Standing PR for board changes. Board-only (backlog/**) — never code.

I push planning work here as it happens instead of opening a PR per change. Merge it whenever suits you; nothing here blocks a dev, so no single commit is urgent. After you merge, I'll open a fresh one from the same branch for the next batch.

What lands here: new tickets and drafts, status/assignee changes, priority and dependency edits, ticket descriptions, and verification notes when I close something.

Contentious calls go in the ticket body rather than this description, so the reasoning survives the merge.

Currently included

DRAFT-14 — per-group transformer fitting, transformer(x) OVER (PARTITION BY g)

Your proposal, routed via Wren from TASK-3 brainstorming. Wren's measurements are recorded verbatim in the ticket — they're what turn it from an idea into something scopeable.

The reframe that matters: a transformer ref is fitted once globally today. Measured — with country a={10,20}, b={30,50} the state holds avg_age=[27.5], not per-country 15/40. That is the correct semantic (it matches sklearn, where a Pipeline step fits once on all training data), so this is a new feature, not a bug.

The split is the spine of the ticket:

• (a) SQL-expressible — StandardScaler, MinMax, MaxAbs, mean-SimpleImputer. Already works end-to-end today. Pure desugar, zero engine work. Measured: PARTITION BY yields one state row per group, batch == infer, and unseen groups return NULL for free via the existing unseen-partition semantic.
• (b) Opaque sklearn objects — genuinely new work. N fitted clones keyed by group, per-group state threaded through the artifact and lookup path.

Four questions need your ruling before it becomes a task:

1. Unseen-group policy for (b) — NULL, global fallback, or error? Wren's point: (a) inherits NULL by construction, so any other choice for (b) creates a semantic split between two halves of one feature.
2. Shorthand OVER (g) vs standard OVER (PARTITION BY g). Wren leans standard — (a) then desugars into already-valid SQL, whereas a shorthand buys new parsing for no new capability.
3. Does (a) ship alone first? It's nearly free and covers the common case.
4. Artifact size for (b) — N clones grows the artifact with group cardinality.

Flagged as distinct from DRAFT-11: that one is about how arguments bind; this is about what data a transformer is fitted on.

🤖 Generated with Claude Code