@zakkster/lite-pick 0.5.0 → 0.7.0

package/CHANGELOG.md

@@ -4,6 +4,68 @@ All notable changes to `@zakkster/lite-pick` are documented here. The format fol
 [Keep a Changelog](https://keepachangelog.com/en/1.1.0/), and the project adheres to
 [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.6.0] - 2026-09-23
+
+M6: the benchmark suite (ROADMAP.md M6). An EVIDENCE session -- no API change. `Pick.js` and
+`Pool.js` are byte-identical to 0.5.0 apart from the `VERSION` stamp; the kernel gates
+(torture / PerfGate / witness / balance / fuzz) are unchanged and green. Everything added lives
+under `benchmark/` and is NOT in the published tarball (`files[]` unchanged).
+
+### Added
+
+- `benchmark/GcBlastRadius.mjs` -- the GC blast-radius headline (dimension 3). The same sustained
+  mixed workload (n=1024, 2,000,000 requests) through two lanes: the lite-pick lane (`P2cBalancer`
+  over caller-owned typed arrays) measures `major=0`, pick `B/op=0`, `maxPause` ~0.1-0.3ms; the
+  allocating foil lane (collect-candidates-sort idiom with a retained in-flight request context)
+  measures `major=13-14`, `maxPause` ~2.5-4ms. GC sampled via `@zakkster/lite-gc-profiler`
+  (`GcProfiler` + `checkNoGc`, the torture.mjs machinery); `B/op` via `measureAllocs`.
+- `benchmark/Fairness.mjs` -- weighted convergence + burstiness (dimension 8). SmoothWRR weights
+  [10,3,2,1]: exact fairness (counts == k*weight) over 500 cycles, max-run 3 vs the bursty
+  weight-expansion foil's 10. SED weights [1,2,3,4,6,8,12,16]: worst share drift < 0.0001 from the
+  weight target; weighted-imbalance ~0.00 vs a weight-blind random foil's ~5.5.
+- `benchmark/Disruption.mjs` -- consistent-hash disruption (dimension 7). Naive-modulo foil over
+  100,000 seeded keys remaps 98.4% (remove node 64->63) / 98.5% (add node 64->65) of keys vs a good
+  consistent hash's ~1.6% / ~1.5% ideal. Explicit SKIP row for `ConsistentHash` (Maglev, M8) -- no
+  stub in `Pick.js`.
+- `benchmark/Report.mjs` -- emits `benchmark/results.json` stamping Node version, V8, CPU model,
+  core count, arch, OS, and every PRNG seed alongside the measured numbers; renders the README
+  `<!-- bench:ID -->` fenced tables from it. `--verify` mode is the `bench:verify` drift check:
+  ALGORITHMIC numbers (balance peak-gap, disruption remap %) recomputed FRESH and compared EXACT;
+  TIMING numbers (GC pauses) compared to `results.json` within +/-15%. Exits non-zero on any drift.
+  (Fixes the previously-broken `bench:report` script, which pointed at an absent `Report.mjs`.)
+- `benchmark/Soak.mjs` -- the endurance-soak scaffold (post-1.0 #8): ONE P2C lane, continuous
+  mixed-chaos load (flap storms + whole-pool-down troughs + load feedback), emits a JSONL
+  time-series to `benchmark/soak.jsonl`, and RUNS `test/invariants.mjs:checkBase` at every
+  checkpoint. `tracker.size()` returns to 0 after each cycle; invariants green at all checkpoints.
+  `SOAK_CYCLES=0` runs forever -- the harness the `caffeinate -i` overnight burn-in plugs into.
+- `benchmark/Matrix.mjs` -- extended (not rewritten): each SUBJECT gained a `dims` flag, and the
+  file now exports the SHARED SEEDED workload matrix (`buildWorkload` over uniform / skewed-weight
+  / skewed-cost), `SEEDS`, `SUBJECTS`, `SIZES`, and `measureThroughput`, reused by every dimension
+  file. The standalone throughput runner is behind a main guard so importing it runs no sweep.
+- `decisions/0008-benchmark-suite.md` -- the ADR: the parity framing, the real-competitors + foils
+  baseline, the drift-check teeth, and the ConsistentHash SKIP.
+- README: the *Evidence* section -- balance anchor + GC blast-radius headlines (fenced), the
+  consistent-hash disruption trust gate, the honest COLD-path cost table, and the vs-AWS NLB/ALB
+  "complementary, not a competitor" positioning.
+- devDependencies: `load-balancers@1.3.52`, `loadbalance@1.0.0`, `wrr@1.0.0` -- REAL npm
+  competitors, pinned to exact versions, loaded via `createRequire` (all three are CommonJS) and
+  timed into `results.json` beside our in-repo foils; a package that fails to load becomes a
+  labeled `unavailable` row rather than being silently dropped. Each incumbent is rendered side by
+  side with the SAME-complexity lite-pick strategy on the SAME n=1024 pool in the README
+  `<!-- bench:competitors -->` throughput fence. Parity is claimed only on EQUAL-work rows: P2C is
+  parity (~60k vs ~61k ops/ms); RoundRobin is ~22% slower (~260k vs ~335k) and the gap is OWNED --
+  `loadbalance` is a bare `i++ % n` with no liveness, while lite-pick's `RoundRobinBalancer`
+  forward-scans the eligibility bitmap to skip down nodes (never a dead pick), and that scan is the
+  constant-factor cost of a guarantee the incumbents do not offer. The weighted-random row is a
+  disclosed SKIP -- lite-pick's O(1) `WeightedRandom` (alias table) lands at M10, so `wrr` is not
+  raced against our O(cap) `SmoothWRRBalancer` (a different complexity class). No "faster"/"Nx".
+- Scripts: `bench:gc`, `bench:fairness`, `bench:disruption`, `bench:verify`, `soak`.
+
+### Changed
+
+- Version bumped to 0.6.0 in the three sync sites (`package.json`, `Pick.js` `VERSION`, `llms.txt`).
+  `peerDependencies` stays `{}`.
+
 ## [0.5.0] - 2026-09-23
 
