Claude Sonnet 5.5 (effort default (no --effort flag) · thinking off) on Backtest a moving-average crossover on a bundled synthetic price file

AClaude Sonnet 5.5default · thinking off · Claude Code 2.1.299 (headless -p) · cloud · 2026-10-20 · by @bobsiopener

Trust 80%Fully autonomoussuccess · first tryCompare

Estimated cost: $0.06
Wall time: 18s
Human step-ins: 0
Outcome: Pass

Compare with other models on this task →

Exact prompt

# Backtest a moving-average crossover on a bundled synthetic price file `prices.csv` is a **synthetic** daily closing-price series (no real market data), and `input.json` gives the rule's parameters. Write `solve.py` (Python 3 standard library only) so that `python solve.py input.json answer.json` backtests the rule exactly as specified. This is an offline exercise: no data downloads, no broker, no trading. We re-run it on fresh files, so do not hard-code anything. ## Input `input.json`: `prices_file` (CSV path, relative to the working directory), `fast`, `slow` (window lengths, fast < slow), `cost_bps` (cost per position change, in basis points), `initial` (starting equity). The CSV has a header `date,close`; dates are ISO `YYYY-MM-DD`. ## Rules 1. Cleaning: rows can be out of order and a date can appear more than once. For a repeated date, the **last** row in the file wins. Then sort by date. Let c[0..n-1] be the closes; `days` = n. 2. SMA_w(t) = mean of c[t-w+1 .. t] (inclusive). Signal s[t] = 1 if SMA_fast(t) > SMA_slow(t) (strictly), else 0. For t < slow - 1, s[t] = 0. 3. Position: p[0] = 0 and p[t] = s[t-1] for t >= 1 (trade on the next day's close). 4. Equity starts at `initial`. For t = 1..n-1: e[t] = e[t-1] * (1 + p[t] * (c[t]/c[t-1] - 1)) * (1 - cost * |p[t] - p[t-1]|), with cost = cost_bps / 10000. The daily strategy return is e[t]/e[t-1] - 1. 5. `trades` = number of t in 1..n-1 with p[t] != p[t-1]. `exposure` = (sum of p[1..n-1]) / (n - 1). 6. `total_return` = e[n-1]/initial - 1; `cagr` = (e[n-1]/initial) ** (252 / (n - 1)) - 1; `max_drawdown` = the largest value of 1 - e[t]/max(e[0..t]) (0 or positive); `sharpe` = mean(daily returns) / population standard deviation (`statistics.pstdev`) of daily returns * sqrt(252), or 0 if the deviation is 0; `buy_hold_return` = c[n-1]/c[0] - 1. 7. Round `final_equity` (= e[n-1]) to 2 decimals and the other fractions to 6 decimals when writing. ## Output `answer.json` keys: `days`, `final_equity`, `total_return`, `trades`, `cagr`, `max_drawdown`, `sharpe`, `exposure`, `buy_hold_return`. ## Chart (required visual) Also write `chart.svg` in the working directory: a static, self-contained SVG (root element `<svg>`, no scripts, no external images, fonts, stylesheets or links) that can be rendered headless. It must contain `<polyline id="equity">` drawing strategy equity by trading day. ## Scoring The `chart_svg_static` check (one more check, counted in partial credit) passes only if `chart.svg` exists after the fixture run, parses, has root `<svg>`, contains the ids above with the stated tags, and has no `<script>`, `<img>`/`<image>`, `<link>`, `<iframe>`/`<object>`/`<embed>`, `on*` handlers, or `href`/`src` pointing anywhere except `#...` or `data:`. Its drawing is not graded. Hidden checks (partial credit = fraction passed; all must pass): `data_cleaning` (days and buy-and-hold return), `equity_and_trades`, `risk_metrics` on the given files; `fresh_run_1`, `fresh_run_2` (all keys on freshly generated files). Tolerances: `final_equity` within 0.011, fractions within 0.00001, `days` and `trades` exact. Each run has a 60-second limit.

Duplicate dates (last wins) and swapped rows, next-day execution, strict crossover, cost per position change, population-deviation Sharpe and running-peak drawdown. 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).

These are synthetic test cases for comparing AI models, not financial advice.

Success criteria: 1. data_cleaning 2. equity_and_trades 3. risk_metrics 4. fresh_run_1 5. fresh_run_2 6. chart_svg_static

Setup notes: Use Case Scout batch 9 case j11-sma-crossover-backtest (offline fixture). Prepared with batch9/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`. Tools: Read/Write/Edit/Glob/Grep/Bash, web disabled. Single run, no retries. Visual output (chart.svg) rendered offline to PNG with google-chrome headless, network blocked, 1280x800 at ~3 s after load. Claude Code's own reported cost for claude-sonnet-5.5: $0.056675. Score: 6/6. 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.

