Claude Haiku 5.5 (effort default (no --effort flag) · thinking on) on Diverging stacked bar chart with exact SVG geometry

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

Trust 80%Fully autonomoussuccess · first trySupersededCompare

Estimated cost: $0.0075
Wall time: 30s
Human step-ins: 0
Outcome: Pass

Compare with other models on this task →

Superseded. This run was rescored on 2026-10-20 and replaced by a newer run. It no longer counts toward rankings, use-case tables or the data export; it is kept here for the record.

Exact prompt

# Diverging stacked bar chart with exact geometry `input.json`: `categories` (list of names) and `series` (each `{name, color, values}`, one value per category; `null` means 0). Write `solve.py` (Python 3 standard library only); `python solve.py input.json answer.json` must write `answer.json` and a static `chart.svg` in the working directory. Fresh inputs are generated. ## Rules 1. Canvas 720x360. Plot area x0=60, y0=20, width 600, height 300. 2. For each category, `pos` = sum of positive values and `neg` = sum of negative values (null and 0 add nothing). `totals` = [[pos, neg], ...] in category order. 3. Raw domain: lo = min(0, min neg), hi = max(0, max pos). 4. Nice ticks: raw = (hi - lo)/5 (use 1 if hi == lo); mag = 10^floor(log10(raw)); f = raw/mag; m = 1 if f<=1, 2 if f<=2, 2.5 if f<=2.5, 5 if f<=5, otherwise 10; step = m*mag. Axis min = floor(lo/step)*step, axis max = ceil(hi/step)*step (if they are equal, add one step). Round the quotient to 9 decimals before floor/ceil so float noise doesn't add a step. Ticks run from axis min to axis max in steps (each rounded to 10 decimals). `domain` = [axis min, axis max]. 5. Tick label: the smallest d in 0..6 such that step rounded to d decimals equals step; format each tick with d decimals; a label equal to zero prints as `0`. 6. y(v) = y0 + 300*(axis max - v)/(axis max - axis min). 7. Band = 600/number of categories. Bar x = x0 + i*band + 0.15*band; width = 0.7*band. 8. Stacking: series in the given order. Positive values stack **upward from 0** on the positive stack; negative values stack **downward from 0** on a separate negative stack. A value of 0 or null draws no bar. 9. A bar from value bot to top has y = y(top) and height = y(bot) - y(top). Round x, y, width and height to 2 decimals. 10. Bar id `bar-<category index>-<series index>` (0-based), with fill = the series color. ## Outputs - `answer.json`: `{"domain", "ticks", "tick_labels", "totals", "bars": [{"id","x","y","width","height","fill"}, ...]}`, with bars in category order and then series order. - `chart.svg`: root `<svg>`, no scripts, external resources or animation. It contains `<g id="bars">` holding one `<rect id="bar-i-j" x y width height fill>` per bar, with numeric attributes equal to the JSON values (within 0.01). For tick k: `<line id="tick-k">` at y(tick) and `<text id="tick-label-k">` holding the label. Also `<line id="zero-line">` at y(0). ## Scoring Hidden checks (partial credit; all must pass): `stack_totals` (1e-6), `nice_ticks_domain` (exact labels), `bar_geometry_json` (0.011), `chart_svg_static` (parses as static SVG with the required ids and tags), `chart_svg_geometry` (rect attributes and tick-label texts read from the SVG itself), `fresh_run_1`, `fresh_run_2` (domain, labels, bars and SVG geometry on new data). 30-second limit per run.

Fixes the subjective look of a chart to exact numbers. Separate +/- stacks, the 2.5 step, float-noise floors and label decimals are easy to get slightly wrong. 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. stack_totals 2. nice_ticks_domain 3. bar_geometry_json 4. chart_svg_static 5. chart_svg_geometry 6. fresh_run_1 7. fresh_run_2

