App: release beta.10 with TensorFold 1.1.4

app/README.md

@@ -4,8 +4,8 @@ An agentic AI coding app for the Mac, with the TensorFold engine built in. It al
 any OpenAI‑compatible LLM endpoint, local or cloud. The look comes from the TensorFold brand:
 graphite and navy surfaces, an electric-blue primary and the fold gradient.
 
-The current desktop release is [beta 9](https://github.com/ashhart/TensorFold/releases/tag/v1.0.0-beta.9).
-It launches the bundled native Zig engine directly.
+The current desktop release is [beta 10](https://github.com/ashhart/TensorFold/releases/tag/v1.0.0-beta.10).
+It launches the bundled TensorFold 1.1.4 native Zig engine directly.
 
 It reads, edits, and runs your code through a tool‑use loop, with streaming responses,
 live diffs, and per‑action approval.

app/build/engine.json

@@ -1,10 +1,10 @@
 {
-  "version": "1.1.0",
+  "version": "1.1.4",
   "platform": "macos-arm64",
-  "archive": "tensorfold-1.1.0-macos-arm64.tar.gz",
-  "root": "tensorfold-1.1.0-macos-arm64",
-  "url": "https://github.com/ashhart/TensorFold/releases/download/v1.1.0/tensorfold-1.1.0-macos-arm64.tar.gz",
-  "sha256": "99c66f56fcde07a7dd8df6ea555bcbd406d0537b7d92a453112a54bef5a72707",
+  "archive": "tensorfold-1.1.4-macos-arm64.tar.gz",
+  "root": "tensorfold-1.1.4-macos-arm64",
+  "url": "https://github.com/ashhart/TensorFold/releases/download/v1.1.4/tensorfold-1.1.4-macos-arm64.tar.gz",
+  "sha256": "e60096c8af989f26eb622bde3099ff6ed3ecc37cdd46ade0397b3afd16145c6f",
   "required": [
     "bin/tensorfold-native",
     "lib/libtfgrammar.dylib",

app/docs/beta/1.0.0-beta.10.md

@@ -0,0 +1,22 @@
+# TensorFold app 1.0.0-beta.10
+
+This macOS app beta bundles the released TensorFold 1.1.4 native engine for Apple silicon on macOS 13.3 or later.
+The engine archive is pinned to its published SHA-256 and its internal file checksums are verified before packaging.
+
+## In this beta
+
+- Update the bundled native engine from 1.1.0 to 1.1.4. See the [engine release notes](https://github.com/ashhart/TensorFold/releases/tag/v1.1.4) for its platform-specific changes and measurements.
+- Benchmarks use the engine's community-v2 protocol, recording request routes, cache format and natural sampled stops. Saved community-v1 results remain readable with their original protocol label.
+- Google sign-in accepts Google's issuer callback parameter and shows a useful explanation when sign-in fails.
+- Update checks and downloads work through fake-IP proxies while retaining the destination checks used for Hub downloads.
+- The model library uses the engine's geometry admission checks before offering a checkpoint for loading.
+- Update the shipped HTTP, email and schema-validation dependencies to patched versions.
+
+## Limits and qualification
+
+The beta.9 limitations still apply, including disabled Cortheon web research, experimental Living Weights and incomplete distributed-inference qualification.
+Google and messaging connections require user-owned accounts and permissions; synthetic service checks do not establish every live-account workflow.
+Window-close flashing reported in Full mode remains under investigation.
+No new mobile or watch installer is included.
+
+Download the signed Mac installer from https://tensorfold.dev/download/mac, or use the in-app update prompt after the signed feed detects this version.

