live / results / 2026-08-26-qwen-qwen3.8-27b-installed.json
live / results / 2026-08-26-qwen-qwen3.8-27b-installed.json
{
"date": "2026-08-26",
"modelId": "qwen/qwen3.8-27b",
"driver": "installed",
"contextLength": 16384,
"results": [
{
"scenario": "S1",
"title": "inspect → read → edit → verify",
"driver": "installed",
"success": true,
"checks": [
{
"name": "bug fixed in src/math.js",
"ok": true
},
{
"name": "node --test passes",
"ok": true
},
{
"name": "test file untouched",
"ok": true
},
{
"name": "at least one applied transaction",
"ok": true
},
{
"name": "inspected before editing",
"ok": true
},
{
"name": "ran node --test through workspace_command",
"ok": true
}
],
"metrics": {
"toolCalls": 12,
"toolCallsByName": {
"workspace_notes": 2,
"workspace_inspect": 4,
"workspace_tasks": 4,
"workspace_edit": 1,
"workspace_command": 1
},
"rounds": 9,
"invalidToolRequests": 0,
"envelopeViolations": 0,
"wallMs": 26558,
"contextOverflow": false,
"promptTokens": 65008,
"predictedTokens": 2074
},
"contextLength": 16384
},
{
"scenario": "S3",
"title": "plan mode: propose, /accept, execute",
"driver": "installed",
"success": true,
"checks": [
{
"name": "no mutation before approval",
"ok": true
},
{
"name": "plan proposed in phase 1",
"ok": true
},
{
"name": "exactly one plan approved",
"ok": true
},
{
"name": "bug fixed in src/math.js",
"ok": true
},
{
"name": "node --test passes",
"ok": true
},
{
"name": "test file untouched",
"ok": true
},
{
"name": "plan proposals after /accept: 0",
"ok": true
},
{
"name": "plan final status: completed",
"ok": true
},
{
"name": "workspace_plan complete called: true",
"ok": true
},
{
"name": "APPROVAL_REQUIRED before approval: 0",
"ok": true
},
{
"name": "stray planned transactions: 0",
"ok": true
}
],
"metrics": {
"toolCalls": 10,
"toolCallsByName": {
"workspace_notes": 2,
"workspace_inspect": 4,
"workspace_plan": 2,
"workspace_edit": 1,
"workspace_command": 1
},
"rounds": 8,
"invalidToolRequests": 0,
"envelopeViolations": 0,
"wallMs": 24473,
"contextOverflow": false,
"promptTokens": 60406,
"predictedTokens": 1895
},
"contextLength": 16384
},
{
"scenario": "S9",
"title": "full project from an empty directory",
"driver": "installed",
"success": false,
"checks": [
{
"name": "package.json parses with scripts.test running node --test",
"ok": true
},
{
"name": "at least one file under src/",
"ok": true
},
{
"name": "at least one test file under tests/",
"ok": true
},
{
"name": "node --test exits 0 with at least three ok lines",
"ok": true
},
{
"name": "score: 4/4",
"ok": true
},
{
"name": "files created: package.json, src/wordcount.js, tests/wordcount.test.js",
"ok": true
},
{
"name": "ok lines: 3",
"ok": true
},
{
"name": "node --test exit: 0",
"ok": true
},
{
"name": "package private: true",
"ok": true
},
{
"name": "applied transactions: 1",
"ok": true
},
{
"name": "model ran node --test: yes (1)",
"ok": true
}
],
"metrics": {
"toolCalls": 12,
"toolCallsByName": {
"workspace_notes": 1,
"workspace_inspect": 1,
"workspace_tasks": 6,
"workspace_edit": 1,
"workspace_command": 2,
"workspace_vcs": 1
},
"rounds": 9,
"invalidToolRequests": 1,
"envelopeViolations": 0,
"wallMs": 44030,
"contextOverflow": true,
"promptTokens": 73281,
"predictedTokens": 3812
},
"contextLength": 16384,
"error": "context overflow: prompt 17368 > 16384"
}
],
"note": "Config verification: the installed session's tool list proves each scenario's `pluginConfig` reached the plugin —\n8 tools with no `workspace_plan` in S1 and S9 (permissionMode 2, auto) and 9 tools including it in S3 (permissionMode 1, plan).\n\nS9 ended on a hard prompt overflow: once the accumulated prompt exceeds the window LM Studio's engine returns 400,\nwhich `contextOverflowPolicy: \"stopAtLimit\"` cannot prevent (it stops generation, it cannot shrink an already-oversized prompt).\nAll 11 checks still passed — wordcount was built and `node --test`-green on disk before the overflow.\nTwo in-process control runs at the same context length reproduced the same hard overflow (prompt 18337 and 16629 > 16384, 4/4 scored both times),\nso this is S9 sitting at the edge of a 16384-token window for this model, not something the installed/SDK boundary introduces;\nthe committed in-process S9 row that passed with a warning was a sample that happened to land under the ceiling.\nS1 and S3 match the in-process rows check for check, and S3 matches its tool-call profile exactly."