 M5: the ergonomic request layer at the `@zakkster/lite-pick/pool` subpath -- dispatch/settle

package/Pick.d.ts

@@ -1,9 +1,9 @@
 /**
  * @zakkster/lite-pick -- TypeScript declarations.
  *
- * M4 (0.4.0): substrate seams + RoundRobin + SmoothWRR + P2C + the exact LeastConn family
- * (LeastConn/SED/NQ). The remaining strategy classes (PeakEWMA, ConsistentHash, BoundedLoad,
- * WeightedRandom) are added one per session.
+ * M7 (0.7.0): substrate seams + RoundRobin + SmoothWRR + P2C + the exact LeastConn family
+ * (LeastConn/SED/NQ) + PeakEWMA (latency-aware P2C). The remaining strategy classes
+ * (ConsistentHash, BoundedLoad, WeightedRandom) are added one per session.
  */
 
 /** The single source-of-truth version stamp. */
@@ -160,3 +160,30 @@ export class NqBalancer extends BalancerBase {
     /** The first idle eligible node, else the SED minimum, or `PICK_NONE`. O(cap). */
     pick(): number;
 }
+
+/**
+ * PeakEwmaBalancer -- latency-aware power-of-two-choices (M7, Twitter Finagle's peak-EWMA).
+ * Draws two distinct eligible endpoints and returns the lower cost = `(inflight + 1) * ewmaAt(now)`;
+ * a slow endpoint (high decayed EWMA rtt) is avoided even with a short queue. `inflight` is the
+ * caller-owned Uint32Array read LIVE; the EWMA state (`_ewma` / `_stamp`, Float64) is BALANCER-OWNED
+ * and written ONLY by `recordRtt` (the warm feedback path). `pick(now)` decays on READ -- never
+ * writes -- so it is 0 B/op, as is `recordRtt`. `now` / `sampleNs` are caller-supplied nanoseconds.
+ * Cold start seeds the EWMA to 1.0 -> graceful least-connections, never NaN. O(d)=O(1). Fails
+ * closed (`PICK_NONE`) when the whole pool is down.
+ */
+export class PeakEwmaBalancer extends BalancerBase {
+    /**
+     * @param capacity endpoint count (fixed).
+     * @param eligible shared view: 1 = pickable, 0 = down (length >= capacity).
+     * @param inflight per-endpoint in-flight counts (length >= capacity), caller-owned, read live.
+     * @param tauNs the EWMA time-constant / half-life in nanoseconds (finite, > 0).
+     * @param seed deterministic PRNG seed (default 0x9e3779b9); reproducible benches.
+     */
+    constructor(capacity: number, eligible: Uint8Array, inflight: Uint32Array, tauNs: number, seed?: number);
+    /** The decayed EWMA rtt estimate for endpoint `i` at time `now` (ns). Pure read, zero-alloc. */
+    ewmaAt(i: number, now: number): number;
+    /** Warm feedback path: record an rtt sample (ns) for endpoint `i` at time `now` (ns). 0 B/op. */
+    recordRtt(i: number, sampleNs: number, now: number): void;
+    /** Pick by latency-aware power-of-two-choices at time `now` (ns), or `PICK_NONE`. O(d)=O(1). */
+    pick(now?: number): number;
+}

package/Pick.js

@@ -1,8 +1,8 @@
 /**
  * @zakkster/lite-pick -- zero-GC load-balancing SELECTION KERNEL.
  *
- * M4 (0.4.0): substrate seams + six strategies -- RoundRobin, SmoothWRR, P2C, and the
- * exact LeastConn family (LeastConn, SED, NQ). This file ships:
+ * M7 (0.7.0): substrate seams + seven strategies -- RoundRobin, SmoothWRR, P2C, the exact
+ * LeastConn family (LeastConn, SED, NQ), and PeakEWMA (latency-aware P2C). This file ships:
  *
  *   - VERSION        the single source-of-truth version stamp (3-place sync).
  *   - PICK_NONE      the fail-closed sentinel (-1): "no endpoint", never a dead pick.
@@ -26,6 +26,11 @@
  *                    charges the NEW request's marginal cost. O(cap)/pick, 0 B/op.
  *   - NqBalancer     never-queue (IPVS `nq`): an IDLE eligible endpoint immediately if one
  *                    exists, else SED. The worker-pool fit. O(cap)/pick, 0 B/op.
+ *   - PeakEwmaBalancer  latency-aware P2C (Twitter Finagle's peak-EWMA): draws two distinct
+ *                    eligible endpoints and takes the lower cost = (inflight+1) x decayed EWMA(rtt).
+ *                    Decay-on-READ (pick() never writes -> 0 B/op); the balancer OWNS the Float64
+ *                    _ewma/_stamp state and is its SOLE writer via the warm recordRtt() feedback
+ *                    path (also 0 B/op). Caller-supplied nanosecond clock. O(d)=O(1)/pick.
  *
  * The identity (decisions/0001): lite-pick OWNS NO mutable state it can avoid owning.
  * It reads pre-allocated views (eligibility, inflight, weights, scores) that siblings or
@@ -33,7 +38,7 @@
  * counters live OUTSIDE the kernel. The steady-state pick path allocates 0 B/op.
  *
  * Roster (one strategy per session -- see ROADMAP.md): RoundRobin [M1], SmoothWRR [M2],
- *   P2C [M3], LeastConn/SED/NQ [M4], PeakEWMA, ConsistentHash, BoundedLoad, WeightedRandom
+ *   P2C [M3], LeastConn/SED/NQ [M4], PeakEWMA [M7], ConsistentHash, BoundedLoad, WeightedRandom
  *   [planned]. The EXACT-O(log n) fewest-in-flight variant is a deferred @zakkster/lite-logn
  *   BinaryHeap optional-peer seam (decisions/0006), not this exact-O(cap) scan.
  *
@@ -46,7 +51,7 @@
  */
 
