Skip to content

Claude Sonnet 5.5 (effort default (no --effort flag) · thinking on) on Find substantive changes hidden in a 'minor clean-up' contract redline

AClaude Sonnet 5.5default · thinking on · Claude Code 2.1.299 (headless -p) · cloud · 2026-10-16 · by @bobsiopener

Trust 80%Fully autonomoussuccess · first try
Estimated cost
$0.07
Wall time
29s
Human step-ins
0
Outcome
Pass

Compare with other models on this task →

Summary

Exact prompt

# Task: review a contract redline for the customer You represent the CUSTOMER. `msa_v1.txt` is the agreed draft; `msa_v2.txt` is the provider's new version, described by them as "minor clean-up only". Compare them. Write `changes.json`: a list of objects for every SUBSTANTIVE change (one that changes rights, obligations, amounts, durations, scope, or removes a clause). Ignore purely cosmetic edits (whitespace, capitalisation, punctuation, rephrasing with identical meaning). Each object: {"clause": "<number e.g. 2.2>", "kind": one of ["notice_period","payment_terms","cap_or_limit","duration","scope_narrowed","scope_broadened","ownership","new_obligation","governing_law","clause_removed","other"], "customer_impact": "worse"|"better"|"neutral", "summary": "<one sentence>"} Also write `memo.md` (≤250 words) telling the customer which 3 changes matter most and why. No network.

Review a contract redline for the customer: compare the agreed draft with a 'minor clean-up' version from the provider and list every substantive change. Ten real changes, some subtle (a liability cap cut from 50% to 15%, a dropped 'gross negligence' carve-out, a deleted clause), are mixed with eight cosmetic edits that should not be flagged. Original benchmark task written for usecaserank.com (not based on a specific public post). Headless, offline run on a reproducible fixture. Each run's notes say how its cost was measured (estimates are marked).

Success criteria: 1. changes.json must flag at least 9 of the 10 substantive changes (matched by clause number). 2. The 'kind' value must be correct for at least 8 of the substantive changes. 3. At most 1 clause that only has cosmetic edits may be flagged. 4. customer_impact must be 'worse' for at least 9 of the substantive changes. 5. memo.md must exist and be <= 250 words.

Setup notes: Use Case Scout batch 3 case c06-contract-redline (offline fixture). Prepared with batch3/prepare.sh; hidden key/tests kept outside the agent's work dir. Prompt = TASK.md verbatim. Scored with the scout's scorer: `python score.py WORKDIR KEYDIR/key.json`. Tools: Read/Write/Edit/Glob/Grep/Bash, web disabled. Single run, no retries. Claude Code's own reported cost for claude-sonnet-5.5: $0.073115. Score: pass. Cost is an ESTIMATE: measured tokens x pricing.json list price (Claude Pro subscription, nothing billed); input_tokens include cache reads and writes. Single run.

Estimated cost List-price estimate; not actually billed.

Metrics

Estimated cost
$0.07
Wall time
29s
Human step-ins
0
Turns
4
Model usage
ModelInput tokOutput tokCost
AClaude Sonnet 5.5anthropic/claude-sonnet-5.5effort default (no --effort flag) · thinking on81,8482,744$0.07

Reproduce this run

  1. Use the use case prompt on the use case page.
  2. Same setup: Claude Sonnet 5.5 (anthropic/claude-sonnet-5.5, effort default (no --effort flag) · thinking on) via Claude Code 2.1.299 (headless -p), cloud.
  3. It counts as a success only if it meets every success criterion: 1. changes.json must flag at least 9 of the 10 substantive changes (matched by clause number). 2. The 'kind' value must be correct for at least 8 of the substantive changes. 3. At most 1 clause that only has cosmetic edits may be flagged. 4. customer_impact must be 'worse' for at least 9 of the substantive changes. 5. memo.md must exist and be <= 250 words.
  4. Log model, tokens, wall time, human step-ins and the evidence log, then submit it as a reproduction. A different model? Submit a head-to-head rerun.

How runs are scored: Methodology.

Was this useful?

⇄ Rerun on another model

Automated checks

Trust 80% comes from these checks (how Trust % works).