}
{
"date": "2026-08-26",
"modelId": "qwen/qwen3.8-27b",
"driver": "installed",
"contextLength": 16384,
"results": [
{
"scenario": "S1",
"title": "inspect → read → edit → verify",
"driver": "installed",
"success": true,
"checks": [
{
"name": "bug fixed in src/math.js",
"ok": true
},
{
"name": "node --test passes",
"ok": true
},
{
"name": "test file untouched",
"ok": true
},
{
"name": "at least one applied transaction",
"ok": true
},
{
"name": "inspected before editing",
"ok": true
},
{
"name": "ran node --test through workspace_command",
"ok": true
}
],
"metrics": {
"toolCalls": 12,
"toolCallsByName": {
"workspace_notes": 2,
"workspace_inspect": 4,
"workspace_tasks": 4,
"workspace_edit": 1,
"workspace_command": 1
},
"rounds": 9,
"invalidToolRequests": 0,
"envelopeViolations": 0,
"wallMs": 26558,
"contextOverflow": false,
"promptTokens": 65008,
"predictedTokens": 2074
},
"contextLength": 16384
},
{
"scenario": "S3",
"title": "plan mode: propose, /accept, execute",
"driver": "installed",
"success": true,
"checks": [
{
"name": "no mutation before approval",
"ok": true
},
{
"name": "plan proposed in phase 1",
"ok": true
},
{
"name": "exactly one plan approved",
"ok": true
},
{
"name": "bug fixed in src/math.js",
"ok": true
},
{
"name": "node --test passes",
"ok": true
},
{
"name": "test file untouched",
"ok": true
},
{
"name": "plan proposals after /accept: 0",
"ok": true
},
{
"name": "plan final status: completed",
"ok": true
},
{
"name": "workspace_plan complete called: true",
"ok": true
},
{
"name": "APPROVAL_REQUIRED before approval: 0",
"ok": true
},
{
"name": "stray planned transactions: 0",
"ok": true
}
],
"metrics": {
"toolCalls": 10,
"toolCallsByName": {
"workspace_notes": 2,
"workspace_inspect": 4,
"workspace_plan": 2,
"workspace_edit": 1,
"workspace_command": 1
},
"rounds": 8,
"invalidToolRequests": 0,
"envelopeViolations": 0,
"wallMs": 24473,
"contextOverflow": false,
"promptTokens": 60406,
"predictedTokens": 1895
},
"contextLength": 16384
},
{
"scenario": "S9",
"title": "full project from an empty directory",
"driver": "installed",
"success": false,
"checks": [
{
"name": "package.json parses with scripts.test running node --test",
"ok": true
},
{
"name": "at least one file under src/",
"ok": true
},
{
"name": "at least one test file under tests/",
"ok": true
},
{
"name": "node --test exits 0 with at least three ok lines",
"ok": true
},
{
"name": "score: 4/4",
"ok": true
},
{
"name": "files created: package.json, src/wordcount.js, tests/wordcount.test.js",
"ok": true
},
{
"name": "ok lines: 3",
"ok": true
},
{
"name": "node --test exit: 0",
"ok": true
},
{
"name": "package private: true",
"ok": true
},
{
"name": "applied transactions: 1",
"ok": true
},
{
"name": "model ran node --test: yes (1)",
"ok": true
}
],
"metrics": {
"toolCalls": 12,
"toolCallsByName": {
"workspace_notes": 1,
"workspace_inspect": 1,
"workspace_tasks": 6,
"workspace_edit": 1,
"workspace_command": 2,
"workspace_vcs": 1
},
"rounds": 9,
"invalidToolRequests": 1,
"envelopeViolations": 0,
"wallMs": 44030,
"contextOverflow": true,
"promptTokens": 73281,
"predictedTokens": 3812
},
"contextLength": 16384,
"error": "context overflow: prompt 17368 > 16384"
}
],
"note": "Config verification: the installed session's tool list proves each scenario's `pluginConfig` reached the plugin —\n8 tools with no `workspace_plan` in S1 and S9 (permissionMode 2, auto) and 9 tools including it in S3 (permissionMode 1, plan).\n\nS9 ended on a hard prompt overflow: once the accumulated prompt exceeds the window LM Studio's engine returns 400,\nwhich `contextOverflowPolicy: \"stopAtLimit\"` cannot prevent (it stops generation, it cannot shrink an already-oversized prompt).\nAll 11 checks still passed — wordcount was built and `node --test`-green on disk before the overflow.\nTwo in-process control runs at the same context length reproduced the same hard overflow (prompt 18337 and 16629 > 16384, 4/4 scored both times),\nso this is S9 sitting at the edge of a 16384-token window for this model, not something the installed/SDK boundary introduces;\nthe committed in-process S9 row that passed with a warning was a sample that happened to land under the ceiling.\nS1 and S3 match the in-process rows check for check, and S3 matches its tool-call profile exactly."
}