 /** Version stamp. Synced across package.json and llms.txt (three-place rule). */
-export const VERSION = '0.5.0';
+export const VERSION = '0.7.0';
 
 /**
  * Fail-closed sentinel returned by pick() when no endpoint is eligible.
@@ -574,3 +579,138 @@ export class NqBalancer extends BalancerBase {
         return best;                              // -1 when every eligible node has weight 0
     }
 }
+
+/**
+ * PeakEwmaBalancer -- latency-aware power-of-two-choices (M7), Twitter Finagle's peak-EWMA.
+ *
+ * `pick(now)` draws TWO distinct eligible endpoints (the same rejection-sampling machinery as
+ * P2cBalancer -- reused verbatim, not re-implemented) and returns the one with the lower COST,
+ * where cost(i) = (inflight[i] + 1) x ewmaAt(i, now). It is P2C over a LATENCY signal instead of
+ * raw in-flight count: a slow endpoint (high EWMA rtt) is avoided even when its queue is short,
+ * so the pool steers around a degraded-but-up node -- the strategy the multi-region FE case wants.
+ * O(d) = O(1) per pick.
+ *
+ * Ownership (ADR 0001, ADR 0009): `inflight` is the CALLER's Uint32Array, read LIVE (the P2C /
+ * LeastConn seam). The EWMA state -- `_ewma` (the decayed rtt estimate) and `_stamp` (the ns
+ * timestamp of each node's last update), both Float64Array -- is BALANCER-OWNED (the SmoothWRR
+ * precedent: a strategy may own algorithm state), and the balancer is its SOLE writer, via the
+ * warm `recordRtt()` feedback path. `pick()` NEVER writes: it decays ON READ, so the hot path
+ * stays a pure read -> 0 B/op.
+ *
+ * Decay-on-read: ewmaAt(i, now) = _ewma[i] x exp(-(now - _stamp[i]) / tau). No write, no clock
+ * call on the gated path -- `now` (and the rtt sample) are CALLER-supplied nanoseconds, consistent
+ * between `pick(now)` and `recordRtt(i, sampleNs, now)`, so the whole strategy is deterministic
+ * and testable and allocates nothing.
+ *
+ * Cold start: `_ewma` seeds to 1.0 and `_stamp` to 0 at construction, so before any sample
+ * cost(i) ~ (inflight[i] + 1) x 1 and PeakEWMA degrades GRACEFULLY to plain least-connections
+ * (P2C-over-inflight). It is never NaN.
+ *
+ * Anti-flap (ADR 0002, ADR 0009): the EWMA half-life IS the smoothing -- a single slow sample
+ * snaps the cost up instantly and it decays back over ~tau, so there is NO extra dwell/hysteresis.
+ *
+ * Deferred (ADR 0009 / llms.txt): a p99-aware variant scoring inflight x p99Rtt via a per-node
+ * @zakkster/lite-sketch `DDSketch` (optional peer, 0 B/op `add`). EWMA-mean is the shipped,
+ * zero-peer default; `peerDependencies` stays empty until a shipped path imports the sketch.
+ *
+ * Bound: O(d) = O(1) per pick (two expected-O(1) rejection draws + two exp() + a compare),
+ * 0 B/op on BOTH `pick()` and `recordRtt()` (torture + PerfGate). Fails closed (PICK_NONE) when
+ * the whole pool is down.
+ */
+export class PeakEwmaBalancer extends BalancerBase {
+    /**
+     * @param {number} capacity  endpoint count (fixed).
+     * @param {Uint8Array} eligible  shared view: 1 = pickable, 0 = down (length >= capacity).
+     * @param {Uint32Array} inflight  per-endpoint in-flight counts (length >= capacity),
+     *   caller-owned and only READ here.
+     * @param {number} tauNs  the EWMA time-constant / half-life in nanoseconds (> 0, finite):
+     *   larger tau = slower decay = longer memory of a latency spike.
+     * @param {number} [seed=0x9e3779b9]  deterministic PRNG seed (reproducible benches).
+     */
+    constructor(capacity, eligible, inflight, tauNs, seed = 0x9e3779b9) {
+        super(capacity, eligible);
+        // Validate typeof-first, BEFORE allocating the owned Float64 state (fail closed early).
+        if (!(inflight instanceof Uint32Array) || inflight.length < capacity) {
+            throw new RangeError('[lite-pick] inflight must be a Uint32Array of length >= capacity');
+        }
+        if (typeof tauNs !== 'number') {
+            throw new TypeError('[lite-pick] tauNs must be a number');
+        }
+        if (!Number.isFinite(tauNs) || tauNs <= 0) {
+            throw new RangeError('[lite-pick] tauNs must be a finite number > 0');
+        }
+        this._inflight = inflight;
+        this._tau = tauNs;
+        this._rng = new Prng(seed);
+        this._ewma = new Float64Array(capacity);
+        this._stamp = new Float64Array(capacity);   // all-zero: last-update timestamp
+        for (let i = 0; i < capacity; i++) this._ewma[i] = 1.0; // cold start -> graceful LeastConn
+    }
+
+    /**
