@miller-tech/uap 1.0.0 → 1.3.0

package/dist/utils/concurrency-pool.d.ts

@@ -0,0 +1,51 @@
+/**
+ * Concurrency Pool
+ *
+ * Shared utility for bounded-concurrency parallel execution.
+ * Replaces duplicated Promise.all batching patterns across the codebase.
+ *
+ * Uses getMaxParallel() for auto-detection with UAP_MAX_PARALLEL env override.
+ */
+/**
+ * Map over items with bounded concurrency.
+ *
+ * Unlike Promise.all(items.map(fn)), this limits the number of
+ * in-flight promises to prevent overwhelming local inference
+ * endpoints or exhausting file descriptors.
+ *
+ * @param items - Array of items to process
+ * @param fn - Async function to apply to each item
+ * @param options - Concurrency options
+ * @returns Results in the same order as input items
+ *
+ * @example
+ * ```ts
+ * // Auto-detect concurrency from vCPUs
+ * const results = await concurrentMap(urls, url => fetch(url));
+ *
+ * // Explicit limit
+ * const results = await concurrentMap(tasks, runTask, { maxConcurrent: 4 });
+ *
+ * // CPU-bound mode (reserves cores for OS)
+ * const results = await concurrentMap(files, compress, { mode: 'cpu' });
+ * ```
+ */
+export declare function concurrentMap<T, R>(items: T[], fn: (item: T, index: number) => Promise<R>, options?: {
+    /** Maximum concurrent operations. Overrides auto-detection. */
+    maxConcurrent?: number;
+    /** 'cpu' reserves cores for OS/inference, 'io' allows higher concurrency */
+    mode?: 'cpu' | 'io';
+}): Promise<R[]>;
+/**
+ * Map over items with bounded concurrency, settling all promises.
+ *
+ * Like concurrentMap but uses Promise.allSettled semantics --
+ * failures don't abort other in-flight operations.
+ *
+ * @returns Array of PromiseSettledResult in input order
+ */
+export declare function concurrentMapSettled<T, R>(items: T[], fn: (item: T, index: number) => Promise<R>, options?: {
+    maxConcurrent?: number;
+    mode?: 'cpu' | 'io';
+}): Promise<PromiseSettledResult<R>[]>;
+//# sourceMappingURL=concurrency-pool.d.ts.map

package/dist/utils/concurrency-pool.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"concurrency-pool.d.ts","sourceRoot":"","sources":["../../src/utils/concurrency-pool.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAIH;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AACH,wBAAsB,aAAa,CAAC,CAAC,EAAE,CAAC,EACtC,KAAK,EAAE,CAAC,EAAE,EACV,EAAE,EAAE,CAAC,IAAI,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,KAAK,OAAO,CAAC,CAAC,CAAC,EAC1C,OAAO,CAAC,EAAE;IACR,+DAA+D;IAC/D,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,4EAA4E;IAC5E,IAAI,CAAC,EAAE,KAAK,GAAG,IAAI,CAAC;CACrB,GACA,OAAO,CAAC,CAAC,EAAE,CAAC,CAkBd;AAED;;;;;;;GAOG;AACH,wBAAsB,oBAAoB,CAAC,CAAC,EAAE,CAAC,EAC7C,KAAK,EAAE,CAAC,EAAE,EACV,EAAE,EAAE,CAAC,IAAI,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,KAAK,OAAO,CAAC,CAAC,CAAC,EAC1C,OAAO,CAAC,EAAE;IACR,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,IAAI,CAAC,EAAE,KAAK,GAAG,IAAI,CAAC;CACrB,GACA,OAAO,CAAC,oBAAoB,CAAC,CAAC,CAAC,EAAE,CAAC,CAuBpC"}