Output

Claude Sonnet 5.5 (effort default (no --effort flag) · thinking off) on Backtest a moving-average crossover on a bundled synthetic price fileSuccess

Metrics

Estimated cost: $0.06
Wall time: 18s
Human step-ins: 0
Turns: 3

Model usage
Model	Input tok	Output tok	Cost
AClaude Sonnet 5.5anthropic/claude-sonnet-5.5effort default (no --effort flag) · thinking off	59,595	1,812	$0.06

Reproduce this run

Use the use case prompt on the use case page.
Same setup: Claude Sonnet 5.5 (anthropic/claude-sonnet-5.5, effort default (no --effort flag) · thinking off) via Claude Code 2.1.299 (headless -p), cloud.
It counts as a success only if it meets every success criterion: 1. data_cleaning 2. equity_and_trades 3. risk_metrics 4. fresh_run_1 5. fresh_run_2 6. chart_svg_static
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).

Run date is not in the future : passed
Model is in the dated pricing table : passed
Cost recomputed from tokens matches : passed
Cached tokens are consistent : passed
Output throughput is plausible : passed
Autonomy matches step-ins : passed
Step-ins do not exceed turns : passed
Retries agree with first-try result : passed
Evidence log SHA-256 matches : passed
Evidence log is substantial : passed

Evidence log

[
 {
  "duration_api_ms": 17450,
  "stop_reason": "end_turn",
  "session_id": "[REDACTED]",
  "total_cost_usd": 0.0566754,
  "usage": {
   "input_tokens": 6,
   "cache_creation_input_tokens": 8355,
   "cache_read_input_tokens": 51234,
   "output_tokens": 1812,
   "output_tokens_details": {
    "thinking_tokens": 0
   },
   "server_tool_use": {
    "web_search_requests": 0,
    "web_fetch_requests": 0
   },
   "service_tier": "standard",
   "cache_creation": {
    "ephemeral_1h_input_tokens": 8355,
    "ephemeral_5m_input_tokens": 0
   },
   "inference_geo": "global",
   "iterations": [
    {
     "input_tokens": 2,
     "output_tokens": 441,
     "cache_read_input_tokens": 19469,
     "cache_creation_input_tokens": 1425,
     "cache_creation": {
      "ephemeral_5m_input_tokens": 0,
      "ephemeral_1h_input_tokens": 1425
     },
     "type": "message"
    }
   ],
   "speed": "standard",
   "fallback_credit": null
  },
  "modelUsage": {
   "claude-sonnet-5-5": {
    "inputTokens": 6,
    "outputTokens": 1812,
    "cacheReadInputTokens": 51234,
    "cacheCreationInputTokens": 8355,
    "webSearchRequests": 0,
    "costUSD": 0.0566754,
    "contextWindow": 1000000,
    "maxOutputTokens": 128000,
    "thinkingTokens": 0,
    "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": 3,
  "subtype": "success",
  "api_error_status": null,
  "result": "I wrote `solve.py`. Running `python solve.py input.json answer.json` on the bundled files writes `answer.json` and `chart.svg` without errors. I couldn't check the numbers against the hidden checks, and I haven't opened `chart.svg` in a renderer or run a validator on it.\n\nResults on the given files:\n\n| Key | Value |\n|---|---|\n| `days` | 509 |\n| `final_equity` | 8589.77 |\n| `total_return` | -0.141023 |\n| `trades` | 9 |\n| `cagr` | -0.072635 |\n| `max_drawdown` | 0.294082 |\n| `sharpe` | -0.450763 |\n| `exposure` | 0.356299 |\n| `buy_hold_return` | -0.260528 |\n\n- **Rules:** I implemented them as written. For a repeated date the last row wins, then rows are sorted by date. The signal uses strict `>` on the two moving averages, and the position lags the signal by one day.\n- **Malformed rows:** I added one thing the spec doesn't mention. A row whose close isn't a number is skipped instead of crashing the run.\n- **Chart:** `chart.svg` is a self-contained SVG with a `<polyline id=\"equity\">` of strategy equity by trading day. It has a white background and a title, and no scripts, images or links.",
  "ttft_ms": 1949,
  "type": "result",
  "duration_ms": 17803,
  "uuid": "[REDACTED]",
  "ttft_stream_ms": 1473,
  "time_to_request_ms": 245,
  "first_content_frame_ms": 1474,
  "queued_turn_count": 0,
  "result_index": 0
 }
]