Skip to content

Claude Haiku 5.5 (effort default (no --effort flag) · thinking on) on Build a TV home screen that shows a smart home's device status instead of ads

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

Trust 80%Fully autonomoussuccess · first trySuperseded
Estimated cost
$0.02
Wall time
2m
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Superseded.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.

Summary

Exact prompt

# TV home screen for a (synthetic) smart home `home.json` describes one synthetic home: `as_of`, `local_utc_offset_hours`, `night`, `offline_after_minutes`, `temp_drift_f`, `rooms` (display order) and `devices`. Only these listed devices exist; nothing is contacted. Write `solve.py` (Python 3 standard library only, no network). `python solve.py input.json answer.json` (first argument unused; read `home.json` from the current directory) writes `answer.json` and `home.html`. Fresh homes are generated when scoring. ## Rules 1. Timestamps carry their own offsets (`Z`, `-04:00`, `-07:00`, `+01:00` ...); compare instants, not wall-clock text. 2. A device is `offline` when as_of - last_seen > `offline_after_minutes` (strictly), else `online`. 3. Tile state (`data-state`) for an online device: light/plug `on`|`off`; lock `locked`|`unlocked`; contact `open`|`closed`; leak `wet`|`dry`; camera `recording`|`idle`; thermostat = current temperature in Fahrenheit with 1 decimal (e.g. `68.5`). Celsius values convert as F = C x 9/5 + 32, rounded half-up to 1 decimal (do the same for setpoints). An offline device's state is `unknown`. 4. It is night when the local hour of as_of (using `local_utc_offset_hours`) is >= night.start_hour or < night.end_hour. 5. Alerts (offline devices raise only `offline`, nothing else). Kinds, in severity order: `leak` (leak sensor wet), `unlocked_at_night` (lock unlocked at night), `door_open` (contact open and as_of - open_since > 30 min), `offline`, `temp_drift` (|current F - setpoint F| > temp_drift_f, both 1-decimal F values), `low_battery` (battery not null and < 20). One device can raise several alerts. 6. Alert order: severity, then the device's room position in `rooms`, then device name (case-insensitive), then id. 7. Summary: `lights_on` (online lights that are on), `online`, `offline`, `alerts` (count), `avg_temp_f` (mean of online thermostats' 1-decimal F values, rounded half-up to 1 decimal; null if none). 8. Room slug: lower-case, every run of non-alphanumeric characters becomes `-`, leading/trailing `-` removed (`Kid's Room` -> `kid-s-room`). ## answer.json `{"rooms": [{"name", "slug", "devices": [{"id", "status", "state"}] sorted by name case-insensitive}] in rooms order, "alerts": [{"device": id, "kind"}] in alert order, "summary": {"lights_on", "online", "offline", "alerts", "avg_temp_f"}}` ## Visual output: home.html (a 1920x1080 TV screen) Static and self-contained: no `<script>`, `<img>`, `<link>`, `<iframe>`/`<object>`/`<embed>`, `on*` handlers, external URLs, CSS url(http)/@import or fetches. - One `<section id="room-<slug>">` per room, in `rooms` order, containing one element `id="dev-<id>"` per device with `data-status` and `data-state` as above. - An element `id="alerts"` containing one element `id="alert-1"`, `id="alert-2"`, ... per alert in order, each with `data-device` and `data-kind`. - An element `id="summary"` with `data-lights-on`, `data-online`, `data-offline`, `data-alerts`, `data-avg-temp-f` (1 decimal, or `none`). - In a 1920x1080 browser viewport the page must fit without scrolling, `#alerts` must be fully on screen, and every device tile must be at least 40x40 px and on screen. Styling is otherwise up to you. ## Scoring Hidden checks (partial credit; all must pass): `html_static_rooms_in_order`, `device_tiles_status_state`, `alerts_rules_and_order`, `summary_counts`, `browser_fits_1080p_no_network` (headless browser, network blocked) on the given home; `fresh_run_1`, `fresh_run_2` (answer.json and home.html on new homes). 30-second limit per run.

