Credit
The Independent 100 list was curated by dan (@irl_danB) and published on 2026-09-24 at independent.prose.md (announcement). The ranks and section names are his, and number 100 is his open wild-card slot. This study is ours and is not endorsed by him or by anyone on the list.
What they publicly do
Head of Evals at Every. He is an O'Reilly author on prompt engineering and DSPy and runs an AI course [Reported: X profile x.com/hammer_mt; GitHub profile github.com/hammer-mt]. He publishes eval methods and small open tools, including Claude Code skills and agent loops [Verified: github.com/hammer-mt/eval-pack].
- X profile: x.com/hammer_mt
- GitHub: github.com/hammer-mt
- Site: saxifrage.xyz
- Blog: saxifrage.xyz/blog
- Also: every.to/p/how-to-grade-ai-and-why-you-should
Public work
- The Pairwise Gold Triad (eval pattern) blog no license stated last activity 2026-03-18
An eval for subjective tasks. A fixed LLM judge picks whichever of incumbent or challenger is closer to a human gold reference, which gives a stable landscape for hill climbing. Test cases are built by fill-in-the-middle extraction from a writer's own paragraphs, and the loop plugs into autoresearch. The DSPy code is in the post.
- eval-pack repo no license file last activity 2026-07-09 · 3 GitHub stars on 2026-09-24
A Claude Code/Codex skill that makes the agent build one self-contained HTML evidence pack for a PR or feature (screenshots, test output, architecture before/after), used as a verification forcing function.
- chief (Chief Wiggum) repo no license file last activity 2026-01-14 · 3 GitHub stars on 2026-09-24
Two Claude Code instances in an adversarial loop. Ralph implements PRD tasks. Chief supervises, reverts bad commits and writes lessons into CLAUDE.md. Both run in a Docker sandbox.
- DSPyUI repo MIT last activity 2026-06-06 · 232 GitHub stars on 2026-09-24
A user interface for building and compiling DSPy programs.
- context-engineering-dspy-book repo MIT last activity 2026-09-15 · 3 GitHub stars on 2026-09-24
Notebooks for the O'Reilly book 'Context Engineering with DSPy'.
- thumb repo no license file last activity 2025-05-24 · 29 GitHub stars on 2026-09-24
A small prompt A/B testing library with blind human rating and cost and token stats.
Licenses of GitHub repositories are from the authenticated GitHub license API on 2026-09-25, one row per repository in evidence/licenses.jsonl (“no license file” means the API found none; “license file present, not recognised by GitHub” means GitHub could not classify it). Other licenses are as the source states. Check the license file before reusing anything.
Relevance to our work
5 of 5 lenses: evals, verification, agents, writing, content
The pairwise gold triad is a method that is adoptable now for subjective evals such as brand voice and video scripts, and eval-pack and chief are small examples of verification and supervisor-loop practice. DSPyUI and the book code are MIT.
Scored 0 to 5 against the study’s relevance rubric: 5 means public, openly licensed work squarely in our lanes that we could study or adopt now, and 1 means little connection to our work. The score is one reviewer’s judgement of how much this public work bears on our own work. It is not a ranking of people and says nothing about the person or the quality of their work.
What we would study next
- Prototype the pairwise gold triad on a brand-voice task: fill-in-the-middle gold set from a writer's own paragraphs, fixed judge, incumbent/challenger loop (saxifrage.xyz/post/pairwise-gold-triad-eval). It gives brand-voice and script writing a comparable, hill-climbable subjective eval. It needs only about 20 gold examples. Effort: medium.
- Read eval-pack's evidence-pack contract as a model for acceptance evidence and closeout receipts (github.com/hammer-mt/eval-pack). It applies 'verify, don't trust' to closeouts, packaged for human review speed. There may be format ideas to take. Effort: small.
- Read chief.sh/ralph.sh as a minimal supervisor-worker loop and compare it with NTM orchestration (github.com/hammer-mt/chief). A small reference design for revert-and-teach supervision. Useful as a baseline for heavier orchestration such as NTM. Effort: small.
Before reusing anything: DSPyUI and context-engineering-dspy-book are MIT. eval-pack, chief and thumb have no LICENSE file on GitHub (all rights reserved by default): study the ideas and reimplement, don't copy the code. chief runs Claude Code with --permission-mode acceptEdits inside Docker, so review the sandbox before any use.
Sources
- github.com/hammer-mt
- saxifrage.xyz
- saxifrage.xyz/post/pairwise-gold-triad-eval
- github.com/hammer-mt/eval-pack
- github.com/hammer-mt/chief
- github.com/hammer-mt/DSPyUI
- github.com/hammer-mt/context-engineering-dspy-book
- github.com/hammer-mt/thumb
Overall confidence in this record: High. Sources were read on 2026-09-24 and every link was checked on 2026-09-25. The full record is line 46 of people.jsonl.
Limits
Not verified: The Head of Evals role at Every beyond his own bios; the student count; thumb's PyPI license metadata (the GitHub API reports none).
This record comes from a first-pass survey of public material by an AI agent session on 2026-09-24, edited for publication on 2026-09-25. Nothing was installed or run for it. Roles, stars and licenses change; see the study’s limits.
Request a correction
Is something about this record wrong or out of date, or would you rather not be covered? Open a correction (a GitHub issue form with this page’s address filled in). A request to remove a record is honoured without argument; accepted corrections are recorded in the changelog with the date.