@bivelio/savings-layer 0.4.13 → 0.4.15

package/dist/index.d.ts

@@ -58,6 +58,19 @@ declare class SavingsLedger {
     sentInputLocal: number;
     /** Tokens que la capa quitó de ese prompt, en cuenta LOCAL. */
     removedInputLocal: number;
+    /**
+     * Tokens (LOCAL ruler) that tool_result ELISION removed from the sent prompt.
+     * They exist to be SUBTRACTED from `removedInputLocal`: live deletion really
+     * lowers `sentInputLocal`, so without this correction the elision saving would
+     * enter the reconstructed baseline and the commission, labelled "estimated".
+     * The doctrine is the opposite — "live count, shadow money": elision is a COUNT
+     * (ledger 2 `eliminatedInput`), its dollar is measured by the shadow lane out
+     * of commission, and `basis` stays "measured". Deliberately distinct from the
+     * tool-selector drop, which DOES count in the baseline (a declared tool that is
+     * not sent is an attributable saving; a superseded tool_result is redundancy
+     * not yet billed as verifiable against the invoice).
+     */
+    elididoLocal: number;
     /**
      * ¿El estimador que produjo las dos cifras anteriores cuenta con la regla del
      * proveedor? Lo fija el pipeline al resolver el tokenizador del modelo.
@@ -169,7 +182,7 @@ declare class SavingsLedger {
  * catalogue stays out of the way.
  */
 /** When these prices were last checked against the providers' public rates. */
-declare const PRECIOS_AL_DIA = "2026-08-24";
+declare const PRECIOS_AL_DIA = "2026-09-06";
 /**
  * Key = the model id as the provider publishes it, without a vendor prefix.
  * Lookup tries the exact id, then the id without its provider prefix, then the
@@ -296,6 +309,27 @@ interface SavingsToggles {
     exactCache?: boolean;
     semanticCache?: boolean | "safe-only";
     compression?: boolean | "auto";
+    /**
+     * OPT-IN. Elide old, large tool_results from a multi-turn context before the
+     * request is sent, replacing each one's body with a short stable stub while
+     * keeping its `role:"tool"`, `toolCallId` and position. This is the v0
+     * (age × size) policy the shadow lane measures — productionized to actually
+     * shorten the outgoing prompt, on the generic tool_result segment nothing else
+     * in the pipeline touches.
+     *
+     * IT ONLY RUNS WHEN THE POLICY ALLOWS IT. Like the semantic cache, elision is
+     * disabled for write/destructive tools, freshness-sensitive, personalized or
+     * stochastic requests, and for `freshnessBucket: "day"` — dropping the output
+     * of a past action, or old context a fresh answer still needs, is exactly the
+     * degradation the layer must not cause.
+     *
+     * FIDELITY FLOOR. The newest tool_results are never elided (a keep-floor of at
+     * least 8, imported from the provider dial S4a); v0's own age≥2 rule is weaker
+     * than that floor, so the floor wins. What it removes is a COUNT on the panel
+     * (`eliminated_tokens`); its money value is estimated only in the shadow lane
+     * and never enters your commission — your invoice stays the source of truth.
+     */
+    elision?: boolean;
     serialization?: boolean | "auto";
     singleflight?: boolean;
     deterministicResolvers?: boolean;
@@ -1255,6 +1289,13 @@ interface Policy {
     allowSingleflight: boolean;
     /** May we run natural-language compression on the context? */
     allowCompression: boolean;
+    /**
+     * May the SDK elide old, large tool_results before resending? Computed, not
+     * hardcoded like `allowCompression`: eliding drops whole tool outputs, so it
+     * honors the same vetoes as the semantic cache — write tools, freshness,
+     * personalization, stochastic sampling and `day`.
+     */
+    allowElision: boolean;
     /** Does the request demand current/live data? */
     requiresFreshData: boolean;
     /** Does the request have (or request) side effects? */
@@ -2362,6 +2403,22 @@ interface AnthropicOptions {
      * pointless: the extra breakpoints are rejected by the API, not ignored.
      */
     maxCacheBreakpoints?: number;
+    /**
+     * When true, requests that the pipeline decided to mark also carry the
+     * request-level `cache_control` (Anthropic's automatic caching): the provider
+     * places one extra breakpoint on the last cacheable block and moves it
+     * forward as the conversation grows, so the history beyond the declared
+     * stable prefix is cached incrementally too. It spends one of the four
+     * breakpoint slots and never fires when the economic gate declined to mark.
+     *
+     * Off by default, and deliberately so: the moving breakpoint writes the
+     * volatile tail of EVERY marked request at the 1.25x write premium. That
+     * pays for itself on conversational or agentic traffic — the next turn reads
+     * the whole history at 0.1x — and loses money on one-shot traffic where
+     * nothing ever reads it back. It is also rejected by legacy Bedrock
+     * deployments. Turn it on when your traffic is multi-turn.
+     */
+    automaticCache?: boolean;
     /**
      * Whether a request carrying BOTH the caller's tools and an output contract
      * may send `tool_choice: {type: "any"}`, which obliges the model to answer
@@ -3262,6 +3319,6 @@ declare function createSavingsLayer(config: Omit<SavingsLayerConfig, "tokenizer"
  * constant to package.json so the two can never drift.
  */
 