app/electron/agent/engine/benchmark/controller.ts

@@ -125,7 +125,7 @@ export class ModelBenchmarks {
       job.receipt = receipt ? parseBenchmarkReceipt(JSON.stringify(receipt), { ...input, engineVersion: checkpoint?.engineVersion ?? BENCHMARK_ENGINE_VERSION }, this.now()) : null
       if (job.receipt) {
         if (checkpoint && (job.receipt.model.revision !== checkpoint.revision || job.receipt.model.config_sha256 !== checkpoint.configSha256)) { job.receipt = null; throw new Error('The cached checkpoint changed during the benchmark.') }
-        job.snapshot.metadata = structuredClone({ runtime: job.receipt.runtime, model: job.receipt.model, hardware: job.receipt.hardware, settings: job.receipt.settings })
+        job.snapshot.metadata = structuredClone({ suite: job.receipt.suite, runtime: job.receipt.runtime, model: job.receipt.model, hardware: job.receipt.hardware, settings: job.receipt.settings })
         job.snapshot.cells = benchmarkCells(job.receipt)
         job.snapshot.receiptAvailable = true
         job.snapshot.protocolComplete = benchmarkProtocolComplete(job.receipt)

app/electron/agent/engine/benchmark/fake-benchmark.mjs

@@ -8,7 +8,7 @@ import { fileURLToPath } from 'node:url'
 let args = process.argv.slice(2)
 let scenario = 'complete'
 if (args[0]?.startsWith('scenario:')) scenario = args.shift().slice(9)
-const engineVersion = scenario === 'old-version' ? '1.0.12' : scenario === 'unknown-version' ? '9.9.9' : '1.1.0'
+const engineVersion = scenario === 'old-version' ? '1.0.12' : scenario === 'unknown-version' ? '9.9.9' : JSON.parse(readFileSync(new URL('../../../../build/engine.json', import.meta.url), 'utf8')).version
 if (args[0] === '--version') { console.log(`tensorfold-native ${engineVersion}`); process.exit(0) }
 if (args[0] !== 'benchmark') {
   const fixture = path.resolve(path.dirname(fileURLToPath(import.meta.url)), '../test-fixtures/fake-engine.mjs')
@@ -19,8 +19,8 @@ const serial = args.includes('--serial')
 const root = path.join(process.env.HF_HUB_CACHE ?? '', `models--${model.replace('/', '--')}`)
 let revision = 'a'.repeat(40), config = Buffer.from('{}')
 try { revision = readFileSync(path.join(root, 'refs', 'main'), 'utf8').trim(); config = readFileSync(path.join(root, 'snapshots', revision, 'config.json')) } catch {}
