EVIDENCE AND EVALUATION / Undominated.ai
Undominated · Reasoning token audit
Audit reasoning token expansion ratios, budget caps, separate billing vs output billing, and hidden multiplier surcharges across reasoning models.
“Audit reasoning token expansion ratios, budget caps, separate billing vs output billing, and hidden multiplier surcharges across reasoning models.”
01 / THE REASONING
Why this made the selection.
- Pairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.
- Created for recurring evidence and comparison failures: missing facts remain unknown and every conclusion keeps its scope.
- It names a concrete failure mode: Validate budget constraints: check if reasoning token count exceeds maximum declared budget (`max_thinking_tokens` or `reasoning_effort`).
02 / THE REVIEW RECORD
What we actually inspected.
Source review has boundaries.
A clear record is more useful than a “safe” badge.
Material inspected
- skills/undominated-reasoning-token-audit/SKILL.md
- skills/undominated-reasoning-token-audit/scripts/check.py
- skills/undominated-reasoning-token-audit/examples/synthetic.json
Our findings
- First-party resource authored by Undominated.ai; this is our own verification record, not an independent endorsement.
- The download preserves the complete source and supporting files. Its manifest hashes identify the exact bytes.
- One input rule the file states is: `reasoningPerMillion`: non-negative decimal string or integer (required if `billingModel` is `separate_reasoning`).
- One condition that blocks a pass is: Effective output rate {effective_output_rate} exceeds maxEffectiveRatePerMillion {max_rate}.
Not established by this review
- Model compliance with these instructions on arbitrary tasks
- Native integration in every agent host
- Public GitHub source identity
- Remote installation through the standard Skills CLI
- The new npm resource CLI installation
The review applies to the material and revision named here. A newer upstream release can change its behavior.
03 / PUT IT TO WORK
Add a skill to your workflow.
- Download the complete bundle from https://undominated.ai/resources/skills/undominated-reasoning-token-audit/bundle.zip and extract it into a new directory.
- Place the extracted directory at .agents/skills/undominated-reasoning-token-audit/ or your host's documented skills directory. Preserve SKILL.md, scripts, examples and licence files together.
- Run the documented synthetic example from the skill directory with Python 3.10 or later before using your own evidence.
Before you start
- An Agent Skills compatible host
- Python 3.10 or later for the optional checker
THE COMPLETE REVIEWED DEFINITION
Read it before you reuse it.
Original source bytes, with attribution.
Review the host-specific setup notes above.
# Reasoning Token Audit
Audit reasoning token expansion ratios, budget caps, separate billing vs output billing, and hidden multiplier surcharges across reasoning models.
## Workflow
1. Inspect reported token usage: prompt tokens, visible completion tokens, and reasoning/thinking tokens (`completion_tokens_details.reasoning_tokens` or equivalent).
2. Validate provider billing model:
- Separate reasoning rate vs bundled output rate.
- Surcharge ratio $\text{Ratio} = \text{Reasoning Tokens} / \text{Completion Tokens}$.
- True effective output cost per million tokens:
$$\text{Effective Rate} = \frac{(\text{Reasoning Tokens} \times \text{Rate}_{\text{reasoning}}) + (\text{Visible Tokens} \times \text{Rate}_{\text{output}})}{\text{Visible Tokens}}$$
3. Validate budget constraints: check if reasoning token count exceeds maximum declared budget (`max_thinking_tokens` or `reasoning_effort`).
4. Output structured audit verification receipt.
## Synthetic Input Format
JSON object with:
- `schemaVersion`: integer (1)
- `model`: model identifier string
- `provider`: provider identifier string
- `billingModel`: `"bundled_output"` | `"separate_reasoning"`
- `rates`:
- `promptPerMillion`: non-negative decimal string or integer
- `outputPerMillion`: non-negative decimal string or integer
- `reasoningPerMillion`: non-negative decimal string or integer (required if `billingModel` is `separate_reasoning`)
- `trajectory`:
- `promptTokens`: non-negative integer
- `visibleTokens`: positive integer
- `reasoningTokens`: non-negative integer
- `budgetCap`: optional positive integer
- `assertions`:
- `maxExpansionRatio`: optional decimal string (upper bound on reasoning / visible tokens)
- `maxEffectiveRatePerMillion`: optional decimal string
## Exit Codes
- `0`: Valid input and all assertions passed.
- `1`: Assertion failed (e.g., expansion ratio exceeded, budget cap breached, effective cost ceiling passed).
- `2`: Malformed input or path traversal violation.
## Verification
```sh
python3 scripts/check.py examples/synthetic.json
```
The download contains SKILL.md. Extract the whole bundle; the supporting files are required. Inspect the included MANIFEST.json for file hashes.
By Undominated.ai. Exact upstream source ↗ · Licence · Attribution
SHA-256 060fbda03cff5d7023838034c851284cedc34c14055f559a3c230a91d1f136fa
Read the applicable licence
MIT License Copyright (c) 2026 Undominated.ai Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
04 / FOLLOW THE EVIDENCE
The source trail.
Our notes are separate from the original resource.
Check upstream before adopting a new version.
https://undominated.ai/resources/skills/undominated-reasoning-token-audit/definition.md
Supports: Workflow and input requirements, Stated limitations
- MIT licence ↗Checked
https://undominated.ai/resources/skills/undominated-reasoning-token-audit/LICENSE.txt
Supports: Redistribution terms and attribution
- Complete source bundle ↗Checked
https://undominated.ai/resources/skills/undominated-reasoning-token-audit/bundle.zip
Supports: Full local source, supporting files and integrity manifest