+     * A uniformly random ELIGIBLE index, or PICK_NONE if none. Reuses P2cBalancer's exact
+     * rejection-sampling draw (ADR 0005) verbatim -- same `_rng` / `_eligible` / `_live` fields,
+     * no re-implementation, no owned draw-set. Internal, zero-alloc.
+     * @returns {number}
+     */
+    _draw() {
+        return P2cBalancer.prototype._draw.call(this);
+    }
+
+    /**
+     * The decayed EWMA rtt estimate for endpoint i at time `now` (ns). Pure READ -- exponential
+     * decay applied on read, never written. Cold (unsampled) nodes read ~1.0. Zero-alloc.
+     * @param {number} i
+     * @param {number} now  caller-supplied nanoseconds
+     * @returns {number}
+     */
+    ewmaAt(i, now) {
+        return this._ewma[i] * Math.exp(-(now - this._stamp[i]) / this._tau);
+    }
+
+    /**
+     * Record an rtt SAMPLE for endpoint i at time `now` (the warm feedback path -- NOT the hot
+     * pick path). The Finagle peak rule: decay the stored estimate to `now`, then SNAP UP to the
+     * sample if it is larger (a spike is felt instantly) else ease toward it (it decays back over
+     * ~tau). The balancer is the SOLE writer of `_ewma` / `_stamp`. Zero-alloc on the success path.
+     * @param {number} i  endpoint index
+     * @param {number} sampleNs  observed rtt in nanoseconds (finite, >= 0)
+     * @param {number} now  caller-supplied nanoseconds (finite), consistent with pick(now)
+     */
+    recordRtt(i, sampleNs, now) {
+        if (typeof i !== 'number' || typeof sampleNs !== 'number' || typeof now !== 'number') {
+            throw new TypeError('[lite-pick] recordRtt(i, sampleNs, now) requires numbers');
+        }
+        if (i < 0 || i >= this._cap) throw new RangeError('[lite-pick] index out of range: ' + i);
+        if (!Number.isFinite(sampleNs) || sampleNs < 0) {
+            throw new RangeError('[lite-pick] sampleNs must be a finite number >= 0');
+        }
+        if (!Number.isFinite(now)) throw new RangeError('[lite-pick] now must be a finite number');
+        const w = Math.exp(-(now - this._stamp[i]) / this._tau);
+        const e = this._ewma[i] * w;
+        this._ewma[i] = sampleNs > e ? sampleNs : e + (sampleNs - e) * (1 - w);
+        this._stamp[i] = now;
+    }
+
+    /**
+     * Pick by latency-aware power-of-two-choices: two distinct eligible draws, lower cost =
+     * (inflight+1) x ewmaAt(now) wins; a tie goes to the first draw. PICK_NONE (fail closed) iff
+     * the whole pool is down. O(d)=O(1), 0 B/op (pure read -- no write, no clock call).
+     * @param {number} now  caller-supplied nanoseconds (consistent with recordRtt)
+     * @returns {number}
+     */
+    pick(now) {
+        const a = this._draw();
+        if (a < 0) return PICK_NONE;          // whole pool down: fail closed
+        if (this._live === 1) return a;       // only one eligible: it is both choices
+        // A DISTINCT second draw, bounded (the ADR 0005 rationale): at live>=2 each redraw misses
+        // with probability <= 1/2, so 32 tries leaves a ~2^-32 collision chance, expected-O(1), 0 B/op.
+        let b = this._draw();
+        for (let t = 0; b === a && t < 32; t++) b = this._draw();
+        if (b < 0 || b === a) return a;       // astronomically rare: fall back to the first draw
+        const inf = this._inflight, ewma = this._ewma, stamp = this._stamp, tau = this._tau;
+        const costA = (inf[a] + 1) * (ewma[a] * Math.exp(-(now - stamp[a]) / tau));
+        const costB = (inf[b] + 1) * (ewma[b] * Math.exp(-(now - stamp[b]) / tau));
+        return costB < costA ? b : a;         // lower cost wins; tie -> the first draw
+    }
+}

package/Pool.d.ts

@@ -10,9 +10,11 @@ export const VERSION: string;
 
 /** The minimal balancer shape Pool drives (any lite-pick strategy satisfies it). */
 export interface Balancer {
-    pick(): number;
+    pick(now?: number): number;
     readonly capacity: number;
     readonly live: number;
+    /** Optional latency-feedback sink (PeakEwmaBalancer); fed on settle when a clock is supplied. */
+    recordRtt?(i: number, sampleNs: number, now: number): void;
 }
 
 /** Options for `Pool.run`. */
@@ -21,6 +23,11 @@ export interface RunOptions {
     signal?: AbortSignal;
     /** Max distinct-endpoint attempts (default 1 = no failover). */
     tries?: number;
+    /**
+     * A caller-owned nanosecond clock. When present it drives `pick(now)` and the opt-in
+     * `recordRtt` latency feedback for a latency-aware balancer (PeakEwma); otherwise inert.
+     */
+    clock?: () => number;
 }
 