-const receipt = { schema_version: 1, suite: { id: 'community-v1', sha256: '9bb717da00666b90b4006174abd9ed46108affc37456b43c64a224a9c938d5be' }, run_id: '00000000-0000-4000-8000-000000000001', created_at: '2026-01-01T00:00:00Z', runtime: { tensorfold_version: scenario === 'mismatched-receipt' ? '1.0.12' : engineVersion, backend: 'mlx', python_version: 'none', platform: 'macos', dependencies: {} }, model: { repo_id: model, revision, family: 'qwen3_5', config_sha256: createHash('sha256').update(config).digest('hex'), tokenizer_sha256: null, quantization: 'mlx-4bit-g64', drafter_repo_id: null, local: false }, hardware: { cpu_model: 'Synthetic CPU', system_memory_bytes: 16 * 1024 ** 3, memory_type: 'unified', gpus: [{ name: 'Synthetic GPU', memory_bytes: null, cores: 8 }], gpu_count: 1, source: 'detected' }, settings: { tokens: 256, repetitions: 5, temperatures: [1, 0], serial, context_tokens: 4096, managed: true, rank_count: 1 }, samples: [] }
-const sample = (fixture_id, temperature, repeat) => ({ fixture_id, temperature, repeat, seed: 1234 + Math.max(0, repeat), status: 'ok', prompt_tokens: 24, completion_tokens: 256, cached_tokens: 0, delivery_seconds: 2.55, delivery_tps: 100, ttft_seconds: 0.1, end_to_end_seconds: 2.8, server_decode_tps: 110, server_decode_seconds: null, prefill_seconds: null, token_sha: 'd'.repeat(12), error_code: null })
+const receipt = { schema_version: 1, suite: { id: 'community-v2', sha256: '51136f5d6775b2ff427834c6b1038b584de27e0a267cfc088ff2694ea069dcdc' }, run_id: '00000000-0000-4000-8000-000000000001', created_at: '2026-01-01T00:00:00Z', runtime: { tensorfold_version: scenario === 'mismatched-receipt' ? '1.0.12' : engineVersion, backend: 'mlx', python_version: 'none', platform: 'macos', dependencies: {} }, model: { repo_id: model, revision, family: 'qwen3_5', config_sha256: createHash('sha256').update(config).digest('hex'), tokenizer_sha256: null, quantization: 'mlx-4bit-g64', drafter_repo_id: null, local: false }, hardware: { cpu_model: 'Synthetic CPU', system_memory_bytes: 16 * 1024 ** 3, memory_type: 'unified', gpus: [{ name: 'Synthetic GPU', memory_bytes: null, cores: 8 }], gpu_count: 1, source: 'detected' }, settings: { tokens: 256, repetitions: 5, temperatures: [1, 0], serial, context_tokens: 4096, managed: true, rank_count: 1, kv_cache: 'bf16' }, samples: [] }
+const sample = (fixture_id, temperature, repeat) => ({ fixture_id, temperature, repeat, seed: 1234 + Math.max(0, repeat), status: 'ok', prompt_tokens: 24, completion_tokens: 256, cached_tokens: 0, delivery_seconds: 2.55, delivery_tps: 100, ttft_seconds: 0.1, end_to_end_seconds: 2.8, server_decode_tps: 110, server_decode_seconds: null, prefill_seconds: null, token_sha: 'd'.repeat(12), error_code: null, finish_reason: 'length', route: fixture_id === 'chat' ? '/v1/chat/completions' : '/v1/completions' })
 const save = () => writeFileSync(output, JSON.stringify(receipt), { mode: 0o600, flag: 'wx' })
 const say = text => scenario === 'stderr' ? console.error(text) : console.log(text)
 if (scenario === 'missing') process.exit(1)

