docs(19): §19.6 describes a layout reality diverged from - #41
Merged
Conversation
…it went The diagram plans five things under `training/`. Two exist, two are unwritten, and one was built in Rust instead. Four empty directories would have satisfied the diagram and told a reader nothing, so they are deliberately not created. `datasets/` is the interesting one: the transcript miner shipped at M19.4 as `panday_harness::mining` + `cargo xtask mine`, not as Python. That is the right place for it — it parses panday's own JSONL event logs, whose shape is a Rust type with golden fixtures, and a Python re-implementation would be a second parser for a moving format that disagrees the first time an event gains a field. `export/`'s signing half is also Rust (`panday models sign`). `recipes/` and `registry.py` are unwritten and blocked on a base-model choice and GPU hours; committing unsloth hyperparameters for a model nobody has picked would be config-shaped fiction. ADR-001's boundary holds — Rust serves, Python trains — the line just sits further into Python than the diagram drew it. Claude-Session: https://claude.ai/code/session_017kFpYDqvz6sKGSkM4YKaRf
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Task 6 of the run brief: "
training/is 1 of the 5 directories docs/19 §19.6 mandates. Create them with real content or change docs/19 to stop requiring them. Either is fine; pretending is not."Changed the doc, because two of the four missing pieces were built in Rust and the other two are blocked on hardware. Four empty directories would have satisfied the diagram and told a reader nothing.
datasets/panday_harness::mining+cargo xtask mine(M19.4)evals/recipes/export/panday models signregistry.pydatasets/is the one worth reading. The miner parses panday's own JSONL event logs, whose shape is a Rust type with golden fixtures — a Python re-implementation would be a second parser for a moving format, and the two would disagree the first time an event gained a field. So it landing in Rust was correct, not a shortcut.The pattern: the deterministic, format-owning half migrated to Rust; the model-shaped half is waiting on a model. ADR-001's boundary still holds, the line just sits further into Python than the diagram drew it.
Also, on task 8
The brief asks me to define M0.2, which it says docs/23 cites without defining. Already done —
docs/23-roadmap.md:137defines it properly in M0.1's style, with a four-clause state table and an honest "three hold as far as CI can hold them; two need a human." Landed in PR #30. The brief's premise was stale; no change needed.Doc-only change:
fmt --checklocally, CI is the gate.https://claude.ai/code/session_017kFpYDqvz6sKGSkM4YKaRf