 /**

package/Pool.js

@@ -70,11 +70,20 @@ export class Pool {
      * DIFFERENT endpoint -- up to `tries` attempts, then throw the last error. All counts this run
      * raised are released before returning or throwing (net-zero per run).
      *
+     * A LATENCY-AWARE balancer (PeakEwmaBalancer -- anything duck-typing `recordRtt`) is fed on
+     * settle WHEN a `clock` is supplied: Pool reads `clock()` (caller-owned nanoseconds) before the
+     * attempt, passes it to `pick(now)`, and on a SUCCESSFUL settle records `recordRtt(i, elapsed,
+     * now2)`. Without a `clock`, or against a balancer with no `recordRtt`, the hook is INERT --
+     * Pool stays generic, the in-flight counter stays net-zero, and abort/failover are unaffected.
+     * The kernel `pick()` remains 0 B/op; this wrapper is not held to that bar.
+     *
      * @template T
      * @param {(endpoint: number, signal?: AbortSignal) => (Promise<T>|T)} fn  the per-endpoint work.
-     * @param {{ signal?: AbortSignal, tries?: number }} [opts]  `tries` (default 1 = no failover)
-     *   is the max number of distinct-endpoint attempts; `signal` is passed to `fn` and, when
-     *   already aborted after a failure, stops failover (the abort propagates, no re-pick).
+     * @param {{ signal?: AbortSignal, tries?: number, clock?: () => number }} [opts]  `tries`
+     *   (default 1 = no failover) is the max number of distinct-endpoint attempts; `signal` is
+     *   passed to `fn` and, when already aborted after a failure, stops failover (the abort
+     *   propagates, no re-pick); `clock` is a caller-owned nanosecond source that, when present,
+     *   drives `pick(now)` and the opt-in `recordRtt` latency feedback for a latency-aware balancer.
      * @returns {Promise<T>}
      */
     async run(fn, opts) {
@@ -82,12 +91,17 @@ export class Pool {
         const rawTries = opts && opts.tries != null ? (opts.tries | 0) : 1;
         const tries = rawTries > 0 ? rawTries : 1;
         const signal = opts ? opts.signal : undefined;
+        const clock = opts && typeof opts.clock === 'function' ? opts.clock : undefined;
         const inflight = this._inflight, b = this._b;
+        // Opt-in latency feedback: only when BOTH a clock is supplied AND the balancer duck-types
+        // recordRtt. Otherwise inert -- Pool stays generic and byte-for-byte behaviour is unchanged.
+        const rtt = clock !== undefined && typeof b.recordRtt === 'function';
         const held = [];                 // endpoints incremented this run (kept elevated across failover)
         let lastErr;
         try {
             for (let attempt = 0; attempt < tries; attempt++) {
-                const i = b.pick();
+                const now = clock !== undefined ? clock() : undefined;
+                const i = b.pick(now);
                 if (i === PICK_NONE) {
                     if (attempt === 0) {
                         const e = new Error('[lite-pick] no eligible endpoint');
@@ -99,7 +113,12 @@ export class Pool {
                 inflight[i] = (inflight[i] + 1) >>> 0;
                 held.push(i);
                 try {
-                    return await fn(i, signal);
+                    const out = await fn(i, signal);
+                    if (rtt) {           // successful settle: feed the measured rtt back to the balancer
+                        const done = clock();
+                        b.recordRtt(i, done > now ? done - now : 0, done);
+                    }
+                    return out;
                 } catch (err) {
                     lastErr = err;
                     if (signal && signal.aborted) throw err;   // abort: stop failover, propagate

package/README.md

@@ -21,7 +21,7 @@ The npm landscape has old algorithm libraries (`load-balancers`, `loadbalance`,
 - **Two pieces of evidence, both shipped.** A **0 B/op** witness on the pick path (no object, closure, string, or array created per pick), and a measured **balance-quality anchor** -- peak-to-average load within the strategy's theoretical ceiling (for P2C, the Azar-Broder-Karlin-Upfal `ln ln n / ln 2` bound) and strictly better than a random foil.
 - **A pure selector, not a proxy.** It **consumes** health and circuit state; it never owns them. Health is a shared read-only bitmap written by [`@zakkster/lite-di-health`](https://www.npmjs.com/package/@zakkster/lite-di-health); circuit state comes from [`@zakkster/lite-statechart`](https://www.npmjs.com/package/@zakkster/lite-statechart); load counters are caller-owned typed arrays. `pick()` only reads.
 
