consolidation: decide fate of hollow patterns/ shim vs canonical contrib/patterns/

[Gradata]

Problem

src/gradata/patterns/__init__.py is a back-compat shim that re-exports from src/gradata/contrib/patterns/. The real Layer 0 is contrib/patterns/; the top-level patterns/ directory is effectively hollow.

Impact:

• help(gradata.patterns) / tab-completion on patterns. yields nothing because the shim has no __all__
• Callers who import gradata.patterns.X silently get gradata.contrib.patterns.X via lazy hops
• Any future refactor of contrib/patterns/ must also maintain the shim or silently break callers

Proposed resolution (pick one)

A. Promote canonical: Delete src/gradata/patterns/, move contrib/patterns/ → src/gradata/patterns/, add deprecation notice to the old gradata.contrib.patterns import path.

B. Make the shim honest: Keep both locations, but give patterns/__init__.py a real __all__ so discovery works.

Option A is cleaner but breaks external imports of gradata.contrib.patterns.X. Option B keeps back-compat but locks in the bifurcation.

Context

Identified by council review of the autoresearch/consolidation branch. Needs a product-level decision — not a mechanical refactor.

[Gradata]

Audit — state of patterns/ shim

Current file on branch (autoresearch/consolidation):

"""Back-compat shim — patterns moved to gradata.contrib.patterns."""

def __getattr__(name: str):
    # Forward to contrib.patterns, fall through to gradata.rules
    ...

Internal callers in src/: zero.
Test-only callers:

• tests/test_bug_fixes.py:333-334 — smoke test asserting symbols import through the shim
• tests/test_spec_compliance.py:312 — from gradata.patterns import pipeline, reflection, guardrails

External callers (uv.lock/PyPI consumers): unknown. This is a 0.x SDK — any caller that hasn't migrated to gradata.contrib.patterns breaks if we remove the shim.

Options

1. Keep indefinitely (zero cost — the file is 26 lines, lazy-loaded). Cost: permanent 'ghost namespace' in the public surface.
2. Add DeprecationWarning on first __getattr__ access + schedule removal at v0.8.0 (same window as integrations/ adapters per consolidation: integrations/ orphan directory (no __init__.py) — merge into middleware/ or add contract #111). Cost: one warning filter in the two smoke tests.
3. Remove now. Cost: any external caller breaks silently at runtime (not import time — __getattr__ means import succeeds, then AttributeError on first use).

Recommendation

Option 2 — mirror the integrations/ deprecation pattern so the public surface shrinks on a predictable cadence. Low effort, reversible, and gives downstream users the audit cue they need.

Not autonomous-closing — this is a product/API-surface call for Oliver.

[Gradata]

Closed via #114. Council-decided Option B: DeprecationWarning + remove at v0.8.0. Runtime warning in getattr (not docs-only) so pinned-version users get the signal before v0.8.0 removes the shim.