app/electron/agent/engine/benchmark/fixtures.ts

@@ -4,5 +4,5 @@ import { BENCHMARK_ENGINE_VERSION } from './version'
 
 /** Synthetic receipt factory for regression tests, with no user or model output. */
 export function syntheticBenchmarkReceipt(model = 'example-org/synthetic-model'): ModelBenchmarkReceipt {
-  return { schema_version: 1, suite: { id: 'community-v1', sha256: BENCHMARK_SUITE_SHA }, run_id: '00000000-0000-4000-8000-000000000001', created_at: '2026-01-01T00:00:00Z', runtime: { tensorfold_version: BENCHMARK_ENGINE_VERSION, backend: 'mlx', python_version: 'none', platform: 'macos', dependencies: {} }, model: { repo_id: model, revision: 'a'.repeat(40), family: 'qwen3_5', config_sha256: 'b'.repeat(64), tokenizer_sha256: 'c'.repeat(64), quantization: 'mlx-affine-4bit-g64', drafter_repo_id: null, local: false }, hardware: { cpu_model: 'Synthetic CPU', system_memory_bytes: 16 * 1024 ** 3, memory_type: 'unified', gpus: [{ name: 'Synthetic GPU', memory_bytes: null, cores: 8 }], gpu_count: 1, source: 'detected' }, settings: { tokens: 256, repetitions: 5, temperatures: [1, 0], serial: false, context_tokens: 4096, managed: true, rank_count: 1 }, samples: [1, 0].flatMap(temperature => (['code', 'chat'] as const).flatMap(fixture_id => Array.from({ length: 5 }, (_, repeat) => ({ fixture_id, temperature, repeat, seed: 1234 + repeat, status: 'ok' as const, prompt_tokens: 24, completion_tokens: 256, cached_tokens: 0, delivery_seconds: 2.55, delivery_tps: 100, ttft_seconds: 0.1 + repeat * 0.1, end_to_end_seconds: 2.8, server_decode_tps: 110, server_decode_seconds: null, prefill_seconds: null, token_sha: 'd'.repeat(12), error_code: null })))) }
+  return { schema_version: 1, suite: { id: 'community-v2', sha256: BENCHMARK_SUITE_SHA }, run_id: '00000000-0000-4000-8000-000000000001', created_at: '2026-01-01T00:00:00Z', runtime: { tensorfold_version: BENCHMARK_ENGINE_VERSION, backend: 'mlx', python_version: 'none', platform: 'macos', dependencies: {} }, model: { repo_id: model, revision: 'a'.repeat(40), family: 'qwen3_5', config_sha256: 'b'.repeat(64), tokenizer_sha256: 'c'.repeat(64), quantization: 'mlx-affine-4bit-g64', drafter_repo_id: null, local: false }, hardware: { cpu_model: 'Synthetic CPU', system_memory_bytes: 16 * 1024 ** 3, memory_type: 'unified', gpus: [{ name: 'Synthetic GPU', memory_bytes: null, cores: 8 }], gpu_count: 1, source: 'detected' }, settings: { tokens: 256, repetitions: 5, temperatures: [1, 0], serial: false, context_tokens: 4096, managed: true, rank_count: 1, kv_cache: 'bf16' }, samples: [1, 0].flatMap(temperature => (['code', 'chat'] as const).flatMap(fixture_id => Array.from({ length: 5 }, (_, repeat) => ({ fixture_id, temperature, repeat, seed: 1234 + repeat, status: 'ok' as const, prompt_tokens: 24, completion_tokens: 256, cached_tokens: 0, delivery_seconds: 2.55, delivery_tps: 100, ttft_seconds: 0.1 + repeat * 0.1, end_to_end_seconds: 2.8, server_decode_tps: 110, server_decode_seconds: null, prefill_seconds: null, token_sha: 'd'.repeat(12), error_code: null, finish_reason: 'length' as const, route: fixture_id === 'chat' ? '/v1/chat/completions' as const : '/v1/completions' as const })))) }
 }

app/electron/agent/engine/benchmark/receipt.test.ts

@@ -1,6 +1,6 @@
 import assert from 'node:assert/strict'
 import { syntheticBenchmarkReceipt } from './fixtures'
-import { parseBenchmarkReceipt, benchmarkCells, benchmarkProtocolComplete } from './receipt'
+import { parseBenchmarkReceipt, benchmarkCells, benchmarkProtocolComplete, LEGACY_BENCHMARK_SUITE_SHA } from './receipt'
 import { assertBundledBenchmarkVersion, BENCHMARK_ENGINE_VERSION } from './version'
 
