Claude Haiku 5.5 (effort default (no --effort flag) · thinking on) on Finish a JSON Schema subset validator (types, JSON equality, pointer-escaped paths, $ref recursion, combinators)
AClaude Haiku 5.5default · thinking on · Claude Code 2.1.299 (headless -p) · cloud · 2026-10-19 · by @bobsiopener
- Estimated cost
- $0.02
- Wall time
- 1m
- Human step-ins
- 0
- Outcome
- Pass
Compare with other models on this task →
Summary
Exact prompt
# Task: finish the JSON Schema validator `validator.py` contains `validate(schema, instance)`, a validator for a subset of JSON Schema used by our config tooling. It handles only a few keywords and gets several of those wrong. Make it implement `SCHEMA_RULES.md` exactly (rewrite freely). Keep the function name, return format and the `SchemaError` class. Python standard library only; do not install or vendor `jsonschema`. No network. The grader runs a hidden unittest suite that calls `validate` on many schema/instance pairs and compares the returned errors (as a multiset of `(path, keyword)` pairs; order does not matter, duplicates do).
Boolean-vs-number equality, integer-valued floats, type-failure short-circuit, ~0/~1 path escaping, per-member additionalProperties errors, single anyOf/oneOf errors and recursive $ref are each pinned by exact (path, keyword) multisets. 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. All 11 hidden unittest cases pass (boolean schemas/unknown keywords, types, type short-circuit, equality, numeric and string keywords, objects and escaped paths, arrays and uniqueItems, nested paths, $ref and recursion, combinators).
Setup notes: Use Case Scout batch 7 case g10-json-schema-validator (offline fixture). Prepared with batch7/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: $0.016829. Score: 11/11. 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
- $0.02
- Wall time
- 1m
- Human step-ins
- 0
- Turns
- 8
| Model | Input tok | Output tok | Cost |
|---|---|---|---|
| AClaude Haiku 5.5anthropic/claude-haiku-5.5effort default (no --effort flag) · thinking on | 250,616 | 18,382 | $0.02 |
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. All 11 hidden unittest cases pass (boolean schemas/unknown keywords, types, type short-circuit, equality, numeric and string keywords, objects and escaped paths, arrays and uniqueItems, nested paths, $ref and recursion, combinators).
- 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?
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": 79354,
"stop_reason": "end_turn",
"session_id": "[REDACTED]",
"total_cost_usd": 0.016829359999999998,
"usage": {
"input_tokens": 16,
"cache_creation_input_tokens": 27004,
"cache_read_input_tokens": 223596,
"output_tokens": 18382,
"output_tokens_details": {
"thinking_tokens": 11191
},
"server_tool_use": {
"web_search_requests": 0,
"web_fetch_requests": 0
},
"service_tier": "standard",
"cache_creation": {
"ephemeral_1h_input_tokens": 27004,
"ephemeral_5m_input_tokens": 0
},
"inference_geo": "global",
"iterations": [
{
"input_tokens": 2,
"output_tokens": 732,
"cache_read_input_tokens": 39126,
"cache_creation_input_tokens": 417,
"cache_creation": {
"ephemeral_5m_input_tokens": 0,
"ephemeral_1h_input_tokens": 417
},
"type": "message"
}
],
"speed": "standard",
"fallback_credit": null
},
"modelUsage": {
"claude-haiku-5-5": {
"inputTokens": 16,
"outputTokens": 18382,
"cacheReadInputTokens": 223596,
"cacheCreationInputTokens": 27004,
"webSearchRequests": 0,
"costUSD": 0.016829359999999998,
"contextWindow": 1000000,
"maxOutputTokens": 128000,
"thinkingTokens": 11191,
"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": 8,
"subtype": "success",
"api_error_status": null,
"result": "I rewrote `validator.py` so it follows `SCHEMA_RULES.md` section by section. The public signature is still `validate(schema, instance)`, errors are still `{\"path\", \"keyword\"}` dicts, and `SchemaError` is unchanged. It uses only the standard library.\n\nI checked it with about 30 hand-written schema/instance cases, run from a scratch directory outside the work tree. All pass, including the trickier rules:\n- `1` vs `true` and `1` vs `1.0` in `enum`, `const` and `uniqueItems`.\n- `integer` accepts `2.0` but rejects `2.5` and `true`.\n- `type` failure reports only the `type` error and skips the other keywords in that schema object.\n- `required` errors sit at the object's path; property paths escape `~` and `/`.\n- `anyOf`, `oneOf` and `not` each give one error with no sub-errors.\n- `$ref` to `#` and to `#/$defs/...` works, including recursive schemas.\n\nThe grader's hidden suite was not available, so this is my own check, not the grader's.\n\nJudgment calls the spec left open:\n- **`$ref`** accepts any JSON pointer from the root, not only `#/$defs/<name>`. An unresolvable one raises `SchemaError`, as does a `$ref` that cycles back to the same instance without consuming any of it (for example `{\"$ref\": \"#\"}` on its own).\n- **Malformed schemas** raise `SchemaError`. This covers an unknown type name, a non-list `enum`/`required`/`allOf`, a bad regex in `pattern`, a non-number `minimum`, and a non-schema `items`. Keyword values are checked only when that keyword is evaluated, so a malformed keyword on an instance it never applies to passes silently.\n- **`exclusiveMinimum` / `exclusiveMaximum`** are treated as numbers only, per the spec, so the draft-4 boolean form raises `SchemaError`.\n- **`required`** emits one error per list entry, so a duplicated missing name produces two errors.\n- **Empty `anyOf` and `oneOf`** produce an error, since no subschema is valid and the count is not exactly 1.",
"ttft_ms": 1262,
"type": "result",
"duration_ms": 79883,
"uuid": "[REDACTED]",
"ttft_stream_ms": 1074,
"time_to_request_ms": 252,
"first_content_frame_ms": 1075,
"queued_turn_count": 0,
"result_index": 0
}
]