a sibling flag.

null means UNMEASURED — no comparable decision existed. It does not mean zero regret; that is 0, and the two are now distinguishable.

optimalRate

readonly optimalRate: number | null;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:121

Share of optimal decisions, or null when unmeasured — see above.

regretPerModel

readonly regretPerModel: Record<string, number>;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:123

Regret per model (how much worse each model performed vs best)

suboptimalDecisions

readonly suboptimalDecisions: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:119

Number of suboptimal decisions. Stays numeric: 0 is the true count.

totalDecisions

readonly totalDecisions: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:117

Number of decisions analyzed. Stays numeric: 0 is the true count.

StatisticalOptions

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:214

Options for statistical calculations.

Properties

alpha?

readonly alpha?: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:218

Alpha level for significance testing (default: 0.05)

confidence?

readonly confidence?: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:216

Confidence level for intervals (default: 0.95)

minSampleSize?

readonly minSampleSize?: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:220

Minimum sample size for valid inference (default: 30)

useContinuityCorrection?

readonly useContinuityCorrection?: boolean;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:222

Use continuity correction for proportions (default: true)

StoredModelStats

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:103

Aggregated model statistics from stored data.

Each average/rate is null when its own inputs are empty. Reward can be unmeasured even when outcomes exist. Null must not be ranked as zero; exclude unmeasured values or place them after measured values.

Properties

avgLatencyMs

readonly avgLatencyMs: number | null;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:109

avgQualityScore

readonly avgQualityScore: number | null;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:108

avgReward

readonly avgReward: number | null;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:107

model

readonly model: "claude" | "gemini" | "codex" | "opencode";

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:104

successRate

readonly successRate: number | null;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:110

totalDecisions

readonly totalDecisions: number;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:105

totalOutcomes

readonly totalOutcomes: number;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:106

StoredReward

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:85

Stored computed reward record.

Properties

baseReward

readonly baseReward: number;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:89

efficiencyBonus

readonly efficiencyBonus: number;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:92

qualityBonus

readonly qualityBonus: number;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:90

retryPenalty

readonly retryPenalty: number;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:93

reward

readonly reward: number;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:88

routingDecisionId

readonly routingDecisionId: string;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:86

speedBonus

readonly speedBonus: number;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:91

timestamp

readonly timestamp: string;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:87

StoredRoutingDecision

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:48

Stored routing decision record.

Properties

alternativeModels

readonly alternativeModels: readonly RoutingArmId[];

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:54

confidence

readonly confidence: number;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:55

id

readonly id: string;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:49

reason

readonly reason: string;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:56

requestId?

readonly requestId?: string;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:58

routerType

readonly routerType: RouterType;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:52

routerTypeMeasured?

readonly routerTypeMeasured?: boolean;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:65

Persisted attribution evidence (#5915), retained for legacy router labels. Old topsis labels stay topsis because they may be measured or fallback; the label alone cannot recover attribution. NULL or absent database evidence reads back as false. An omitted field on input is stored as unmeasured.

selectedModel

readonly selectedModel: RoutingArmId;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:53

taskProfile

readonly taskProfile: Record<string, unknown>;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:57

timestamp

readonly timestamp: string;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:51

traceId

readonly traceId: string;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:50

StoredTaskOutcome

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:71

Stored task outcome record.

Properties

durationMs

readonly durationMs: number;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:77

errorMessage?

readonly errorMessage?: string;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:79

outcomeClass

readonly outcomeClass: OutcomeClass;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:74

qualityScore

readonly qualityScore: number;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:76

routingDecisionId

readonly routingDecisionId: string;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:72

success

readonly success: boolean;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:75

timestamp

readonly timestamp: string;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:73

tokenUsage

readonly tokenUsage: number;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:78

TaskOutcome

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:162

Task outcome for a routing decision.

Properties

durationMs

readonly durationMs: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:174

Execution duration in milliseconds

errorMessage?

readonly errorMessage?: string;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:178

Error message if failed

outcomeClass

readonly outcomeClass: OutcomeClass;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:168

Outcome classification

qualityScore

readonly qualityScore: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:172

Quality score (0-1)

qualitySignals

readonly qualitySignals: QualitySignals;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:180

Extracted quality signals

routingDecisionId

readonly routingDecisionId: string;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:164

Reference to the routing decision

success

readonly success: boolean;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:170

Overall success indicator

timestamp

readonly timestamp: string;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:166

Timestamp of outcome

tokenUsage

readonly tokenUsage: number;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:176

Token usage

traceId

readonly traceId: string;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:182

Trace ID for correlation

VariantStats

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:87

Variant statistics.