package/dist/utils/concurrency-pool.js

@@ -0,0 +1,80 @@
+/**
+ * Concurrency Pool
+ *
+ * Shared utility for bounded-concurrency parallel execution.
+ * Replaces duplicated Promise.all batching patterns across the codebase.
+ *
+ * Uses getMaxParallel() for auto-detection with UAP_MAX_PARALLEL env override.
+ */
+import { getMaxParallel } from './system-resources.js';
+/**
+ * Map over items with bounded concurrency.
+ *
+ * Unlike Promise.all(items.map(fn)), this limits the number of
+ * in-flight promises to prevent overwhelming local inference
+ * endpoints or exhausting file descriptors.
+ *
+ * @param items - Array of items to process
+ * @param fn - Async function to apply to each item
+ * @param options - Concurrency options
+ * @returns Results in the same order as input items
+ *
+ * @example
+ * ```ts
+ * // Auto-detect concurrency from vCPUs
+ * const results = await concurrentMap(urls, url => fetch(url));
+ *
+ * // Explicit limit
+ * const results = await concurrentMap(tasks, runTask, { maxConcurrent: 4 });
+ *
+ * // CPU-bound mode (reserves cores for OS)
+ * const results = await concurrentMap(files, compress, { mode: 'cpu' });
+ * ```
+ */
+export async function concurrentMap(items, fn, options) {
+    if (items.length === 0)
+        return [];
+    const max = options?.maxConcurrent ?? getMaxParallel(options?.mode ?? 'io');
+    const results = new Array(items.length);
+    let nextIndex = 0;
+    const worker = async () => {
+        while (nextIndex < items.length) {
+            const i = nextIndex++;
+            results[i] = await fn(items[i], i);
+        }
+    };
+    const workerCount = Math.min(max, items.length);
+    await Promise.all(Array.from({ length: workerCount }, () => worker()));
+    return results;
+}
+/**
+ * Map over items with bounded concurrency, settling all promises.
+ *
+ * Like concurrentMap but uses Promise.allSettled semantics --
+ * failures don't abort other in-flight operations.
+ *
+ * @returns Array of PromiseSettledResult in input order
+ */
+export async function concurrentMapSettled(items, fn, options) {
+    if (items.length === 0)
+        return [];
+    const max = options?.maxConcurrent ?? getMaxParallel(options?.mode ?? 'io');
+    const results = new Array(items.length);
+    let nextIndex = 0;
+    const worker = async () => {
+        while (nextIndex < items.length) {
+            const i = nextIndex++;
+            try {
+                const value = await fn(items[i], i);
+                results[i] = { status: 'fulfilled', value };
+            }
+            catch (reason) {
+                results[i] = { status: 'rejected', reason };
+            }
+        }
+    };
+    const workerCount = Math.min(max, items.length);
+    await Promise.all(Array.from({ length: workerCount }, () => worker()));
+    return results;
+}
+//# sourceMappingURL=concurrency-pool.js.map