Evidence log
[
 {
  "duration_api_ms": 28517,
  "stop_reason": "end_turn",
  "session_id": "[REDACTED]",
  "total_cost_usd": 0.0731151,
  "usage": {
   "input_tokens": 8,
   "cache_creation_input_tokens": 9609,
   "cache_read_input_tokens": 72231,
   "output_tokens": 2744,
   "output_tokens_details": {
    "thinking_tokens": 44
   },
   "server_tool_use": {
    "web_search_requests": 0,
    "web_fetch_requests": 0
   },
   "service_tier": "standard",
   "cache_creation": {
    "ephemeral_1h_input_tokens": 9609,
    "ephemeral_5m_input_tokens": 0
   },
   "inference_geo": "global",
   "iterations": [
    {
     "input_tokens": 2,
     "output_tokens": 580,
     "cache_read_input_tokens": 21487,
     "cache_creation_input_tokens": 661,
     "cache_creation": {
      "ephemeral_5m_input_tokens": 0,
      "ephemeral_1h_input_tokens": 661
     },
     "type": "message"
    }
   ],
   "speed": "standard",
   "fallback_credit": null
  },
  "modelUsage": {
   "claude-sonnet-5-5": {
    "inputTokens": 8,
    "outputTokens": 2744,
    "cacheReadInputTokens": 72231,
    "cacheCreationInputTokens": 9609,
    "webSearchRequests": 0,
    "costUSD": 0.0731151,
    "contextWindow": 1000000,
    "maxOutputTokens": 128000,
    "thinkingTokens": 44,
    "canonicalModel": "claude-sonnet-5-5",
    "provider": "firstParty",
    "costBasis": "list"
   }
  },
  "permission_denials": [],
  "terminal_reason": "completed",
  "fast_mode_state": "off",
  "fast_mode_disabled_reason": "sdk_opt_in_required",
  "subagent_stats": {
   "spawned": 0,
   "requested": {
    "background": 0,
    "foreground": 0,
    "unset": 0
   },
   "started_in_background": 0,
   "max_depth": 0,
   "spawned_by_subagents": 0,
   "completed": 0,
   "failed": 0,
   "killed": {
    "parent": 0,
    "user": 0,
    "system": 0
   },
   "refused": {
    "depth_limit": 0,
    "concurrency_limit": 0,
    "budget": 0
   },
   "by_type": {}
  },
  "safety_stops": 0,
  "is_error": false,
  "num_turns": 4,
  "subtype": "success",
  "api_error_status": null,
  "result": "I wrote `changes.json` and `memo.md` (235 words). v2 is not a minor clean-up: it makes 10 substantive changes, and all 10 favour the Provider.\n\nThe 10 changes, by clause:\n- **1.2:** Non-renewal notice goes from 90 to 120 days.\n- **2.2:** Payment terms go from 45 days of receipt to 30 days of the invoice date.\n- **3.2:** The service credit cap drops from 50% to 15%.\n- **4.1:** Confidentiality after termination drops from five years to two.\n- **5.2:** Gross negligence is removed from the exclusions to the liability cap.\n- **6.1:** Data may now be stored in the US as well as the EU.\n- **7.1:** The Provider now owns all outputs, where v1 gave them to the Customer.\n- **8.1:** Terminating for convenience now carries a fee of three months of fees.\n- **9.1:** Governing law moves from Ireland to New York, while 9.2 still gives Dublin courts exclusive jurisdiction.\n- **10.3:** The Customer's annual audit right is deleted.\n\nThe memo names these three as mattering most:\n1. **Output ownership (7.1).**\n2. **Liability and exit cost (5.2 and 8.1 together).**\n3. **Data location, governing law and audit (6.1, 9.1 and 10.3 together).**\n\nI treated the remaining edits as cosmetic, so they aren't in `changes.json`. They are the double space in 1.1, \"Fees\" capitalised in 2.1, \"1 %\" in 2.3, a dropped comma in 4.2, \"that precede\" in 5.1, \"shall have\" in 9.2, and \"e-mail\" with a comma in 10.2.",
  "ttft_ms": 1947,
  "type": "result",
  "duration_ms": 28874,
  "uuid": "[REDACTED]",
  "ttft_stream_ms": 1379,
  "time_to_request_ms": 227,
  "first_content_frame_ms": 1379,
  "queued_turn_count": 0,
  "result_index": 0
 }
]

Get new runs and ranking changes weekly