Properties

avgLatencyMs

readonly avgLatencyMs: number;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:101

Average latency in ms

avgReward

readonly avgReward: number;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:99

Average reward

n

readonly n: number;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:93

Number of observations

name

readonly name: string;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:91

Variant name

successes

readonly successes: number;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:95

Success count

successRate

readonly successRate: number;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:97

Success rate

sumLatencyMs

readonly sumLatencyMs: number;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:105

Sum of latencies (for incremental computation)

sumReward

readonly sumReward: number;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:103

Sum of rewards (for incremental computation)

variantId

readonly variantId: string;

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:89

Variant ID

WinLossAnalysis

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:129

Win/loss analysis comparing routing choices.

Properties

losses

readonly losses: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:135

Number of times this model lost (not best outcome)

model

readonly model: string;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:131

Model name

ties

readonly ties: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:137

Number of ties

winRate

readonly winRate: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:139

Win rate

winRateCI

readonly winRateCI: ConfidenceInterval;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:141

Confidence interval for win rate

wins

readonly wins: number;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:133

Number of times this model won (best outcome)

Type Aliases

EffectThresholds

type EffectThresholds = Pick<DistillerConfig, 
  | "failureRateThreshold"
  | "successRateThreshold"
| "latencyRatioThreshold">;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:93

The detector thresholds effectFor normalises a metric against.

ExperimentStatus

type ExperimentStatus = "draft" | "running" | "paused" | "completed" | "archived";

Defined in: packages/nexus-agents/src/learning/ab-test-types.ts:16

Experiment status states.

OutcomeClass

type OutcomeClass = "success" | "partial" | "failure" | "timeout" | "error";

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:50

Task outcome classification.

OutcomeProcessedCallback

type OutcomeProcessedCallback = (decision, outcome, reward) => void;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:351

Callback for when an outcome is processed.

Parameters

decision

outcome

reward

Returns

void

PatternType

type PatternType = "failure-rate" | "success-rate" | "latency-spike";

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:19

The type of pattern detected from outcomes.

RouterType

type RouterType = "linucb" | "preference" | "cascade" | "topsis" | "unattributed";

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:25

Router type that made the routing decision, or unattributed when no scoring stage explains it. The producerless quality member is removed.

Legacy stored quality labels migrate to unattributed. Stored topsis labels remain topsis: before this change the fallback used the same label as measured TOPSIS, so the label alone cannot recover historical attribution. Retain routerTypeMeasured as the persisted evidence for those rows (#5915).

RulesSnapshot

type RulesSnapshot = z.infer<typeof RulesSnapshotSchema>;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-persistence.ts:71

Learning exports - Closed-loop feedback and routing improvement Split from index.ts for file size compliance (Issue #285)

RuleStatus

type RuleStatus = "draft" | "active" | "expired";

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:16

Status lifecycle for a distilled rule.

StrategyAction

type StrategyAction = "penalize" | "boost" | "avoid";

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:22

Action to take when a rule matches a routing candidate.

Variables

DEFAULT_DISTILLER_CONFIG

const DEFAULT_DISTILLER_CONFIG: DistillerConfig;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-types.ts:99

Default distiller configuration.

DEFAULT_FEEDBACK_COLLECTOR_CONFIG

const DEFAULT_FEEDBACK_COLLECTOR_CONFIG: FeedbackCollectorConfig;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:274

Default feedback collector configuration.

DEFAULT_FEEDBACK_INTEGRATION_CONFIG

const DEFAULT_FEEDBACK_INTEGRATION_CONFIG: FeedbackIntegrationConfig;

Defined in: packages/nexus-agents/src/learning/feedback-integration-types.ts:84

Default configuration.

DEFAULT_OUTCOME_STORAGE_CONFIG

const DEFAULT_OUTCOME_STORAGE_CONFIG: {
  autoPruneInterval: 3600000;
  maxRecords: 100000;
};

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:285

Default configuration values.