package/dist/utils/concurrency-pool.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"concurrency-pool.js","sourceRoot":"","sources":["../../src/utils/concurrency-pool.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,EAAE,cAAc,EAAE,MAAM,uBAAuB,CAAC;AAEvD;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AACH,MAAM,CAAC,KAAK,UAAU,aAAa,CACjC,KAAU,EACV,EAA0C,EAC1C,OAKC;IAED,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IAElC,MAAM,GAAG,GAAG,OAAO,EAAE,aAAa,IAAI,cAAc,CAAC,OAAO,EAAE,IAAI,IAAI,IAAI,CAAC,CAAC;IAC5E,MAAM,OAAO,GAAQ,IAAI,KAAK,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;IAC7C,IAAI,SAAS,GAAG,CAAC,CAAC;IAElB,MAAM,MAAM,GAAG,KAAK,IAAmB,EAAE;QACvC,OAAO,SAAS,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC;YAChC,MAAM,CAAC,GAAG,SAAS,EAAE,CAAC;YACtB,OAAO,CAAC,CAAC,CAAC,GAAG,MAAM,EAAE,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;QACrC,CAAC;IACH,CAAC,CAAC;IAEF,MAAM,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;IAChD,MAAM,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,WAAW,EAAE,EAAE,GAAG,EAAE,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IAEvE,OAAO,OAAO,CAAC;AACjB,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,CAAC,KAAK,UAAU,oBAAoB,CACxC,KAAU,EACV,EAA0C,EAC1C,OAGC;IAED,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IAElC,MAAM,GAAG,GAAG,OAAO,EAAE,aAAa,IAAI,cAAc,CAAC,OAAO,EAAE,IAAI,IAAI,IAAI,CAAC,CAAC;IAC5E,MAAM,OAAO,GAA8B,IAAI,KAAK,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;IACnE,IAAI,SAAS,GAAG,CAAC,CAAC;IAElB,MAAM,MAAM,GAAG,KAAK,IAAmB,EAAE;QACvC,OAAO,SAAS,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC;YAChC,MAAM,CAAC,GAAG,SAAS,EAAE,CAAC;YACtB,IAAI,CAAC;gBACH,MAAM,KAAK,GAAG,MAAM,EAAE,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;gBACpC,OAAO,CAAC,CAAC,CAAC,GAAG,EAAE,MAAM,EAAE,WAAW,EAAE,KAAK,EAAE,CAAC;YAC9C,CAAC;YAAC,OAAO,MAAM,EAAE,CAAC;gBAChB,OAAO,CAAC,CAAC,CAAC,GAAG,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,EAAE,CAAC;YAC9C,CAAC;QACH,CAAC;IACH,CAAC,CAAC;IAEF,MAAM,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;IAChD,MAAM,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,WAAW,EAAE,EAAE,GAAG,EAAE,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IAEvE,OAAO,OAAO,CAAC;AACjB,CAAC"}

package/dist/utils/system-resources.d.ts

@@ -0,0 +1,47 @@
+/**
+ * System Resource Detection
+ *
+ * Provides vCPU, VRAM, and memory detection with caching.
+ * Used to auto-tune parallelism across the UAP system.
+ *
+ * Env override: UAP_MAX_PARALLEL always takes precedence.
+ * Precedence: env var → config → auto-detect → hardcoded default
+ */
+export interface SystemResources {
+    /** Number of logical CPUs (vCPUs / hardware threads) */
+    vCPUs: number;
+    /** GPU VRAM in GB (0 if no GPU detected) */
+    vramGB: number;
+    /** System RAM in GB */
+    memoryGB: number;
+}
+/**
+ * Detect system resources (cached after first call).
+ */
+export declare function detectSystemResources(): SystemResources;
+/**
+ * Compute safe parallelism ceiling.
+ *
+ * @param mode - 'cpu' for compute-bound work (reserves cores for OS + inference),
+ *               'io' for IO-bound work like API calls (higher concurrency safe)
+ * @returns Maximum number of concurrent operations
+ *
+ * Precedence:
+ *   1. UAP_MAX_PARALLEL env var (always wins)
+ *   2. Auto-detected from os.cpus()
+ *   3. Hardcoded fallback (3)
+ */
+export declare function getMaxParallel(mode?: 'cpu' | 'io'): number;
+/**
+ * Check if parallelism is globally enabled.
+ *
+ * Precedence:
+ *   1. UAP_PARALLEL env var ('false' disables)
+ *   2. Default: true
+ */
+export declare function isParallelEnabled(): boolean;
+/**
+ * Reset cached resources (for testing).
+ */
+export declare function resetResourceCache(): void;
+//# sourceMappingURL=system-resources.d.ts.map

package/dist/utils/system-resources.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"system-resources.d.ts","sourceRoot":"","sources":["../../src/utils/system-resources.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAKH,MAAM,WAAW,eAAe;IAC9B,wDAAwD;IACxD,KAAK,EAAE,MAAM,CAAC;IACd,4CAA4C;IAC5C,MAAM,EAAE,MAAM,CAAC;IACf,uBAAuB;IACvB,QAAQ,EAAE,MAAM,CAAC;CAClB;AAID;;GAEG;AACH,wBAAgB,qBAAqB,IAAI,eAAe,CA+BvD;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,cAAc,CAAC,IAAI,GAAE,KAAK,GAAG,IAAW,GAAG,MAAM,CAiBhE;AAED;;;;;;GAMG;AACH,wBAAgB,iBAAiB,IAAI,OAAO,CAE3C;AAED;;GAEG;AACH,wBAAgB,kBAAkB,IAAI,IAAI,CAEzC"}