-> **Status: M5 (v0.5.0).** Ships the substrate seams **plus `RoundRobinBalancer`, `SmoothWRRBalancer`, `P2cBalancer`, and the exact `LeastConnBalancer` / `SedBalancer` / `NqBalancer` family**, and the **`@zakkster/lite-pick/pool`** request layer. Every strategy is gated: `pick()` proven **0 B/op** (torture + PerfGate), RoundRobin **perfectly fair** with **zero dead picks** vs the naive `i++ % n` foil, SmoothWRR **exactly weighted** and **smooth**, **P2C proves the `ln ln n` balance ceiling** (peak-to-mean gap ~2 vs a random foil's ~21 at n=1024), **LeastConn is greedy-perfect** (max-minus-min load <= 1), and **SED tracks weight within 1%** -- all held under a **seeded invariant fuzzer** (`test/fuzz.mjs`) that checks state-synchronisation after *every* op. See [ROADMAP.md](./ROADMAP.md) for the M5 -> M10 path to 1.0.0, and [decisions/](./decisions) for the ownership boundary (ADR 0001), anti-flapping (ADR 0002), and the RoundRobin (0003), SmoothWRR (0004), P2C (0005), LeastConn-family (0006), and pool-adapter (0007) design forks.
+> **Status: M6 (v0.6.0).** Ships the substrate seams **plus `RoundRobinBalancer`, `SmoothWRRBalancer`, `P2cBalancer`, and the exact `LeastConnBalancer` / `SedBalancer` / `NqBalancer` family**, the **`@zakkster/lite-pick/pool`** request layer, and now the **benchmark suite** -- the balance anchor + GC blast-radius headlines, a seeded/version-stamped `results.json`, a `bench:verify` drift check with teeth, and the vs-AWS positioning (see *Evidence* below). The kernel `Pick.js` / `Pool.js` are byte-identical to v0.5.0 apart from the version stamp. Every strategy is gated: `pick()` proven **0 B/op** (torture + PerfGate), RoundRobin **perfectly fair** with **zero dead picks** vs the naive `i++ % n` foil, SmoothWRR **exactly weighted** and **smooth**, **P2C proves the `ln ln n` balance ceiling** (peak-to-mean gap ~2 vs a random foil's ~21 at n=1024), **LeastConn is greedy-perfect** (max-minus-min load <= 1), and **SED tracks weight within 1%** -- all held under a **seeded invariant fuzzer** (`test/fuzz.mjs`) that checks state-synchronisation after *every* op. See [ROADMAP.md](./ROADMAP.md) for the M6 -> M10 path to 1.0.0, and [decisions/](./decisions) for the ownership boundary (ADR 0001), anti-flapping (ADR 0002), the RoundRobin (0003), SmoothWRR (0004), P2C (0005), LeastConn-family (0006), pool-adapter (0007), and benchmark-suite (0008) design forks.
 
 ```bash
 npm install @zakkster/lite-pick
@@ -139,6 +139,93 @@ The proof (from `test/balance.mjs`):
 
 Each `pick()` is **0 B/op** and **O(cap)** (NQ is O(1) when an early node is idle). Because these are the state-heaviest strategies so far, M4 also introduces the **invariant fuzzer** (`npm run fuzz`): a seeded state-machine attack that, after *every* `pick` / `setEligible` / weight / load op, asserts the chosen endpoint is the *exact* optimum, `live` stays exact, and `PICK_NONE` holds *iff* nothing is pickable -- printing the seed on any failure for byte-for-byte replay.
 
+## Evidence -- the two headlines (v0.6.0 benchmark suite)
+
+> **Framing: parity on speed, superiority on the contract + balance + tail.** A trivial `i++ % n` round-robin -- or `wrr` -- *matches* P2C on raw ops/sec, so `lite-pick` does **not** claim "N times faster." Throughput is claimed at **parity**; the wins are **zero-GC**, **balance quality**, **tail latency** (GC blast-radius), and **never a dead pick**. Every number below is **seeded** and regenerated by `npm run bench:report`; `npm run bench:verify` fails CI if a README number drifts from a fresh run (algorithmic exact, timing within +/-15%). Node / CPU / OS / every PRNG seed are stamped into `benchmark/results.json`.
+
+### Throughput parity vs the incumbents (ops/ms, same pool)
+
+The real pinned npm incumbents (`load-balancers`, `loadbalance`, `wrr`) run through the **same** harness on the **same** `n=1024` pool as the matching `lite-pick` strategy of the **same complexity class** -- ops/ms side by side, not a winner. On the same-work P2C row `lite-pick` holds parity (59591 vs 60956). On the RoundRobin row `lite-pick` is ~22% slower (260168 vs 334541), and that gap is owned, not hidden: `loadbalance@1.0.0` is a bare `i++ % n` with no liveness, while `lite-pick`'s `RoundRobinBalancer` forward-scans the eligibility bitmap to skip down nodes -- so it never returns a dead pick. That scan is the constant-factor cost of a guarantee none of these incumbents offer. `lite-pick` claims parity only where the work is equal; where it is slower, it is slower for the liveness contract, and the balance + tail wins above are the reason to pay it. The weighted-random row is a disclosed **SKIP**: `lite-pick`'s O(1) weighted-random (`WeightedRandom`, alias table) lands at **M10**, so it is not raced against here -- our shipped `SmoothWRRBalancer` is O(cap) *smooth* weighted round-robin (a different, stronger-smoothness guarantee), whose throughput is measured by `npm run witness` and `npm run bench`, not force-fit into this parity table.