The page looks right long before the data is right: one naive timestamp parse flips several devices online/offline and changes every alert, count and tile. 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. html_static_rooms_in_order 2. device_tiles_status_state 3. alerts_rules_and_order 4. summary_counts 5. browser_fits_1080p_no_network 6. fresh_run_1 7. fresh_run_2

Setup notes: Use Case Scout batch 13 case n01-tv-home-screen (offline, synthetic fixture). Prepared with batch13/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 0.34 (8 cores). home.html rendered offline with google-chrome headless (CDP pipe, network blocked): PNG 1920x1080 at ~3 s after load; MP4 ~6 s captured at 1920x1080, H.264 1280x720. Claude Code's own reported cost for claude-haiku-5.5: $0.022289. 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
$0.02
Wall time
2m
Human step-ins
0
Turns
11
Model usage
ModelInput tokOutput tokCost
AClaude Haiku 5.5anthropic/claude-haiku-5.5effort default (no --effort flag) · thinking on386,31022,653$0.02

Reproduce this run

  1. Use the use case prompt on the use case page.
  2. 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.
  3. It counts as a success only if it meets every success criterion: 1. html_static_rooms_in_order 2. device_tiles_status_state 3. alerts_rules_and_order 4. summary_counts 5. browser_fits_1080p_no_network 6. fresh_run_1 7. fresh_run_2
  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": 105970,
  "stop_reason": "end_turn",
  "session_id": "[REDACTED]",
  "total_cost_usd": 0.022289039999999996,
  "usage": {
   "input_tokens": 20,
   "cache_creation_input_tokens": 37356,
   "cache_read_input_tokens": 348934,
   "output_tokens": 22653,
   "output_tokens_details": {
    "thinking_tokens": 13774
   },
   "server_tool_use": {
    "web_search_requests": 0,
    "web_fetch_requests": 0
   },
   "service_tier": "standard",
   "cache_creation": {
    "ephemeral_1h_input_tokens": 37356,
    "ephemeral_5m_input_tokens": 0
   },
   "inference_geo": "global",
   "iterations": [
    {
     "input_tokens": 2,
     "output_tokens": 865,
     "cache_read_input_tokens": 48626,
     "cache_creation_input_tokens": 1269,
     "cache_creation": {
      "ephemeral_5m_input_tokens": 0,
      "ephemeral_1h_input_tokens": 1269
     },
     "type": "message"
    }
   ],
   "speed": "standard",
   "fallback_credit": null
  },
  "modelUsage": {
   "claude-haiku-5-5": {
    "inputTokens": 20,
    "outputTokens": 22653,
    "cacheReadInputTokens": 348934,
    "cacheCreationInputTokens": 37356,
    "webSearchRequests": 0,
    "costUSD": 0.022289039999999996,
    "contextWindow": 1000000,
    "maxOutputTokens": 128000,
    "thinkingTokens": 13774,
    "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": 11,
  "subtype": "success",
  "api_error_status": null,
  "result": "`solve.py` is written and runs in well under the 30-second limit. It produces `answer.json` and `home.html` for the given `home.json`. I checked the answer values by hand against the rules and checked the page in headless Chrome at 1920x1080.\n\n**Results for this home**\n- Summary: 1 light on, 21 online, 5 offline, 10 alerts, average temperature 68.9 \u00b0F.\n- Alerts in order: Kitchen leak (Kitchen Leak 8), Bath #2 leak (Bath #2 Leak 21), five offline devices, two temp-drift thermostats (Garage 3 and 5), and Kid's Room Lock 2 for low battery.\n- Browser: no page overflow, all 26 tiles are 135x92 and on screen, and `#alerts` fits. The static checks pass: no script, image, link, iframe, external URL or `on*` handler.\n\n**Interpretation choices**\n- **`door_open`:** the data has no `open_since` field, so this alert can't be evaluated. I read it from the device or its `state` if present, and otherwise never raise it. No contact is open in this home, so it has no effect here.\n- **`lights_on`:** counts only devices of type `light`. The plug is not counted.\n- **Rounding:** the average is the mean of the already-rounded 1-decimal values, rounded half-up. Negative temperatures would round away from zero.\n- **Output location:** `home.html` is written next to `answer.json`, which is the current directory in the given invocation.\n\n**Limits**\n- I only browser-tested this home. The fresh homes used in scoring are untested.\n- The alerts panel holds about 24 alerts in two columns. Beyond that, the extra alerts would be clipped and not visible.\n\nFiles are in `./usecase-runner/runs/n01-claude-haiku/work/`: `solve.py`, `answer.json`, `home.html`.",
  "ttft_ms": 1973,
  "type": "result",
  "duration_ms": 108479,
  "uuid": "[REDACTED]",
  "ttft_stream_ms": 1733,
  "time_to_request_ms": 266,
  "first_content_frame_ms": 1733,
  "queued_turn_count": 0,
  "result_index": 0
 }
]

Get new runs and ranking changes weekly