package/dist/utils/system-resources.js

@@ -0,0 +1,92 @@
+/**
+ * System Resource Detection
+ *
+ * Provides vCPU, VRAM, and memory detection with caching.
+ * Used to auto-tune parallelism across the UAP system.
+ *
+ * Env override: UAP_MAX_PARALLEL always takes precedence.
+ * Precedence: env var → config → auto-detect → hardcoded default
+ */
+import { cpus, totalmem } from 'os';
+import { execSync } from 'child_process';
+let _cached = null;
+/**
+ * Detect system resources (cached after first call).
+ */
+export function detectSystemResources() {
+    if (_cached)
+        return _cached;
+    const vCPUs = cpus().length;
+    const memoryGB = Math.round(totalmem() / 1024 ** 3);
+    let vramGB = 0;
+    try {
+        // NVIDIA GPU
+        const out = execSync('nvidia-smi --query-gpu=memory.total --format=csv,noheader,nounits', {
+            encoding: 'utf-8',
+            timeout: 3000,
+            stdio: ['pipe', 'pipe', 'pipe'],
+        });
+        vramGB = Math.round(parseInt(out.trim().split('\n')[0], 10) / 1024);
+    }
+    catch {
+        try {
+            // macOS unified memory (report as VRAM since GPU shares it)
+            const out = execSync('sysctl -n hw.memsize', {
+                encoding: 'utf-8',
+                timeout: 3000,
+                stdio: ['pipe', 'pipe', 'pipe'],
+            });
+            vramGB = Math.min(Math.round(parseInt(out.trim(), 10) / 1024 ** 3), 48);
+        }
+        catch {
+            // No GPU detected
+        }
+    }
+    _cached = { vCPUs, vramGB, memoryGB };
+    return _cached;
+}
+/**
+ * Compute safe parallelism ceiling.
+ *
+ * @param mode - 'cpu' for compute-bound work (reserves cores for OS + inference),
+ *               'io' for IO-bound work like API calls (higher concurrency safe)
+ * @returns Maximum number of concurrent operations
+ *
+ * Precedence:
+ *   1. UAP_MAX_PARALLEL env var (always wins)
+ *   2. Auto-detected from os.cpus()
+ *   3. Hardcoded fallback (3)
+ */
+export function getMaxParallel(mode = 'io') {
+    const envOverride = process.env.UAP_MAX_PARALLEL;
+    if (envOverride) {
+        const parsed = parseInt(envOverride, 10);
+        if (!isNaN(parsed) && parsed > 0)
+            return parsed;
+    }
+    const { vCPUs } = detectSystemResources();
+    if (mode === 'cpu') {
+        // Reserve 2 cores for OS + inference server
+        return Math.max(1, vCPUs - 2);
+    }
+    // IO-bound: safe to use more concurrency, cap at 8 to avoid
+    // overwhelming local inference endpoints
+    return Math.max(1, Math.min(vCPUs, 8));
+}
+/**
+ * Check if parallelism is globally enabled.
+ *
+ * Precedence:
+ *   1. UAP_PARALLEL env var ('false' disables)
+ *   2. Default: true
+ */
+export function isParallelEnabled() {
+    return process.env.UAP_PARALLEL !== 'false';
+}
+/**
+ * Reset cached resources (for testing).
+ */
+export function resetResourceCache() {
+    _cached = null;
+}
+//# sourceMappingURL=system-resources.js.map

