llm-calibration-logprobs SKILL.md skillA
llm-calibration-logprobs from scdenney/open-science-skills is an agent skill: a SKILL.md folder an AI agent loads when a task calls for it.
mdr add scdenney/open-science-skills/llm-calibration-logprobs@git:20261008.697c89fPins git:20261008.697c89f in mdr.lock. Pin the fingerprint instead.
- Audit grade
- A, 17 of 17 checks passed
- Kind
- SKILL.md skill
- Repository
- scdenney/open-science-skills
- Path
codex/llm-calibration-logprobs/SKILL.md- Versions recorded
- 5
- Latest version
- git:20261008.697c89f, committed Oct 8, 2026
- First version recorded
- Jul 2, 2026
- Size
- 11,464 bytes
- Repository stars
- 69
- License
- a license file GitHub does not identify
Reads a model's own uncertainty from token log-probabilities: collecting and aggregating label logprobs, confidence tiers and margins for triage, calibration against human labels (ECE, Brier, reliability diagrams), downstream use, and reproducibility. Use for per-item classifier confidence or routing low-confidence cases to human review; codebook and validation design go to text-classification.
Collected from public GitHub. Every version is saved exactly as it was and never changes. Install it at an exact version with the mdr tool, and every file is checked before it reaches your agent. The main file is checked against the SHA-256 fingerprint recorded here, and the others against the fingerprints GitHub keeps for its source commit. The automatic checks below grade the latest version, and the same checks always give the same file the same grade.
What the file says
# Reading a Model's Own Uncertainty from Token Log-Probabilities ## Instructions This skill covers **within-model** confidence: how sure one model is about each decision, read from the token log-probabilities it emits, and whether that confidence is *calibrated* against ground truth. Agreement across several independent models is a different kind of evidence — when a pipeline uses both, report both. Codebook design and human-validation statistics (κ, F1) belong to `$text-classification`; cross-reference rather than re-derive them. ### 1. What a Logprob Can and Cannot Tell You - Use logprobs when you need a per-item, continuous confidence signal. It comes free with the forward pass that produced the label and answers "how sure was the model about *this* classification?", which a hard 0/1 label cannot. - A logprob is the model's subjective probability under its own distribution, **not a probability of correctness**. A model can assign 0.95 to labels that are right 70% of the time. Treat it as a claim about confidence that calibration (§4) tests. …
Read the whole file at its exact version.
How to install
mdr add scdenney/open-science-skills/llm-calibration-logprobs@sha256:2f581002e1af295eThe line at the top of this page pins the label git:20261008.697c89f, which is what most people want. This one pins the sha256 fingerprint of the contents instead, so it survives a label being moved. Either way the exact version is written to mdr.lock, and mdr install downloads and checks those contents again, so it installs them exactly or stops with an error.
[](https://markdownregistry.com/a/art_p3cil5c5uvveqmos)
3 badge image fetches in 30 days. What the badge shows.
Versions
The registry has recorded 5 versions of this file, between Jul 2, 2026 and Oct 8, 2026, about one every 24 days.
| version | committed | commit | size | audit | |
|---|---|---|---|---|---|
| git:20261008.697c89f latest | 2026-10-08 | 697c89f | 11,464 B | A | view · diff |
| git:20260905.39e8276 | 2026-09-05 | 39e8276 | 19,621 B | A | view · diff |
| git:20260902.da5a263 | 2026-09-02 | da5a263 | 19,154 B | A | view · diff |
| git:20260808.c81b350 | 2026-08-08 | c81b350 | 18,573 B | A | view · diff |
| git:20260702.0d142fe | 2026-07-02 | 0d142fe | 18,741 B | A | view |
Audit of the latest version
- pass: Frontmatter block present
- pass: Frontmatter declares a name
- pass: Frontmatter declares a description
- pass: Size between 200 bytes and 200 KB (11464 bytes)
- pass: No zero-width or bidi control characters
- pass: No instruction hidden inside an HTML comment
- pass: No link to an exfiltration or paste host
- pass: No credential-shaped string
- pass: No instruction to send local credentials anywhere
- pass: No text hidden with inline styles
- pass: No prompt-injection phrasing
- pass: No curl or wget piped into a shell
- pass: No recursive delete of root, home or parent
- pass: No instruction to read or print local credentials
- pass: No base64 blob over 200 characters
- pass: No link to a raw IP address
- pass: No script tag
Source
scdenney/open-science-skills · 69 stars · license a license file GitHub does not identify · pushed 2026-11-05 · branch main
API
GET https://markdownregistry.com/api/v1/artifacts/art_p3cil5c5uvveqmos GET https://markdownregistry.com/api/v1/resolve?ref=scdenney/open-science-skills/llm-calibration-logprobs GET https://markdownregistry.com/api/v1/blob/2f581002e1af295e991326c3b05fd8e762d860d501dfecf097d6626b573c8af6
Your AI assistant does the legwork. You hear about the deals worth your word. Hand yours the standing instructions at modelranch.com and it joins the network that reads files like this one.
Guides: the SKILL.md format and its required fields · what to check before you install a skill · how to pin a skill to an exact version · where to find agent skills · keeping a team on the same skills · which agent file does what.
More from scdenney/open-science-skills
Every file in scdenney/open-science-skills