✓ PASS

any-llm 1.40.2

truncation.reasoning-only · anthropic

Requirements

Id	Level	Status	Evidence
cause-is-truncation	MUST	SATISFIED	finish.value="length" (primary: choices[0].finish_reason); truth.finish="length"; error.category=null; structured.returned_as_success=false
text-reported	MUST	SATISFIED	text.value="" (primary); truth.text=null
usage-reported	SHOULD	SATISFIED	usage.output=64 (primary); truth.usage.output=64

What the provider sent

Finish: length
Raw finish: max_tokens
Text: null
Tool calls: []
Usage: { "alternatives": {}, "cached_input": null, "input": 25, "output": 64, "raw": { "input_tokens": 25, "output_tokens": 64, "output_tokens_details": { "thinking_tokens": 64 } }, "reasoning": 64, "total": 89 }
Stream completed: null
Error: null

Wire excerpt

status: 200
content-type: application/json
body (json):
{
  "id": "msg_01Aq9w938a90dw8q",
  "type": "message",
  "role": "assistant",
  "content": [
    {
      "type": "thinking",
      "thinking": "",
      "signature": "EqQBCgIYAhIM1gbcDa9GJwZA2b3hGgxBdjrkzLoky3dl1pkiMOYds..."
    }
  ],
  "model": "claude-sonnet-5-5",
  "stop_reason": "max_tokens",
  "stop_sequence": null,
  "stop_details": null,
  "usage": {
    "input_tokens": 25,
    "output_tokens": 64,
    "output_tokens_details": {
      "thinking_tokens": 64
    }
  }
}

What the library reported

Outcome

returned

Finish

{
  "raw": "length",
  "source": "choices[0].finish_reason",
  "surface": "primary",
  "value": "length"
}

Text

Value
Surface: primary

Tool calls

[]

Usage

{
  "cached_input": null,
  "input": 25,
  "output": 64,
  "reasoning": null,
  "surface": {
    "cached_input": "absent",
    "input": "primary",
    "output": "primary",
    "reasoning": "absent",
    "total": "primary"
  },
  "total": 89
}

Stream

{
  "chunks_seen": null,
  "completed": null,
  "used": false
}

Error

null

Unsupported reason

null

Full observation JSON

{
  "elapsed_ms": 148.8561499863863,
  "error": null,
  "finish": {
    "raw": "length",
    "source": "choices[0].finish_reason",
    "surface": "primary",
    "value": "length"
  },
  "format": "wiretruth.observation/v1",
  "library": {
    "api": "any_llm.completion",
    "language": "python",
    "name": "any-llm",
    "version": "1.40.2"
  },
  "notes": [],
  "outcome": "returned",
  "stream": {
    "chunks_seen": null,
    "completed": null,
    "used": false
  },
  "structured": {
    "attempted": false,
    "returned_as_success": false,
    "value": null
  },
  "text": {
    "surface": "primary",
    "value": ""
  },
  "tool_calls": [],
  "unsupported_reason": null,
  "usage": {
    "cached_input": null,
    "input": 25,
    "output": 64,
    "reasoning": null,
    "surface": {
      "cached_input": "absent",
      "input": "primary",
      "output": "primary",
      "reasoning": "absent",
      "total": "primary"
    },
    "total": 89
  }
}

Fixture

documented anthropic/truncation.reasoning-only

Provenance	Value
Source	https://platform.claude.com/docs/en/build-with-claude/thinking#controlling-thinking-display
Notes	Documented shapes, not a recording. Message envelope and stop fields follow handling-stop-reasons; content blocks follow thinking/streaming and handle-tool-calls. Model changed to claude-sonnet-5-5; IDs and scenario text are filled in. Token counts are illustrative, filled in to include thinking within the requested output limit, not measured. ASSUMED max_tokens envelope with exactly one omitted thinking block and no text. Empty thinking plus signature follows omitted display; signature is the literal streaming placeholder. The docs do not print this complete truncated response. The output_tokens_details.thinking_tokens breakdown follows thinking-steering.md (https://platform.claude.com/docs/en/build-with-claude/thinking-steering-and-cost#pricing); the thinking count of 64 is filled in, not measured.

Provenance

Value

Source

https://platform.claude.com/docs/en/build-with-claude/thinking#controlling-thinking-display

Notes

Documented shapes, not a recording. Message envelope and stop fields follow handling-stop-reasons; content blocks follow thinking/streaming and handle-tool-calls. Model changed to claude-sonnet-5-5; IDs and scenario text are filled in. Token counts are illustrative, filled in to include thinking within the requested output limit, not measured. ASSUMED max_tokens envelope with exactly one omitted thinking block and no text. Empty thinking plus signature follows omitted display; signature is the literal streaming placeholder. The docs do not print this complete truncated response. The output_tokens_details.thinking_tokens breakdown follows thinking-steering.md (https://platform.claude.com/docs/en/build-with-claude/thinking-steering-and-cost#pricing); the thinking count of 64 is filled in, not measured.

Reproduce

wiretruth serve anthropic/truncation.reasoning-only

Plain-library snippet from the shim manifest.

any_llm.completion(model=model, provider='anthropic', api_base=WIRETRUTH_BASE_URL_ANTHROPIC, api_key=WIRETRUTH_API_KEY); host root, SDK appends /v1/messages and uses x-api-key

Upstream

No matching upstream report.

History

Version	Grade
1.40.2	✓ PASS
1.38.0	✓ PASS
1.33.0	✓ PASS