package/dist/utils/system-resources.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"system-resources.js","sourceRoot":"","sources":["../../src/utils/system-resources.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,IAAI,CAAC;AACpC,OAAO,EAAE,QAAQ,EAAE,MAAM,eAAe,CAAC;AAWzC,IAAI,OAAO,GAA2B,IAAI,CAAC;AAE3C;;GAEG;AACH,MAAM,UAAU,qBAAqB;IACnC,IAAI,OAAO;QAAE,OAAO,OAAO,CAAC;IAE5B,MAAM,KAAK,GAAG,IAAI,EAAE,CAAC,MAAM,CAAC;IAC5B,MAAM,QAAQ,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,EAAE,GAAG,IAAI,IAAI,CAAC,CAAC,CAAC;IAEpD,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,IAAI,CAAC;QACH,aAAa;QACb,MAAM,GAAG,GAAG,QAAQ,CAAC,mEAAmE,EAAE;YACxF,QAAQ,EAAE,OAAO;YACjB,OAAO,EAAE,IAAI;YACb,KAAK,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC;SAChC,CAAC,CAAC;QACH,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,IAAI,CAAC,CAAC;IACtE,CAAC;IAAC,MAAM,CAAC;QACP,IAAI,CAAC;YACH,4DAA4D;YAC5D,MAAM,GAAG,GAAG,QAAQ,CAAC,sBAAsB,EAAE;gBAC3C,QAAQ,EAAE,OAAO;gBACjB,OAAO,EAAE,IAAI;gBACb,KAAK,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC;aAChC,CAAC,CAAC;YACH,MAAM,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,IAAI,EAAE,EAAE,EAAE,CAAC,GAAG,IAAI,IAAI,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;QAC1E,CAAC;QAAC,MAAM,CAAC;YACP,kBAAkB;QACpB,CAAC;IACH,CAAC;IAED,OAAO,GAAG,EAAE,KAAK,EAAE,MAAM,EAAE,QAAQ,EAAE,CAAC;IACtC,OAAO,OAAO,CAAC;AACjB,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,cAAc,CAAC,OAAqB,IAAI;IACtD,MAAM,WAAW,GAAG,OAAO,CAAC,GAAG,CAAC,gBAAgB,CAAC;IACjD,IAAI,WAAW,EAAE,CAAC;QAChB,MAAM,MAAM,GAAG,QAAQ,CAAC,WAAW,EAAE,EAAE,CAAC,CAAC;QACzC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,MAAM,GAAG,CAAC;YAAE,OAAO,MAAM,CAAC;IAClD,CAAC;IAED,MAAM,EAAE,KAAK,EAAE,GAAG,qBAAqB,EAAE,CAAC;IAE1C,IAAI,IAAI,KAAK,KAAK,EAAE,CAAC;QACnB,4CAA4C;QAC5C,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC;IAChC,CAAC;IAED,4DAA4D;IAC5D,yCAAyC;IACzC,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,CAAC;AACzC,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,iBAAiB;IAC/B,OAAO,OAAO,CAAC,GAAG,CAAC,YAAY,KAAK,OAAO,CAAC;AAC9C,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,kBAAkB;IAChC,OAAO,GAAG,IAAI,CAAC;AACjB,CAAC"}