 const receipt = syntheticBenchmarkReceipt()
@@ -12,11 +12,28 @@ assert.equal(benchmarkCells(parsed)[0].successful, 5)
 assert.equal(benchmarkCells(parsed)[0].deliveryTokensPerSecond?.median, 100)
 assert.equal(benchmarkCells(parsed)[0].firstTextSeconds?.median, 0.30000000000000004)
 assert.equal(parsed.samples[0].server_decode_seconds, null, 'missing engine duration stays unknown')
-const current = structuredClone(receipt); current.runtime.tensorfold_version = '1.1.0'
-assert.equal(parseBenchmarkReceipt(JSON.stringify(current), { model: current.model.repo_id, serial: false, engineVersion: '1.1.0' }).runtime.tensorfold_version, '1.1.0')
-const historical = structuredClone(receipt); historical.runtime.tensorfold_version = '1.0.12'
-assert.equal(parseBenchmarkReceipt(JSON.stringify(historical)).runtime.tensorfold_version, '1.0.12', 'reviewed historical receipts remain readable after the engine update')
-assert.throws(() => parseBenchmarkReceipt(JSON.stringify(historical), { model: historical.model.repo_id, serial: false, engineVersion: '1.1.0' }), /invalid|unsupported/, 'a historical receipt cannot substitute for a live run on the new engine')
+assert.equal(parseBenchmarkReceipt(JSON.stringify(receipt), { model: receipt.model.repo_id, serial: false, engineVersion: BENCHMARK_ENGINE_VERSION }).suite.id, 'community-v2')
+for (const version of ['1.0.12', '1.1.0']) {
+  const historical = structuredClone(receipt)
+  historical.runtime.tensorfold_version = version
+  historical.suite = { id: 'community-v1', sha256: LEGACY_BENCHMARK_SUITE_SHA }
+  delete historical.settings.kv_cache
+  for (const sample of historical.samples) { delete sample.finish_reason; delete sample.route }
+  assert.equal(benchmarkProtocolComplete(parseBenchmarkReceipt(JSON.stringify(historical))), true, 'reviewed historical receipts remain readable after the engine update')
+  assert.throws(() => parseBenchmarkReceipt(JSON.stringify(historical), { model: historical.model.repo_id, serial: false, engineVersion: BENCHMARK_ENGINE_VERSION }), /invalid|unsupported/, 'a historical receipt cannot substitute for a live run on the new engine')
+  historical.runtime.tensorfold_version = BENCHMARK_ENGINE_VERSION
+  assert.throws(() => parseBenchmarkReceipt(JSON.stringify(historical)), /invalid|unsupported/, 'the current engine must emit its reviewed v2 schema')
+}
+for (const count of [128, 200, 255]) {
+  const natural = structuredClone(receipt)
+  Object.assign(natural.samples[0], { finish_reason: 'stop', completion_tokens: count, delivery_seconds: (count - 1) / 100 })
+  assert.equal(benchmarkProtocolComplete(parseBenchmarkReceipt(JSON.stringify(natural))), true, 'v2 permits sampled natural stops from 128 tokens')
+}
+const chatOnly = structuredClone(receipt)
+for (const sample of chatOnly.samples) sample.route = '/v1/chat/completions'
+assert.equal(benchmarkProtocolComplete(parseBenchmarkReceipt(JSON.stringify(chatOnly))), true, 'chat-only models may consistently use the chat route for code')
+const fp8 = structuredClone(receipt); fp8.settings.kv_cache = 'fp8'
+assert.equal(parseBenchmarkReceipt(JSON.stringify(fp8)).settings.kv_cache, 'fp8')
 for (const mutate of [
   (value: any) => { value.prompt = 'SYNTHETIC_PRIVATE_PROMPT' },
   (value: any) => { value.model.repo_id = '/private/operator/model' },
@@ -32,6 +49,20 @@ for (const mutate of [
   (value: any) => { value.runtime.dependencies.private_path = '/private/operator' },
   (value: any) => { value.runtime.tensorfold_version = '9.9.9' },
   (value: any) => { value.runtime.tensorfold_version = '1.1.0-preview' },
+  (value: any) => { value.runtime.tensorfold_version = '1.1.0' },
+  (value: any) => { delete value.settings.kv_cache },
+  (value: any) => { value.settings.kv_cache = null },
+  (value: any) => { value.settings.kv_cache = 'fp16' },
+  (value: any) => { delete value.samples[0].route },
+  (value: any) => { value.samples[0].route = '/private/path' },
+  (value: any) => { value.samples[5].route = '/v1/completions' },
+  (value: any) => { value.samples[1].route = '/v1/chat/completions' },
+  (value: any) => { delete value.samples[0].finish_reason },
+  (value: any) => { value.samples[0].finish_reason = null },
+  (value: any) => { value.samples[0].finish_reason = 'unknown' },
+  (value: any) => { Object.assign(value.samples[0], { finish_reason: 'stop', completion_tokens: 127, delivery_seconds: 1.26 }) },
+  (value: any) => { Object.assign(value.samples[0], { finish_reason: 'length', completion_tokens: 128, delivery_seconds: 1.27 }) },
+  (value: any) => { Object.assign(value.samples[10], { finish_reason: 'stop', completion_tokens: 128, delivery_seconds: 1.27 }) },
 ]) {
   const invalid = structuredClone(receipt); mutate(invalid)
   assert.throws(() => parseBenchmarkReceipt(JSON.stringify(invalid)), /invalid|unsupported/)