Type Declaration

autoPruneInterval

readonly autoPruneInterval: 3600000 = 3600000;

maxRecords

readonly maxRecords: 100000 = 100000;

DEFAULT_STATISTICAL_OPTIONS

const DEFAULT_STATISTICAL_OPTIONS: Required<StatisticalOptions>;

Defined in: packages/nexus-agents/src/learning/validation-stats-types.ts:228

Default statistical options.

FeedbackCollectorConfigSchema

const FeedbackCollectorConfigSchema: ZodObject<{
  efficiencyWeight: ZodDefault<ZodNumber>;
  enableAutoReward: ZodDefault<ZodBoolean>;
  maxHistorySize: ZodDefault<ZodNumber>;
  maxPendingDecisions: ZodDefault<ZodNumber>;
  pendingTimeoutMs: ZodDefault<ZodNumber>;
  qualityWeight: ZodDefault<ZodNumber>;
  retryPenalty: ZodDefault<ZodNumber>;
  speedWeight: ZodDefault<ZodNumber>;
  targetDurationMs: ZodDefault<ZodNumber>;
  targetTokenUsage: ZodDefault<ZodNumber>;
}, $strip>;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:290

Zod schema for feedback collector configuration.

FeedbackRoutingDecisionSchema

const FeedbackRoutingDecisionSchema: ZodIntersection<ZodObject<{
  armIndex: ZodOptional<ZodNumber>;
  banditContext: ZodOptional<ZodRecord<ZodString, ZodUnknown>>;
  confidence: ZodOptional<ZodNumber>;
  domain: ZodOptional<ZodString>;
  id: ZodUUID;
  query: ZodString;
  queryFeatures: ZodOptional<ZodRecord<ZodString, ZodUnknown>>;
  selectedModel: ZodString;
  selectedTier: ZodOptional<ZodEnum<{
     strong: "strong";
     weak: "weak";
  }>>;
  timestamp: ZodISODateTime;
  traceId: ZodString;
  ucbScore: ZodOptional<ZodNumber>;
}, $strip>, ZodPipe<ZodObject<{
  routerType: ZodEnum<{
     cascade: "cascade";
     linucb: "linucb";
     preference: "preference";
     quality: "quality";
     topsis: "topsis";
     unattributed: "unattributed";
  }>;
  routerTypeMeasured: ZodOptional<ZodBoolean>;
}, $strip>, ZodTransform<{
  routerType: RouterType;
  routerTypeMeasured?: boolean;
}, {
  routerType:   | "topsis"
     | "linucb"
     | "preference"
     | "quality"
     | "cascade"
     | "unattributed";
  routerTypeMeasured?: boolean;
}>>>;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:136

Zod schema for feedback routing decision.

OutcomeStorageConfigSchema

const OutcomeStorageConfigSchema: ZodObject<{
  autoPruneInterval: ZodOptional<ZodNumber>;
  dbPath: ZodString;
  maxRecords: ZodOptional<ZodNumber>;
}, $strip>;

Defined in: packages/nexus-agents/src/learning/outcome-storage-types.ts:276

Zod schema for OutcomeStorageConfig validation.

QualitySignalsSchema

const QualitySignalsSchema: ZodObject<{
  coherenceScore: ZodOptional<ZodNumber>;
  completionRatio: ZodDefault<ZodNumber>;
  lintErrors: ZodOptional<ZodNumber>;
  retryCount: ZodDefault<ZodNumber>;
  testsPass: ZodOptional<ZodBoolean>;
  userApproved: ZodOptional<ZodBoolean>;
  validStructure: ZodOptional<ZodBoolean>;
}, $strip>;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:75

Zod schema for quality signals.

RulesSnapshotSchema

const RulesSnapshotSchema: ZodObject<{
  eligibleOutcomes: ZodOptional<ZodNumber>;
  rules: ZodArray<ZodObject<{
     action: ZodEnum<{
        avoid: "avoid";
        boost: "boost";
        penalize: "penalize";
     }>;
     category: ZodString;
     cli: ZodUnion<readonly [ZodEnum<{
        claude: "claude";
        codex: "codex";
        gemini: "gemini";
        opencode: "opencode";
     }>, ZodTemplateLiteral<`api:${string}`>]>;
     confidence: ZodNumber;
     createdAt: ZodNumber;
     effect: ZodOptional<ZodNumber>;
     id: ZodString;
     metric: ZodNumber;
     observationCount: ZodNumber;
     patternType: ZodEnum<{
        "failure-rate": "failure-rate";
        "latency-spike": "latency-spike";
        "success-rate": "success-rate";
     }>;
     status: ZodEnum<{
        active: "active";
        draft: "draft";
        expired: "expired";
        promoted: "promoted";
     }>;
     support: ZodOptional<ZodNumber>;
     updatedAt: ZodNumber;
  }, $strip>>;
  savedAt: ZodString;
  version: ZodLiteral<1>;
}, $strip>;

Defined in: packages/nexus-agents/src/learning/strategy-distiller-persistence.ts:60

Versioned snapshot schema for atomic saves.