package/docs/BENCHMARK_GAPS_AND_PLAN.md

@@ -0,0 +1,146 @@
+# UAP Benchmark: Actual Gaps & Execution Plan
+
+**Generated:** 2026-03-17
+**Benchmark:** Harbor Terminal-Bench 2.0 (89 tasks)
+**Primary Target:** Qwen3.5 35B A3B (IQ4_XS)
+
+---
+
+## What Already Exists (DO NOT REBUILD)
+
+| Component                        | File                                                       | Status                                                                                         |
+| -------------------------------- | ---------------------------------------------------------- | ---------------------------------------------------------------------------------------------- |
+| Baseline benchmark (no UAP)      | `scripts/benchmarks/benchmark-qwen35-baseline-no-uap.tsx`  | 403 lines, 94 tasks                                                                            |
+| UAP benchmark (full integration) | `scripts/benchmarks/benchmark-qwen35-uap-3.0-opencode.tsx` | 812 lines, 89 tasks                                                                            |
+| Harbor quick runner (UAP)        | `scripts/benchmarks/run-tbench-qwen35-quick.sh`            | 459 lines, hybrid-adaptive                                                                     |
+| Harbor baseline+UAP runner       | `scripts/benchmarks/run-harbor-qwen35-benchmark.sh`        | Runs both configs sequentially                                                                 |
+| Harbor YAML configs              | `benchmarks/harbor-configs/qwen35_*.yaml`                  | Baseline + UAP pair                                                                            |
+| Comparison report generator      | `scripts/benchmarks/generate-comparison-report.ts`         | 461 lines, p-value tests                                                                       |
+| Full benchmark harness           | `scripts/benchmarks/run-full-benchmark.sh`                 | 413 lines, multi-model A/B                                                                     |
+| Multi-turn agent loop            | `src/benchmarks/multi-turn-loop.ts`                        | 213 lines, `executeWithRetry()`                                                                |
+| Multi-turn + verification        | `src/benchmarks/multi-turn-agent.ts`                       | Wired to dynamic retrieval                                                                     |
+| Improved benchmark runner        | `src/benchmarks/improved-benchmark.ts`                     | 794 lines, wires multi-turn + dynamic retrieval + task classification + hierarchical prompting |
+| Dynamic memory retrieval         | `src/memory/dynamic-retrieval.ts`                          | 1168 lines, 6 memory sources, adaptive depth                                                   |
+| Task classifier                  | `src/memory/task-classifier.ts`                            | 426 lines, 8 categories, ambiguity detection                                                   |
+| Qdrant embeddings                | `src/memory/embeddings.ts`                                 | Fixed, 5 backends with fallback                                                                |
+| Tool call retry (Qwen)           | `tools/agents/scripts/qwen_tool_call_wrapper.py`           | 686 lines, 6 retry strategies                                                                  |
+| Harbor UAP agent                 | `tools/uap_harbor/uap_agent.py`                            | 379 lines, classified preamble                                                                 |
+| Qwen3.5 model presets            | `src/models/types.ts:136-151`                              | `qwen35-a3b` and `qwen35` defined                                                              |
+| Model router                     | `src/models/router.ts`                                     | Qwen3.5 as default executor                                                                    |
+
+---
+
+## Actual Gaps (3 items)
+
+### Gap 1: `improved-benchmark.ts` MODELS array missing Qwen3.5
+
+`src/benchmarks/improved-benchmark.ts:95-99` has the fully wired runner (multi-turn + dynamic retrieval + task classification + hierarchical prompting + verification) but its MODELS array only contains:
