---
title: "Undominated · Reasoning token audit: review, setup & limitations · Undominated.ai"
canonical: https://undominated.ai/skills/undominated-reasoning-token-audit/
description: "Audit reasoning token expansion ratios, budget caps, separate billing vs output billing, and hidden multiplier surcharges across reasoning models."
---

# Undominated · Reasoning token audit: review, setup & limitations · Undominated.ai

> Audit reasoning token expansion ratios, budget caps, separate billing vs output billing, and hidden multiplier surcharges across reasoning models.

[← Explore all skills](/skills/)

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.

 [See setup guidance ↓](#setup)[Original source ↗](https://undominated.ai/resources/skills/undominated-reasoning-token-audit/definition.md)

SOURCE REVIEW

 Reviewed 2026-10-14
 Evidence 3 linked sources
 Publisher Undominated.ai
 Licence [MIT ↗](https://undominated.ai/resources/skills/undominated-reasoning-token-audit/LICENSE.txt)
 Revision 060fbda03cff
 [Read what was—and wasn’t—checked ↓](#review)

“Audit reasoning token expansion ratios, budget caps, separate billing vs output billing, and hidden multiplier surcharges across reasoning models.”

 [Undominated.ai · upstream description ↗](https://undominated.ai/resources/skills/undominated-reasoning-token-audit/definition.md) Our analysis follows below.

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`).

### A good fit for

 - Audit reasoning token usage, thinking budget caps, and effective output rates for reasoning-capable models.
- Audit reasoning token expansion ratios, budget caps, separate billing vs output billing.

### Weigh up before choosing

 - Deterministic local checks over supplied token counts; does not evaluate model reasoning depth or solution correctness.
- Fixture checks establish the documented local behaviour, not performance on an unrestricted set of real tasks.
- Public GitHub source availability is not yet verified; the complete download contains the reviewed source.
- A new npm install path is not yet verified. Use the complete source download.
- `2`: Malformed input or path traversal violation.

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.

[Upstream setup instructions ↗](https://undominated.ai/resources/skills/undominated-reasoning-token-audit/definition.md)
 - 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

### Compatibility

Agent Skills compatible hosts · Python 3.10+

### Implementation

Python · Markdown

### Local, explicit inputs

 - read:user-selected-local-file

### Cost model

MIT source at no charge. Your agent host or model provider may charge for use; the included offline checks require no paid API.

THE COMPLETE REVIEWED DEFINITION

## Read it before you reuse it.

Original source bytes, with attribution. Review the host-specific setup notes above.

 Copy definition ↗ [Download complete resource + licence ↗](/resources/skills/undominated-reasoning-token-audit/bundle.zip)[Raw Markdown ↗](/resources/skills/undominated-reasoning-token-audit/definition.md)
 # 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 ↗](https://undominated.ai/resources/skills/undominated-reasoning-token-audit/definition.md) · [Licence](/resources/skills/undominated-reasoning-token-audit/LICENSE.txt) · [Attribution](/resources/skills/undominated-reasoning-token-audit/ATTRIBUTION.txt)

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.

 - [Original Undominated definition ↗](https://undominated.ai/resources/skills/undominated-reasoning-token-audit/definition.md) Checked 2026-10-07 https://undominated.ai/resources/skills/undominated-reasoning-token-audit/definition.md Supports: Workflow and input requirements, Stated limitations
- [MIT licence ↗](https://undominated.ai/resources/skills/undominated-reasoning-token-audit/LICENSE.txt) Checked 2026-10-07 https://undominated.ai/resources/skills/undominated-reasoning-token-audit/LICENSE.txt Supports: Redistribution terms and attribution
- [Complete source bundle ↗](https://undominated.ai/resources/skills/undominated-reasoning-token-audit/bundle.zip) Checked 2026-10-07 https://undominated.ai/resources/skills/undominated-reasoning-token-audit/bundle.zip Supports: Full local source, supporting files and integrity manifest

KEEP COMPARING

## Other approaches to consider.

Related by category or shared topics. These are alternatives to inspect, not a measured quality order.

 [### Undominated · Precision prompt cost estimator ↗ Calculate inference costs across prompt tokens, completion tokens, cached prompt tokens (read/write), reasoning tokens, and multi-turn trajectories using explicit provider pricing ladders.](/skills/undominated-prompt-cost-estimator/)[### Undominated · A marginal block is not priced ↗ Refuse a marginal-block bill. A whole request uses the rate of the rung that contains its input length. This checker does not price either mode.](/skills/undominated-marginal-not-whole/)[### Undominated · A zero base is not defined ↗ Refuse a multiple whose base is zero. The multiple is not defined. This checker does not emit a quotient.](/skills/undominated-zero-base-undefined/)

 [AI Tools ↗](/tools/)[Skills ↗](/skills/)[Agents ↗](/agents/)[MCP Servers ↗](/mcp-servers/)[Workflows ↗](/workflows/)

## Continue your investigation

 - [Choose a toolkit](/tools/)
- [Inspect execution hosts](/agents/)
- [Check tool access](/mcp-servers/)
- [Check project dependencies](/check-repo/)
