Repository navigation
feat(atlas): re-topic colibri probes — 8 categories, non_coding contrast bucket - #159
Merged
Merged
Conversation
…ast bucket Owner directive 2026-09-27 (hydra_vortex track): - ADDED: technical_architect, ui_development, reasoning_thinking (3 prompts each, lengths kept in the existing narrow band to avoid the prefill length-bias confound). - FOLDED: chinese / german / poetry / law / medicine / casual_chat into a single non_coding contrast category (representative medicine / german / poetry prompts) — keeps a non-coding baseline for the base-rate correction (p(c|e) normalization and spec = 1 - H/log C) while halving the non-coding topic count. - KEPT verbatim: code_python, code_sql, json_format, math_proof. Net: 10 x 3 -> 8 x 3 = 24 probes (sweep.sh / analyze.py read probes.json dynamically; no code changes needed). README provenance updated — the "verbatim from Colibri" claim is now "started from Colibri @a8f2ca62, owner-updated 2026-09-27". NOTE: this changes the topic DEFINITION only. The served atlas still carries the old 10-topic artifact until a 24-probe re-sweep regenerates experts.json / expert-ranks.json (and spec/labels change because C goes 10 -> 8; post-hoc editing the served JSON would leave affinities unnormalized — re-probe is the only correct path). Validation: python3 -m json.tool OK; 8 categories, 3 prompts each. Co-Authored-By: Oh My Pi (opencode-go/mimo-v2.6-flash)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Owner directive 2026-09-27 — update the Colibri probe topics:
technical_architect,ui_development,reasoning_thinking(3 prompts each, narrow length band preserved to avoid the prefill length-bias confound).non_codingcontrast bucket (medicine/german/poetry representative prompts) — keeps the non-coding baseline the base-rate correction needs (p(c|e) normalization, spec = 1 − H/log C) while dropping the six separate non-coding topics.code_python,code_sql,json_format,math_proof.Net: 10×3 → 8×3 = 24 probes.
sweep.sh/analyze.pyreadprobes.jsondynamically — no code change. README provenance updated (no longer claims "verbatim Colibri").Not in this PR
The served atlas still carries the old 10-topic artifact: a 24-probe re-sweep must regenerate
experts.json/expert-ranks.json(C changes 10→8, so spec/labels change; post-hoc editing the served JSON would leave affinities unnormalized).Validation:
python3 -m json.toolOK; 8 categories × 3 prompts.