Claude Haiku 5.5 (effort default (no --effort flag) · thinking on) on Progressive tax on fictional brackets, and pre-tax vs after-tax savings comparison

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

Trust 80%Fully autonomoussuccess · first tryCompare

Estimated cost: $0.0054
Wall time: 18s
Human step-ins: 0
Outcome: Pass

Compare with other models on this task →

Exact prompt

# Progressive tax on fictional brackets, and pre-tax ("traditional") vs after-tax ("Roth") savings `input.json` contains a **fictional** tax system (brackets and standard deductions per filing status), a list of incomes to tabulate, and savings scenarios. Write `solve.py` (Python 3 standard library only) so that `python solve.py input.json answer.json` computes the answers below. Use only the numbers in the input. We re-run it on fresh inputs, so do not hard-code anything. ## Rules 1. Taxable income = max(0, gross income - `standard_deduction[filing]`). 2. Brackets for a filing status are a list of `[lower_bound, rate]` sorted by bound. Rate j applies to the part of taxable income above `lower_bound_j` and up to `lower_bound_{j+1}` (the last bracket has no upper bound). Tax = sum over brackets. 3. Marginal rate = the rate of the last bracket whose lower bound is strictly below taxable income; if taxable income is 0, the first bracket's rate. 4. `tax_table`: for each `[filing, income]` in input order output `{"filing", "income", "tax", "effective_rate", "marginal_rate"}`; effective_rate = tax / income (0 when income is 0), 6 decimals. 5. Scenarios (each has `name`, `filing`, `wages`, `contribution`, `years`, `growth`, `retire_filing`, `retire_other_income`): - `tax_now_traditional` = tax on (wages - contribution); `tax_now_roth` = tax on wages; `tax_saved_now` = their difference; `marginal_rate_now` = marginal rate (rule 3) at the taxable income of (wages - contribution). - `traditional_balance` F = contribution * (1 + growth) ** years. `withdrawal_tax` = tax(retire_other_income + F) - tax(retire_other_income), both with `retire_filing`, taken as one withdrawal in a single year. `traditional_after_tax` = F - withdrawal_tax. - Same out-of-pocket cost: the after-tax account receives (contribution - tax_saved_now), so `roth_after_tax` = (contribution - tax_saved_now) * (1 + growth) ** years. - `winner` = "traditional" or "roth", whichever after-tax value is larger, or "tie" if they differ by less than 0.005. 6. Round money to 2 decimals and rates to 6 only when writing (bracket rates are written as given). ## Output `answer.json`: `{"tax_table": [...], "scenarios": [{"name", "tax_now_traditional", "tax_now_roth", "tax_saved_now", "marginal_rate_now", "traditional_balance", "withdrawal_tax", "traditional_after_tax", "roth_after_tax", "winner"}, ...]}` in input order. ## Scoring Hidden checks (partial credit = fraction passed; all must pass), money within 0.011, rates within 0.000001, winners exact: `tax_table_amounts`, `effective_and_marginal`, `scenario_tax_now`, `scenario_future_values`, `scenario_winners` on the given input; `fresh_run_1`, `fresh_run_2`: the whole answer on fresh inputs (tolerance 0.011). Each run has a 60-second limit.

Incomes placed exactly on bracket boundaries test the strict marginal-rate rule; the equal out-of-pocket comparison and the lump-withdrawal tax at retirement decide winners. 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. tax_table_amounts 2. effective_and_marginal 3. scenario_tax_now 4. scenario_future_values 5. scenario_winners 6. fresh_run_1 7. fresh_run_2

Setup notes: Use Case Scout batch 9 case j05-tax-brackets-roth-vs-traditional (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. Claude Code's own reported cost for claude-haiku-5.5: $0.005416. 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.

Metrics

Estimated cost: $0.0054
Wall time: 18s
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	114,877	3,171	$0.0054

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. tax_table_amounts 2. effective_and_marginal 3. scenario_tax_now 4. scenario_future_values 5. scenario_winners 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": 16556,
  "stop_reason": "end_turn",
  "session_id": "[REDACTED]",
  "total_cost_usd": 0.0054155,
  "usage": {
   "input_tokens": 10,
   "cache_creation_input_tokens": 14107,
   "cache_read_input_tokens": 100760,
   "output_tokens": 3171,
   "output_tokens_details": {
    "thinking_tokens": 966
   },
   "server_tool_use": {
    "web_search_requests": 0,
    "web_fetch_requests": 0
   },
   "service_tier": "standard",
   "cache_creation": {
    "ephemeral_1h_input_tokens": 14107,
    "ephemeral_5m_input_tokens": 0
   },
   "inference_geo": "global",
   "iterations": [
    {
     "input_tokens": 2,
     "output_tokens": 719,
     "cache_read_input_tokens": 24693,
     "cache_creation_input_tokens": 1953,
     "cache_creation": {
      "ephemeral_5m_input_tokens": 0,
      "ephemeral_1h_input_tokens": 1953
     },
     "type": "message"
    }
   ],
   "speed": "standard",
   "fallback_credit": null
  },
  "modelUsage": {
   "claude-haiku-5-5": {
    "inputTokens": 10,
    "outputTokens": 3171,
    "cacheReadInputTokens": 100760,
    "cacheCreationInputTokens": 14107,
    "webSearchRequests": 0,
    "costUSD": 0.0054155,
    "contextWindow": 1000000,
    "maxOutputTokens": 128000,
    "thinkingTokens": 966,
    "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` and ran it on `input.json`. It produces `answer.json` with the `tax_table` and six scenarios in input order. The script uses only the standard library and reads everything from the input, so nothing is hard-coded.\n\nI checked one tax figure by hand: the joint $683,000 row (taxable $651,800) gives $172,707.60, matching the script. I did not check the scenario figures by hand, and I did not run it on fresh inputs.\n\nChoices I made where the spec was silent:\n- **Internal precision:** the script keeps full precision and rounds only when writing. `tax_saved_now` feeds the Roth calculation unrounded.\n- **Winner test:** it compares the unrounded after-tax values against the 0.005 tie threshold.\n- **Marginal rate:** taxable income exactly on a bracket bound takes the lower bracket's rate, since the bound is not strictly below.\n- **Bracket order:** brackets are sorted by lower bound before use, in case a fresh input isn't sorted.\n\nThe folder isn't a git repository, so I did not commit anything.",
  "ttft_ms": 1447,
  "type": "result",
  "duration_ms": 16957,
  "uuid": "[REDACTED]",
  "ttft_stream_ms": 1253,
  "time_to_request_ms": 269,
  "first_content_frame_ms": 1254,
  "queued_turn_count": 0,
  "result_index": 0
 }
]