TaskOutcomeSchema

const TaskOutcomeSchema: ZodObject<{
  durationMs: ZodNumber;
  errorMessage: ZodOptional<ZodString>;
  outcomeClass: ZodEnum<{
     error: "error";
     failure: "failure";
     partial: "partial";
     success: "success";
     timeout: "timeout";
  }>;
  qualityScore: ZodNumber;
  qualitySignals: ZodObject<{
     coherenceScore: ZodOptional<ZodNumber>;
     completionRatio: ZodDefault<ZodNumber>;
     lintErrors: ZodOptional<ZodNumber>;
     retryCount: ZodDefault<ZodNumber>;
     testsPass: ZodOptional<ZodBoolean>;
     userApproved: ZodOptional<ZodBoolean>;
     validStructure: ZodOptional<ZodBoolean>;
  }, $strip>;
  routingDecisionId: ZodUUID;
  success: ZodBoolean;
  timestamp: ZodISODateTime;
  tokenUsage: ZodNumber;
  traceId: ZodString;
}, $strip>;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-types.ts:188

Zod schema for task outcome.

Functions

calculateDistributionStats()

function calculateDistributionStats(values): DistributionStats;

Defined in: packages/nexus-agents/src/learning/validation-stats.ts:173

Calculate descriptive statistics for a distribution.

Parameters

values

readonly number[]

Returns

DistributionStats

calculateMinSampleSize()

function calculateMinSampleSize(
   baselineRate, 
   minimumDetectableEffect, 
   options?
): number;

Defined in: packages/nexus-agents/src/learning/validation-stats.ts:349

Calculate minimum sample size for detecting a difference in proportions. Uses formula for two-proportion z-test power analysis.

Parameters

baselineRate

number

minimumDetectableEffect

number

options?

alpha?

number

power?

number

Returns

number

calculateRegret()

function calculateRegret(decisions): RegretAnalysis;

Defined in: packages/nexus-agents/src/learning/validation-stats.ts:228

Calculate regret analysis comparing actual decisions vs oracle (best possible).

Parameters

decisions

readonly { actualReward: number; chosenModel: string; rewards: Record<string, number>; }[]

Returns

RegretAnalysis

calculateWinLoss()

function calculateWinLoss(
   model, 
   decisions, 
   options?
): WinLossAnalysis;

Defined in: packages/nexus-agents/src/learning/validation-stats.ts:293

Calculate win/loss analysis for a model.

Parameters

model

string

decisions

readonly { actualReward: number; chosenModel: string; rewards: Record<string, number>; }[]

options?

Returns

WinLossAnalysis

compareProportions()

function compareProportions(
   successes1, 
   total1, 
   successes2, 
   total2, 
   options?
): ComparisonResult;

Defined in: packages/nexus-agents/src/learning/validation-stats.ts:128

Compare two proportions using two-proportion z-test.

Parameters

successes1

number

total1

number

successes2

number

total2

number

options?

Returns

ComparisonResult

computeOutcomeReward()

function computeOutcomeReward(collector, outcome): ComputedReward;

Defined in: packages/nexus-agents/src/learning/feedback-integration.ts:511

Helper to compute reward from outcome.

Parameters

collector

IOutcomeFeedback

outcome

Returns

OutcomeFeedbackCollector

createAbTestTracker()

function createAbTestTracker(): IAbTestTracker;

Defined in: packages/nexus-agents/src/learning/ab-test-tracker.ts:413

Create a default A/B test tracker instance.

Returns

IAbTestTracker

createFeedbackIntegration()

function createFeedbackIntegration(config?, collector?): IFeedbackIntegration;

Defined in: packages/nexus-agents/src/learning/feedback-integration.ts:501

Creates a FeedbackIntegration instance.