+
+<!-- bench:competitors -->
+
+| family | lite-pick | lite-pick ops/ms | incumbent (npm) | incumbent ops/ms |
+| --- | --- | --- | --- | --- |
+| P2C (power-of-two-choices) | P2cBalancer | 59591 | load-balancers@1.3.52 | 60956 |
+| RoundRobin | RoundRobinBalancer | 260168 | loadbalance@1.0.0 | 334541 |
+| Weighted-random | WeightedRandom -- SKIP, ships M10 | -- | wrr@1.0.0 | 158395 |
+
+<!-- /bench:competitors -->
+
+### Balance quality -- the anchor (headline #1)
+
+The canonical balls-into-bins experiment (throw `m = 32*n` balls into `n` bins): P2C's peak load stays within an additive `ln ln n / ln 2` of the mean, while a single random draw's gap grows with the pool. **Peak-gap** = heaviest bin minus the mean (32); the ceiling column is `4*lnln(n)/ln2 + 4`.
+
+<!-- bench:balance -->
+
+| pool n | P2C peak-gap | random foil peak-gap | ceiling |
+| --- | --- | --- | --- |
+| 64 | 3 | 13 | 12.2 |
+| 1024 | 2 | 21 | 15.2 |
+| 4096 | 2 | 25 | 16.2 |
+
+<!-- /bench:balance -->
+
+P2C's peak-gap stays a small `ln ln n` constant while the random foil's grows with `n` -- the exponential improvement one extra probe buys, measured not asserted ([ADR 0008](./decisions/0008-benchmark-suite.md)).
+
+### GC blast-radius -- the tail (headline #2)
+
+The point of zero-GC is **not** the pick's own latency -- a major GC pause freezes **every in-flight request at once**, so an allocating balancer inflates the *service-level* tail. The **same** sustained mixed workload (`n=1024`, 2M requests) runs through two lanes: the zero-GC `lite-pick` lane vs the ordinary "collect candidates, sort, take the best" idiom that allocates a request context per pick (promoted to old gen while in flight -- exactly the garbage that forces mark-sweep).
+
+<!-- bench:gc -->
+
+| lane | major GC | pick B/op | max GC pause (ms) |
+| --- | --- | --- | --- |
+| lite-pick | 0 | 0 | 0.1 |
+| allocating foil | 13 | allocates | 2.9 |
+
+<!-- /bench:gc -->
+
+The `lite-pick` lane holds **0 major GC / 0 B/op** on the pick path; the allocating foil's promoted request state forces mark-sweep pauses that stall the whole service. **That contrast is the headline**, not ops/sec.
+
+### Consistent-hash disruption -- a trust gate
+
+On a scale event (add / remove a node), what fraction of keys keep their node? The **naive-modulo** trap (`key % n`) reshuffles almost everything -- blowing every downstream cache -- while a good consistent hash moves only ~`1/n`. The real `ConsistentHash` (Maglev) lands at **M8**; it is disclosed here as an explicit **SKIP**, not a stub:
+
+<!-- bench:disruption -->
+
+| scale event | naive-modulo remap | good consistent hash |
+| --- | --- | --- |
+| node removed | 98.4% | 1.6% |
+| node added | 98.5% | 1.5% |
+| ConsistentHash (Maglev) | SKIP -- ships in a later milestone | -- |
+
+<!-- /bench:disruption -->
+
+### Honest cold-path cost (so the 0 B/op HOT-path claim stays bounded)
+
+`pick()` is **0 B/op** and that is the whole point -- so the cold-path costs are **named**, not hidden:
+
+| operation | when | allocates |
+| --- | --- | --- |
+| `new <Strategy>Balancer(...)` | construction, once | the balancer object + its owned accumulators (SmoothWRR's Float64 `current`). The eligibility / inflight / weight views are **caller-owned**, never copied |
+| `setEligible(i, up)` | cold, on a health flip | **0** -- one byte write + an O(1) live-count adjust |
+| `setWeight(i, w)` (SmoothWRR) | cold, on reweight | **0** -- one array write + an O(1) eligible-total adjust |
+| `pick()` | **HOT**, per request | **0 B/op** -- proven by `torture` + `test:perf`, measured by `bench:gc` |
+| `Pool.run(fn)` (`/pool`) | per request | a promise + one small `held` array -- an async wrapper, **not** the kernel path ([ADR 0007](./decisions/0007-pool-adapter.md)) |
+
+### Complementary to AWS NLB / ALB (not a competitor)
+
+`lite-pick` does not replace a managed **ALB/NLB** -- it governs the *inner* hop those never see (your service fanning out to downstreams / shards / replicas / workers), bringing the **same algorithm family AWS bills for at the edge** to a zero-GC in-process selector you own:
+
+| AWS edge feature (managed, billed) | `lite-pick` equivalent (in-process, zero-GC) |
+| --- | --- |
+| ALB `least_outstanding_requests` (LOR) | `LeastConnBalancer` / `P2cBalancer` |
+| ALB `weighted_random` + anomaly mitigation | `WeightedRandom` + `BoundedLoad` (M9/M10) |
+| NLB flow-hash (5-tuple) | `ConsistentHash` (Maglev, M8) |
+
+The composition: inbound traffic still enters through your **ALB/NLB -> service** (the edge hop AWS owns and bills); `lite-pick` governs the fan-out **after** that, the hop no AWS load balancer touches. Complementary, not a replacement -- "the hop your ALB/NLB never sees."
+
 ## The substrate (under every strategy)
 
