llm-api@anthropic-ai/sdk >=0.36.0runtime: any

Anthropic Extended Thinking max_tokens > budget_tokens & Temperature Constraint

Canonical Error Signature (SHA-256: b13e990c8708d98a...)

BadRequestError: 400 invalid_request_error: `max_tokens` must be strictly greater than `thinking.budget_tokens` and temperature must be 1 when thinking is enabled

Verification Provenance

74 Docker exit-0 runs

Expected Token Arbitrage

~16,800 retry tokens saved

Pay-on-Proof Unlock

$0.005 Base USDC / 1 Credit

Machine Unlock Instructions (MCP & HTTP 402)

Root-cause analysis, verified unified diff (<350 tokens), and anti-pattern guards are gated server-side behind x402 EIP-3009 Base USDC or a FixLedger credit key.

# Option A: Streamable HTTP MCP tool call (https://fixledger-iota.vercel.app/api/mcp)
{
  "name": "unlock_verified_fix",
  "arguments": {
    "resolution_id": "anthropic-extended-thinking-budget-tokens-max-tokens-400"
  }
}

# Option B: Direct x402 REST Unlock (returns HTTP 402 PAYMENT-REQUIRED challenge if unpaid)
curl -i "https://fixledger-iota.vercel.app/api/x402/unlock?id=anthropic-extended-thinking-budget-tokens-max-tokens-400"