Parameters

config?

Partial<FeedbackIntegrationConfig>

collector?

Returns

IFeedbackIntegration

createOutcomeFeedbackCollector()

function createOutcomeFeedbackCollector(config?): OutcomeFeedbackCollector;

Defined in: packages/nexus-agents/src/learning/outcome-feedback.ts:406

Create an OutcomeFeedbackCollector instance.

Parameters

config?

Partial<FeedbackCollectorConfig>

Returns

OutcomeFeedbackCollector

createOutcomeStorage()

function createOutcomeStorage(config): SQLiteOutcomeStorage;

Defined in: packages/nexus-agents/src/learning/outcome-storage.ts:374

Create an SQLite outcome storage instance.

Parameters

config

OutcomeStorageConfig

Returns

SQLiteOutcomeStorage

createRoutingDecision()

function createRoutingDecision(params): FeedbackRoutingDecision;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-helpers.ts:115

Create a routing decision record.

Parameters

params

Omit<FeedbackRoutingDecision, "id" | "timestamp">

Returns

createStrategyDistiller()

function createStrategyDistiller(
   outcomeStore, 
   logger?, 
   config?
): StrategyDistiller;

Defined in: packages/nexus-agents/src/learning/strategy-distiller.ts:459

Factory function for creating StrategyDistiller.

Parameters

outcomeStore

OutcomeStore

logger?

ILogger

config?

Partial<DistillerConfig>

Returns

StrategyDistiller

createTaskOutcome()

function createTaskOutcome(params): TaskOutcome;

Defined in: packages/nexus-agents/src/learning/outcome-feedback-helpers.ts:128

Create a task outcome record.

Parameters

params

Omit<TaskOutcome, "timestamp">

Returns

detectFailurePatterns()

function detectFailurePatterns(groups, threshold): DetectedPattern[];

Defined in: packages/nexus-agents/src/learning/strategy-distiller.ts:136

Detect groups with failure rate above threshold.

Parameters

groups

readonly OutcomeGroup[]

threshold

number

Returns

DetectedPattern[]

detectLatencyPatterns()

function detectLatencyPatterns(groups, threshold): DetectedPattern[];

Defined in: packages/nexus-agents/src/learning/strategy-distiller.ts:182

Detect groups with latency spike (p90/median > threshold).

Parameters

groups

readonly OutcomeGroup[]

threshold

number

Returns

DetectedPattern[]

detectSuccessPatterns()

function detectSuccessPatterns(groups, threshold): DetectedPattern[];

Defined in: packages/nexus-agents/src/learning/strategy-distiller.ts:159

Detect groups with success rate above threshold.

Parameters

groups

readonly OutcomeGroup[]

threshold

number

Returns

DetectedPattern[]

effectFor()

function effectFor(
   patternType, 
   metric, 
   thresholds
): number;

Defined in: packages/nexus-agents/src/learning/strategy-distiller.ts:67

Effect size of a detected pattern in [0, 1] (#5004 finding 3): how far the metric sits past the detector threshold that produced it, normalised.

failure-rate: (rate − failureRateThreshold) / (1 − failureRateThreshold)
success-rate: (rate − successRateThreshold) / (1 − successRateThreshold)
latency-spike: min(1, (ratio − latencyRatioThreshold) / latencyRatioThreshold)

A metric exactly at its threshold is 0: the detector fired, but there is no margin to act on. A metric at the far end of its scale is 1. Never NaN — a threshold with no headroom (failure threshold 1.0, latency threshold 0) or a NaN metric yields 0, so baseDelta × confidence in routing stays a number.

Parameters

patternType

PatternType

metric

number

thresholds

EffectThresholds

Returns

number

meanConfidenceInterval()

function meanConfidenceInterval(values, options?): ConfidenceInterval;

Defined in: packages/nexus-agents/src/learning/validation-stats.ts:82

Calculate confidence interval for a mean.

Parameters

values

readonly number[]

options?

Returns

ConfidenceInterval

proportionConfidenceInterval()

function proportionConfidenceInterval(
   successes, 
   total, 
   options?
): ConfidenceInterval;

Defined in: packages/nexus-agents/src/learning/validation-stats.ts:32

Calculate confidence interval for a proportion (success rate). Uses Wilson score interval for better coverage at extreme proportions.

Parameters

successes

number

total

number

options?