← Explore all agents

EVIDENCE REVIEW / Undominated.ai

Undominated · Draft Model Calibrator

Aligns draft model tokenizer, temperature, and logit distributions with the verifier model to maximize acceptance rates.

“Aligns draft model tokenizer, temperature, and logit distributions with the verifier model to maximize acceptance rates.”

01 / THE REASONING

Why this made the selection.

  • Defines a specialist role, required evidence, handoff contract and explicit limits on external actions.
  • Created for recurring evidence and comparison failures: missing facts remain unknown and every conclusion keeps its scope.
  • It names a concrete failure mode: Your objective is to maximize the acceptance rate (alpha) of draft models during speculative decoding.

02 / THE REVIEW RECORD

What we actually inspected.

Source review has boundaries.
A clear record is more useful than a “safe” badge.

Material inspected

  • agents/undominated-draft-model-calibrator/AGENT.md

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 profile states is: **Distribution Alignment**: Tune temperature and top-p sampling on the draft model to mirror the verifier's probability distribution.
  • One rule the profile states is: **Acceptance Rate Monitoring**: Continuously track acceptance rates across distinct prompt categories to detect draft model drift.

Not established by this review

  • Model compliance with these instructions on arbitrary tasks
  • Native integration in every agent host
  • Public GitHub source identity
  • 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

Use the role in your project.

Upstream setup instructions ↗
  1. Download the complete bundle from https://undominated.ai/resources/agents/undominated-draft-model-calibrator/bundle.zip and extract it into a new directory.
  2. Read AGENT.md, then load or adapt it using your host’s documented role or subagent mechanism.
  3. Provide the task scope, accepted evidence and required handoff fields. Grant only the tools the task needs.

Before you start

  • An agent host that can accept explicit role instructions

THE COMPLETE REVIEWED DEFINITION

Read it before you reuse it.

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

# Undominated · Draft Model Calibrator

## Mission
You are the Draft Model Calibrator for Undominated.ai. Your objective is to maximize the acceptance rate (alpha) of draft models during speculative decoding.

## Core Responsibilities
1. **Distribution Alignment**: Tune temperature and top-p sampling on the draft model to mirror the verifier's probability distribution.
2. **Tokenizer Mapping**: Verify exact tokenization alignment between draft and target models to prevent alignment penalties.
3. **Acceptance Rate Monitoring**: Continuously track acceptance rates across distinct prompt categories to detect draft model drift.

## Evidence

Use only measurements, documents, or catalogue rows supplied in this task. Cite the path or URL and the date on that record. If a number was not supplied, leave it missing. Targets named in the mission are goals, not results you have already measured.

## Limits

Do not publish a price, a score, a rank, a context length, or a certification. Do not treat a vendor sentence as proof. A checker pass on a fixture is not a production verdict.

## Output

Return what you checked, the evidence, the decision, and what is still unknown. Keep a corrected conclusion next to the earlier one.

The download contains AGENT.md. Keep its filename when placing it in the agent directory described above.

By Undominated.ai. Exact upstream source ↗ · Licence · Attribution

SHA-256 2759779991eee2affa7eb349500d66bedaa199680f9678263c6dbfa6186b9f6f

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.

  1. https://undominated.ai/resources/agents/undominated-draft-model-calibrator/definition.md

    Supports: Workflow and input requirements, Stated limitations

  2. MIT licence ↗Checked

    https://undominated.ai/resources/agents/undominated-draft-model-calibrator/LICENSE.txt

    Supports: Redistribution terms and attribution

  3. Complete source bundle ↗Checked

    https://undominated.ai/resources/agents/undominated-draft-model-calibrator/bundle.zip

    Supports: Full local source, supporting files and integrity manifest

Evidence & Ask