Skip to content

docs(rfc-0002): dataset immutability principle + reject --append (cli#156)#238

Merged
saqlainsyed007 merged 1 commit into
developfrom
docs/rfc0002-append-rejected
Jul 13, 2026
Merged

docs(rfc-0002): dataset immutability principle + reject --append (cli#156)#238
saqlainsyed007 merged 1 commit into
developfrom
docs/rfc0002-append-rejected

Conversation

@LukasWodka

@LukasWodka LukasWodka commented Jul 11, 2026

Copy link
Copy Markdown
Contributor

What

Records the dataset-immutability principle in RFC-0002 and turns the --append non-goal from "tracked separately" into an explicit rejection, now that cli#156 is closed obsolete.

Two edits to docs/rfcs/0002-data-ingest-flow.md:

  • §2 Principles — new Principle 6: datasets are immutable snapshots. A dataset is the fixed basis for its use cases + training; reproducibility depends on it not changing under already-trained models. Ingest creates or wholesale-replaces (--overwrite); never appends in place. Growth = a new (versioned) dataset. (The RFC never actually stated this principle before, even though the whole flow depends on it.)
  • §13 Non-goals--append rewritten from "tracked separately, cli#156" to rejected, with the integrity/reproducibility rationale and the pointer to backend#1073 (the canonical decision record) + cli#156 (closed obsolete 2026-07-11).

Why

cli#156 (data ingest --append) was closed as obsolete — in-place append breaks the reproducibility guarantee the leaderboard depends on. The RFC-of-record still framed append as a pending/tracked item, which was misleading. This makes the design-of-record match the decision.

Docs-only (the RFC markdown); no code change. Rev 5.


Note

Low Risk
Markdown RFC updates only; no runtime or API behavior changes.

Overview
RFC-0002 now states that datasets are immutable snapshots as Principle 6: ingest only creates or wholesale-replaces (--overwrite); growth is a new versioned dataset, not in-place mutation.

§13 no longer treats server-side data ingest --append as tracked work—it is explicitly rejected, aligned with cli#156 (closed obsolete) and backend#1073. The rationale is reproducibility and leaderboard integrity after training.

Docs-only (Rev 5); no code changes.

Reviewed by Cursor Bugbot for commit a2f6fa8. Bugbot is set up for automated code reviews on this repo. Configure here.

…#156)

Adds Principle 6 (datasets are immutable snapshots) and rewrites the §13 --append non-goal into an explicit rejection now that cli#156 is closed obsolete. In-place append breaks dataset integrity/reproducibility; growth = dataset versioning, not mutation. Rationale of record: backend#1073.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@saqlainsyed007
saqlainsyed007 merged commit d13befa into develop Jul 13, 2026
20 checks passed
@saqlainsyed007
saqlainsyed007 deleted the docs/rfc0002-append-rejected branch July 13, 2026 10:12
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants