--- title: "Undominated · Test-time search budget audit: review, setup & limitations · Undominated.ai" canonical: https://undominated.ai/skills/undominated-test-time-search-budget-audit/ description: "Audits test-time compute search budgets to ensure marginal accuracy gains justify token expenditure." --- # Undominated · Test-time search budget audit: review, setup & limitations · Undominated.ai > Audits test-time compute search budgets to ensure marginal accuracy gains justify token expenditure. [← Explore all skills](/skills/) EVIDENCE AND EVALUATION / Undominated.ai # Undominated · Test-time search budget audit Audits test-time compute search budgets to ensure marginal accuracy gains justify token expenditure. [See setup guidance ↓](#setup)[Original source ↗](https://undominated.ai/resources/skills/undominated-test-time-search-budget-audit/definition.md) SOURCE REVIEW Reviewed 2026-10-08 Evidence 3 linked sources Publisher Undominated.ai Licence [MIT ↗](https://undominated.ai/resources/skills/undominated-test-time-search-budget-audit/LICENSE.txt) Revision 1bd9d78f15ac [Read what was—and wasn’t—checked ↓](#review) “Audits test-time compute search budgets to ensure marginal accuracy gains justify token expenditure.” [Undominated.ai · upstream description ↗](https://undominated.ai/resources/skills/undominated-test-time-search-budget-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. ### A good fit for - Audit Best-of-N or search-tree reasoning systems to detect diminishing returns in token spending. ### Weigh up before choosing - Evaluates observed pass@k curves against token budgets; does not generate reasoning trees. - 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. - A pass checks the supplied JSON only. Recomputed fields are listed in the skill. It does not certify compliance, safety, or a market price. 02 / THE REVIEW RECORD ## What we actually inspected. Source review has boundaries. A clear record is more useful than a “safe” badge. ### Material inspected - Original definition and MIT licence - Bundled checker source, input contract and synthetic examples ### 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. - The checker type-checks every declared field, recomputes only identities that match the bundled synthetic example, and does not emit a compliance or release flag. ### 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-test-time-search-budget-audit/definition.md) - Download the complete bundle from https://undominated.ai/resources/skills/undominated-test-time-search-budget-audit/bundle.zip and extract it into a new directory. - Place the extracted directory at .agents/skills/undominated-test-time-search-budget-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-test-time-search-budget-audit/bundle.zip)[Raw Markdown ↗](/resources/skills/undominated-test-time-search-budget-audit/definition.md) --- name: undominated-test-time-search-budget-audit description: "Audits test-time compute search budgets to ensure marginal accuracy gains justify token expenditure. Recomputed fields must match the other inputs. A pass is not a compliance, safety, or price verdict." license: MIT metadata: author: Undominated.ai version: "1.1.0" --- # Test-time search budget audit ## Description Audits test-time compute search budgets to ensure marginal accuracy gains justify token expenditure. ## Category `evaluation` ## Tags `evaluation`, `reasoning`, `test-time-compute`, `best-of-n`, `roi` ## When to Use Audit Best-of-N or search-tree reasoning systems to detect diminishing returns in token spending. ## Operational Limitations Evaluates observed pass@k curves against token budgets; does not generate reasoning trees. ## Verification Script Run the deterministic verification checker: ```bash python3 scripts/check.py examples/synthetic.json ``` Exit codes: - `0`: Check passed and thresholds satisfied. - `1`: Threshold breach or operational violation detected. - `2`: Schema error or invalid usage. ## What the checker recomputes The original threshold checks still apply: - `marginal_gain_per_k_tokens` must not be < `min_roi_threshold`. Every other declared field is type-checked. A negative value is rejected except for deltas, drifts, and gains. A type check does not, by itself, prove the measurement. `population` and `observedAt` are required. The bundled examples are synthetic fixtures, not market measurements. ```bash python3 scripts/check.py examples/synthetic.json python3 scripts/check.py examples/contradiction.json python3 scripts/check.py examples/broken.json ``` Exit 0 is a pass, exit 1 is a review, and exit 2 is invalid input. A pass does not certify a provider, a regulation, a release, or a price. 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-test-time-search-budget-audit/definition.md) · [Licence](/resources/skills/undominated-test-time-search-budget-audit/LICENSE.txt) · [Attribution](/resources/skills/undominated-test-time-search-budget-audit/ATTRIBUTION.txt) SHA-256 1bd9d78f15ac592ade1f2818c1ec93b543db0a8fde9dd048e6640c4bf06eb708 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-test-time-search-budget-audit/definition.md) Checked 2026-10-07 https://undominated.ai/resources/skills/undominated-test-time-search-budget-audit/definition.md Supports: Workflow and input requirements, Stated limitations - [MIT licence ↗](https://undominated.ai/resources/skills/undominated-test-time-search-budget-audit/LICENSE.txt) Checked 2026-10-07 https://undominated.ai/resources/skills/undominated-test-time-search-budget-audit/LICENSE.txt Supports: Redistribution terms and attribution - [Complete source bundle ↗](https://undominated.ai/resources/skills/undominated-test-time-search-budget-audit/bundle.zip) Checked 2026-10-07 https://undominated.ai/resources/skills/undominated-test-time-search-budget-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 · Test-Time Compute ROI Audit ↗ Calculates cost-per-accuracy return curves for Best-of-N sampling, search depth scaling, and verifier reranking vs base model upgrading.](/skills/undominated-test-time-compute-roi-audit/)[### Undominated · Process reward model tax audit ↗ Audits Process Reward Model (PRM) step-level verification token overhead versus outcome accuracy gains.](/skills/undominated-process-reward-model-tax-audit/)[### Undominated · Reasoning trace redundancy audit ↗ Audits reasoning trace self-correction loops and verbosity redundancy to prune repetitive reflection tokens.](/skills/undominated-reasoning-trace-redundancy-audit/) [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/)