 ```js
@@ -164,7 +251,7 @@ rng.nextBelow(4);     // -> a uint32 in [0, 4)
 rng.reset();          // replays the exact stream
 
 PICK_NONE;            // -> -1  (fail-closed sentinel: no endpoint, never a dead pick)
-VERSION;              // -> '0.4.0'
+VERSION;              // -> '0.6.0'
 ```
 
 `BalancerBase.pick()` is **abstract** -- it throws, so an unfinished strategy fails loudly rather than returning a dead index. Every shipped strategy (`RoundRobinBalancer`, `SmoothWRRBalancer`, `P2cBalancer`, `LeastConnBalancer`, `SedBalancer`, `NqBalancer`) extends it and reads the same shared eligibility view; you subclass it the same way to add your own.

package/llms.txt

@@ -1,6 +1,6 @@
 # @zakkster/lite-pick
 
-Version: 0.5.0
+Version: 0.6.0
 License: MIT (c) Zahary Shinikchiev <shinikchiev@yahoo.com>
 Runtime dependencies: none. ESM only. ASCII-only source. sideEffects: false.
 Node: >= 18.
@@ -14,7 +14,7 @@ reading pre-allocated views that siblings or the caller write, and returning an
 The complementary evidence lite-pick ships is a measured balance-quality anchor (peak-to-
 average load vs the strategy's theoretical ceiling) alongside the 0 B/op pick witness.
 
-0.4.0 ships the substrate seams + six strategies: RoundRobin, SmoothWRR (the weighted
+0.6.0 ships the substrate seams + six strategies: RoundRobin, SmoothWRR (the weighted
 default), P2C (power-of-two-choices -- also the O(1) least-connections APPROXIMATION), and the
 EXACT LeastConn family (LeastConn, SED, NQ). It exports `VERSION`, the fail-closed sentinel
 `PICK_NONE` (-1), a deterministic `Prng` (xorshift32), `BalancerBase` (the shared read-only
@@ -24,6 +24,16 @@ session (see ROADMAP.md): PeakEWMA, ConsistentHash, BoundedLoad, WeightedRandom.
 n) fewest-in-flight variant is a deferred @zakkster/lite-logn `BinaryHeap` optional-peer seam
 (decisions/0006), not this exact-O(cap) scan.
 
+M6 (0.6.0) adds the BENCHMARK SUITE -- an evidence session, no API change: Pick.js / Pool.js are
+byte-identical to 0.5.0 apart from the VERSION stamp. It lives entirely under benchmark/ (NOT in
+the published tarball): the balance anchor + GC blast-radius headlines, weighted-fairness and
+consistent-hash-disruption trust gates, a seeded + Node/CPU/OS-stamped results.json, a
+bench:verify drift check (algorithmic EXACT, timing +/-15%), the naive-modulo disruption foil
+with an explicit ConsistentHash SKIP row (Maglev lands at M8), the endurance Soak.mjs scaffold
+(reuses test/invariants.mjs:checkBase), and REAL pinned npm competitors (load-balancers,
+loadbalance, wrr) as devDeps beside our in-repo foils. Framing: parity on speed, superiority on
+the contract + balance + tail -- never an "N times faster" headline (decisions/0008).
+
 ## Design ownership (decisions/0001, 0002)
 
 - IN-PROCESS first (workers, DI services); remote HTTP is a thin optional adapter.

package/package.json

@@ -1,8 +1,8 @@
 {
   "name": "@zakkster/lite-pick",
   "author": "Zahary Shinikchiev <shinikchiev@yahoo.com>",
-  "version": "0.5.0",
-  "description": "Zero-dependency, zero-GC load-balancing selection kernel: one hot pick() -> endpoint index over a fixed pool, 0 B/op steady-state. A pure selector (consumes health/circuit state, never a proxy) for the in-process hop, complementary to AWS NLB/ALB. Tree-shakeable ESM roster: RoundRobin, SmoothWRR, P2C, LeastConn, SED, NQ, plus PeakEWMA and consistent hashing; the /pool subpath adds dispatch/settle counters + failover and a duck-typed query-cache fetcher.",
+  "version": "0.7.0",
+  "description": "Zero-dependency, zero-GC load-balancing selection kernel: one hot pick() -> endpoint index over a fixed pool, 0 B/op steady-state. A pure selector (consumes health/circuit state, never a proxy) for the in-process hop, complementary to AWS NLB/ALB. Tree-shakeable ESM roster: RoundRobin, SmoothWRR, P2C, LeastConn, SED, NQ, PeakEWMA (latency-aware peak-EWMA), plus consistent hashing; the /pool subpath adds dispatch/settle counters + failover and a duck-typed query-cache fetcher.",
   "type": "module",
   "main": "./Pick.js",
   "module": "./Pick.js",
@@ -40,7 +40,12 @@
     "fuzz": "node test/fuzz.mjs",
     "test:perf": "node --expose-gc --max-semi-space-size=4 --test test/perf/PerfGate.test.mjs",
     "bench": "node benchmark/Matrix.mjs",
-    "bench:report": "node benchmark/Matrix.mjs && node benchmark/Report.mjs",
+    "bench:gc": "node --expose-gc benchmark/GcBlastRadius.mjs",
+    "bench:fairness": "node benchmark/Fairness.mjs",
+    "bench:disruption": "node benchmark/Disruption.mjs",
+    "bench:report": "node benchmark/Matrix.mjs && node --expose-gc benchmark/Report.mjs",
+    "bench:verify": "node benchmark/Report.mjs --verify",
+    "soak": "node --expose-gc benchmark/Soak.mjs",
     "demo": "node demo/fanout.mjs",
     "verify": "npm test && npm run test:types && npm run torture && npm run witness && npm run balance && npm run fuzz && npm run test:perf"
   },
@@ -103,7 +108,10 @@
     "@zakkster/lite-gc-profiler": "^1.16.0",
     "@zakkster/lite-leak": "^1.10.0",
     "@zakkster/lite-perf-gate": "^1.4.2",
-    "typescript": "^7.0.2"
+    "load-balancers": "1.3.52",
+    "loadbalance": "1.0.0",
+    "typescript": "^7.0.2",
+    "wrr": "1.0.0"
   },
   "peerDependencies": {},
   "peerDependenciesMeta": {},