Setup notes: Use Case Scout batch 12 case m11-stacked-chart-geometry (offline, synthetic fixture). Prepared with batch12/prepare.sh; hidden key/checker kept outside the agent's work dir (temp dir during the run). Prompt = TASK.md verbatim. Scored with the case's scorer: `python score.py WORKDIR KEYDIR` (python3, run from the case dir). Tools: Read/Write/Edit/Glob/Grep/Bash, web disabled. Single run, no retries. 1-min load average at scoring 0.22 (8 cores). chart.svg regenerated for preview by running the agent's solve.py in a scratch copy. chart.svg rendered offline with google-chrome headless (CDP pipe, network blocked): PNG 1280x800 at ~3 s after load; MP4 ~6 s captured at 1280x720, H.264 1280x720. Claude Code's own reported cost for claude-haiku-5.5: $0.007529. Score: 7/7. 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

Metrics

Estimated cost: $0.0075
Wall time: 30s
Human step-ins: 0
Turns: 5

Model usage
Model	Input tok	Output tok	Cost
AClaude Haiku 5.5anthropic/claude-haiku-5.5effort default (no --effort flag) · thinking on	121,489	6,393	$0.0075

Reproduce this run

Use the use case prompt on the use case page.
Same setup: Claude Haiku 5.5 (anthropic/claude-haiku-5.5, effort default (no --effort flag) · thinking on) via Claude Code 2.1.299 (headless -p), cloud.
It counts as a success only if it meets every success criterion: 1. stack_totals 2. nice_ticks_domain 3. bar_geometry_json 4. chart_svg_static 5. chart_svg_geometry 6. fresh_run_1 7. fresh_run_2
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": 28572,
  "stop_reason": "end_turn",
  "session_id": "[REDACTED]",
  "total_cost_usd": 0.00752924,
  "usage": {
   "input_tokens": 10,
   "cache_creation_input_tokens": 16405,
   "cache_read_input_tokens": 105074,
   "output_tokens": 6393,
   "output_tokens_details": {
    "thinking_tokens": 2819
   },
   "server_tool_use": {
    "web_search_requests": 0,
    "web_fetch_requests": 0
   },
   "service_tier": "standard",
   "cache_creation": {
    "ephemeral_1h_input_tokens": 16405,
    "ephemeral_5m_input_tokens": 0
   },
   "inference_geo": "global",
   "iterations": [
    {
     "input_tokens": 2,
     "output_tokens": 624,
     "cache_read_input_tokens": 27683,
     "cache_creation_input_tokens": 1261,
     "cache_creation": {
      "ephemeral_5m_input_tokens": 0,
      "ephemeral_1h_input_tokens": 1261
     },
     "type": "message"
    }
   ],
   "speed": "standard",
   "fallback_credit": null
  },
  "modelUsage": {
   "claude-haiku-5-5": {
    "inputTokens": 10,
    "outputTokens": 6393,
    "cacheReadInputTokens": 105074,
    "cacheCreationInputTokens": 16405,
    "webSearchRequests": 0,
    "costUSD": 0.00752924,
    "contextWindow": 1000000,
    "maxOutputTokens": 128000,
    "thinkingTokens": 2819,
    "canonicalModel": "claude-haiku-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": 5,
  "subtype": "success",
  "api_error_status": null,
  "result": "I wrote `solve.py` in the work directory. It runs in about 0.05 s and produces `answer.json` and `chart.svg` for your input.\n\nI checked the output by hand on your input: the domain is [-50, 200] with step 50, and C1's S1 bar sits at y = 213.2. C4's positive stack runs 0 \u2192 63 \u2192 177 \u2192 180, which matches the rules. I also ran two edge cases: all-null data, which gives domain [0, 1] with no bars, and fractional values, which give the expected ticks. I have not run it against the hidden checks or fresh inputs.\n\nChoices I made where the spec was open:\n- `chart.svg` goes to the current working directory, as the spec says, even when `answer.json` is written elsewhere.\n- Totals and tick values are rounded to 10 decimals to strip float noise. The `step` itself is also rounded to 10 decimals.\n- The SVG adds a `<g id=\"axis\">` wrapper around the tick lines and labels, plus a small text anchor and font size for readability. The required ids and tags are all present.\n\nThis is not a git repository, so I did not commit anything.",
  "ttft_ms": 1481,
  "type": "result",
  "duration_ms": 29167,
  "uuid": "[REDACTED]",
  "ttft_stream_ms": 1279,
  "time_to_request_ms": 290,
  "first_content_frame_ms": 1279,
  "queued_turn_count": 0,
  "result_index": 0
 }
]