-declare const BV_SALA_VERSION = "0.4.13";
+declare const BV_SALA_VERSION = "0.4.15";
 
 export { type AnthropicOptions, AnthropicProvider, BV_SALA_VERSION, type BivelioHandoff, CATALOGO, COMPACT_ROWS_INSTRUCTION, type CacheOutcome, type CacheTier, type CachedAnswer, type CanonicalRequest, type ChatMessage, type CompactRowsDecodeOptions, type CompactRowsOptions, type CompactVerification, type CompressionEconomics, type CompressionPolicyInput, type CompressionRequest, type CompressionResult, type Compressor, type ConstraintBackend, type ConstraintBackendId, type ConstraintPlan, type ConversationState, type CostBasis, type CostModel, DEFAULT_COST_MODEL, DEFAULT_MAX_STALENESS_SECONDS, type DeferredExecutionReport, type DeterministicResolver, EMBEDDED_PUBLIC_KEY, type EmbeddingFn, type EvidenceDoc, type ExactCache, ExactTokenizer, type ExecutionClass, ExecutionRouter, type FormatChoice, FormatSelector, type FormatSelectorOptions, type HandoffFact, HeuristicEstimator, InMemoryExactCache, InMemorySemanticCache, type JsonSchema, LicenseError, type LicenseGate, type LicenseKeyset, type LicenseStatus, type LicenseTokenPayload, type LocalClassifier, MODELOS_POR_CODIFICACION, MODELS_BY_ENCODING, type MemoryDelta, type MessageRole, MockProvider, type MockProviderOptions, type ModelCascadeMode, type ModelCascadeReport, type ModelCascadeSettings, type OpenAICompatibleOptions, OpenAICompatibleProvider, type OutputContract, PRECIOS_AL_DIA, type Policy, type ProtectedSpan, type Provider, type ProviderRequest, type ProviderResponse, type ProviderUsage, QUALITY_STATUSES, type QualityStatus, type RequestContext, type Reranker, type ResolvedDeps, type ResolvedGate, type ResolvedLicense, ResolverRegistry, type ResolverResult, type RetrievalOptions, type RetrievalOutcome, type RetrievalPlan, type Retriever, type RevocationListBody, type RevocationListVerification, type RevocationPollEvent, RevocationPoller, type RevocationPollerOptions, RevocationSnapshot, type RouteDecision, type SavingsLayer, type SavingsLayerConfig, SavingsLedger, type SavingsReport, type SavingsRequest, type SavingsResult, type SavingsToggles, SchemaRegistry, type SemanticCacheOptions, type SemanticHit, type SemanticRecord, type SemanticVerifier, type SerializationFormat, type Serializer, Singleflight, type TokenEstimator, TokenizerRegistry, type ToolCallRequest, type ToolDescriptor, type ToolRisk, type ToolSelection, ToolSelector, type ToolSelectorOptions, TraceCollector, type TraceEvent, UMBRAL_LOCAL, UsageEmitter, type UsageEmitterOptions, type ValidationError, type ValidationResult, adaptiveK, arithmeticResolver, buildExactKey, buildInflightKey, canonicalize, compactRowsSerializer, compresionFiel, computeNextDelay, cosine, createCompactRowsSerializer, createDefaultResolvers, createHandoff, createMockProvider, createSavingsLayer, csvSerializer, decodeCompactRows, dedupe, defaultTokenizerRegistry, deriveSeatId, deriveStatus, detectCi, embebedorLocal, emptyState, encodeCompactRows, evalArithmetic, evaluatePolicy, executeRetrievalPlan, gateFromKey, heuristicEstimator, isKeyset, isStrictCompatible, jsonCompactSerializer, keyTokensMatch, liveGateFromKey, mergeDelta, modelFamilyOf, multisetEqual, negationCount, noopCompressor, normalizeText, numericTokens, precioDeLista, projectContext, providerNativeBackend, randomId, registerExactTokenizer, registrarTokenizadorExacto, renderState, resolveConstraintBackend, resolveDependencies, resolveLicense, sha256, shouldCompress, stableStringify, uniformColumns, validateAgainstSchema, vectorLocal, verifyCompactSignature, verifyLicenseToken, verifyRevocationList, withProtectedSpans, words };