+
+```typescript
+const MODELS: ModelConfig[] = [
+  { id: 'opus-4.5', name: 'Claude Opus 4.5', apiModel: 'claude-opus-4-5-20251101' },
+  { id: 'glm-4.7', name: 'GLM 4.7', apiModel: 'glm-4.7' },
+  { id: 'gpt-5.2-codex', name: 'GPT 5.2 Codex', apiModel: 'gpt-5.2-codex' },
+];
+// Qwen3.5 MISSING
+```
+
+**Fix:** Add Qwen3.5 to the MODELS array. The preset already exists in `src/models/types.ts:136-151`.
+
+### Gap 2: `model-integration.ts` MODELS array missing Qwen3.5 + still single-shot
+
+`src/benchmarks/model-integration.ts:336-361` is the older benchmark runner. It:
+
+- Has no Qwen3.5 in its MODELS array
+- Uses single-shot execution (no multi-turn, no dynamic retrieval)
+
+**Fix:** Add Qwen3.5 to its MODELS array. The multi-turn wiring gap is already solved by `improved-benchmark.ts` -- this file can remain as the "legacy single-shot" runner for comparison purposes.
+
+### Gap 3: No benchmark results exist
+
+`benchmark-results/` directory does not exist. None of the scripts have been executed.
+
+**Fix:** Run the existing scripts.
+
+---
+
+## Execution Plan
+
+### Step 1: Add Qwen3.5 to improved-benchmark.ts MODELS array
+
+**File:** `src/benchmarks/improved-benchmark.ts:95-99`
+
+```typescript
+const MODELS: ModelConfig[] = [
+  { id: 'opus-4.5', name: 'Claude Opus 4.5', apiModel: 'claude-opus-4-5-20251101' },
+  { id: 'glm-4.7', name: 'GLM 4.7', apiModel: 'glm-4.7' },
+  { id: 'gpt-5.2-codex', name: 'GPT 5.2 Codex', apiModel: 'gpt-5.2-codex' },
+  { id: 'qwen35-a3b', name: 'Qwen 3.5 35B A3B', apiModel: 'qwen35-a3b-iq4xs' },
+];
+```
+
+### Step 2: Add Qwen3.5 to model-integration.ts MODELS array
+
+**File:** `src/benchmarks/model-integration.ts:336-361`
+
+```typescript
+{
+  id: 'qwen35-a3b',
+  name: 'Qwen 3.5 35B A3B',
+  provider: 'local',
+  apiModel: 'qwen35-a3b-iq4xs',
+},
+```
+
+### Step 3: Run existing benchmarks
+
+```bash
+# Option A: Quick Qwen3.5 baseline + UAP via Harbor (recommended first)
+./scripts/benchmarks/run-harbor-qwen35-benchmark.sh
+
+# Option B: Direct API baseline (no Harbor containers)
+npx tsx scripts/benchmarks/benchmark-qwen35-baseline-no-uap.tsx
+
+# Option C: Direct API UAP-enhanced
+npx tsx scripts/benchmarks/benchmark-qwen35-uap-3.0-opencode.tsx
+
+# Option D: Improved benchmark with multi-turn + dynamic retrieval (all models)
+npx tsx src/benchmarks/improved-benchmark.ts
+
+# Option E: Full Harbor harness (all models, baseline vs UAP)
+./scripts/benchmarks/run-full-benchmark.sh --model qwen35-a3b-iq4xs
+```
+
+### Step 4: Generate comparison report
+
+```bash
+npx tsx scripts/benchmarks/generate-comparison-report.ts \
+  --baseline benchmark-results/qwen35_baseline_no_uap/ \
+  --uap benchmark-results/qwen35_uap_3.0_opencode/
+```
+
+---
+
+## What This Plan Does NOT Do (because it already exists)
+
+- Build a multi-turn agent loop (exists: `src/benchmarks/multi-turn-loop.ts`)
+- Build dynamic memory retrieval (exists: `src/memory/dynamic-retrieval.ts`)
+- Build task classification (exists: `src/memory/task-classifier.ts`)
+- Fix Qdrant embeddings (already fixed: `src/memory/embeddings.ts`)
+- Build Harbor configs (exist: `benchmarks/harbor-configs/qwen35_*.yaml`)
+- Build comparison report generator (exists: `scripts/benchmarks/generate-comparison-report.ts`)
+- Wire multi-turn into benchmark runner (exists: `src/benchmarks/improved-benchmark.ts`)
+- Build tool call retry for Qwen (exists: `tools/agents/scripts/qwen_tool_call_wrapper.py`)
+- Create execution scripts (exist: 6+ scripts in `scripts/benchmarks/`)
+
+---
+
+## Estimated Effort
+
+| Step                                 | Effort         | Type                                   |
+| ------------------------------------ | -------------- | -------------------------------------- |
+| Add Qwen3.5 to improved-benchmark.ts | 2 minutes      | Code change (1 line)                   |
+| Add Qwen3.5 to model-integration.ts  | 2 minutes      | Code change (5 lines)                  |
+| Run benchmarks                       | 2-8 hours      | Execution (depends on model speed)     |
+| Review results                       | 30 minutes     | Analysis                               |
+| **Total**                            | **~3-9 hours** | Mostly waiting for benchmark execution |