hippo-memory 2.0.0 → 2.1.0

package/dist/eval/eval-stats.js

@@ -1,5 +1,9 @@
-/** Statistics and cost accounting for the token-efficiency evals: cost is priced over four buckets (uncached input, cache write, cache read, output).
- *  Uncertainty uses paired, cluster (same-repository tasks are not independent) and paired-ratio bootstraps; every resampler takes a seed. */
+/**
+ * Statistics and cost accounting for the token-efficiency evals: cost is priced over four buckets (uncached input,
+ * cache write, cache read, output).
+ * Uncertainty uses paired, cluster (same-repository tasks are not independent) and paired-ratio bootstraps; every
+ * resampler takes a seed.
+ */
 const DEFAULT_ITERATIONS = 5000;
 const DEFAULT_TWO_LEVEL_ITERATIONS = 10_000;
 const DEFAULT_ALPHA = 0.05;
@@ -48,7 +52,10 @@ function percentileInterval(samples, alpha) {
 function mean(xs) {
     return xs.length === 0 ? 0 : xs.reduce((s, x) => s + x, 0) / xs.length;
 }
-/** Paired bootstrap for the mean of per-task differences (treatment minus control, same task); an interval excluding zero is the bar for a real difference. */
+/**
+ * Paired bootstrap for the mean of per-task differences (treatment minus control, same task); an interval excluding
+ * zero is the bar for a real difference.
+ */
 export function pairedBootstrap(diffs, opts = {}) {
     const iterations = opts.iterations ?? DEFAULT_ITERATIONS;
     const alpha = opts.alpha ?? DEFAULT_ALPHA;
@@ -65,7 +72,10 @@ export function pairedBootstrap(diffs, opts = {}) {
     }
     return { estimate: mean(diffs), ...percentileInterval(samples, alpha), iterations };
 }
-/** Cluster bootstrap for the mean of per-task differences: resamples whole clusters (e.g. one repository), because tasks in a cluster are not independent. */
+/**
+ * Cluster bootstrap for the mean of per-task differences: resamples whole clusters (e.g. one repository), because tasks
+ * in a cluster are not independent.
+ */
 export function clusteredPairedBootstrap(diffsByCluster, opts = {}) {
     const iterations = opts.iterations ?? DEFAULT_ITERATIONS;
     const alpha = opts.alpha ?? DEFAULT_ALPHA;
@@ -94,7 +104,9 @@ function costPerResolved(outcomes) {
     const cost = outcomes.reduce((s, o) => s + o.cost, 0);
     return resolved === 0 ? Number.POSITIVE_INFINITY : cost / resolved;
 }
-/** Cost per resolved task in two arms on the same tasks, with a paired bootstrap (`control[i]` and `treatment[i]` must be the same task).
+/**
+ * Cost per resolved task in two arms on the same tasks, with a paired bootstrap (`control[i]` and `treatment[i]` must
+ * be the same task).
  *  Resamples in which an arm resolves nothing are dropped; `iterations` reports how many were kept. */
 export function costPerResolvedDelta(control, treatment, opts = {}) {
     if (control.length !== treatment.length) {

package/dist/eval/eval-suite.d.ts

@@ -1,4 +1,6 @@
-/** Self-contained feature eval suite: a synthetic corpus with known ground truth, per-feature metrics and regression detection against a saved baseline.
+/**
+ * Self-contained feature eval suite: a synthetic corpus with known ground truth, per-feature metrics and regression
+ * detection against a saved baseline.
  *  Zero API calls (no LLM judge, no embeddings), deterministic (fixed timestamps, content, IDs) and fast. */
 import { type MemoryEntry } from '../core/memory.js';
 export type FeatureCategory = 'direct-recall' | 'extraction-preference' | 'dag-drilldown' | 'temporal' | 'noise-resistance' | 'multi-hop';