[jonahgabriel]

Summary

OMN-20329. node_projection_read_effect now serves contract-declared exposures from the local SQLite store when the runtime projection binding's database_url is a sqlite: or file: DSN (local MVP mode 1). Decision D1 (a) routes the local dashboard's reads through this node (OMN-20159), so this is the read path for that mode. Before this change the resolver refused every non-Postgres binding with projection_binding_unsupported.

• ports/sqlite_row_source.py: SqliteTableRowSource serves the same window, tenant scoping, row shape and named refusals as the Postgres TableRowSource, in SQLite's dialect. It opens the file read-only per read and never creates a missing file (projection_database_unavailable).
• Before reading, it checks the table and every column the read names against pragma_table_info. SQLite reads a double-quoted unknown column as a string literal, so without the check a missing column is served as its own name. It answers projection_column_missing instead.
• resolve_projection_read_source: postgres/postgresql resolve to TableRowSource unchanged; sqlite/file resolve to SqliteTableRowSource; any other scheme is still refused projection_binding_unsupported; no binding is still projection_binding_unconfigured (no fallback).
• sqlite_path_from_dsn moves to the shared projection/sqlite_database.py, and the delegate orchestrator's evidence DB resolver imports it instead of keeping a private copy, so the writer and the reader resolve a DSN to the same file.

dod_evidence

• tests/unit/nodes/node_projection_read_effect/ (incl. new test_projection_read_effect_sqlite_omn20329.py), test_evidence_db_resolution_omn14015.py, test_golden_chain_projection_read_effect.py: 39 passed on lab host h202.
• tests/integration/nodes/node_projection_read_effect/test_projection_read_effect_real_postgres_omn20159.py: 5 passed against a disposable postgres:16-alpine on h202, including the new case proving a Postgres binding still reads and close() still drains its pools.
• Real local store on h202 (~/.omninode/delegation/delegation.sqlite, 22 delegation_events rows, one tenant), read through the node with a sqlite: binding: before the catalogue check the decisions exposure answered ok with 5 rows; after it, ok=False 503 projection_column_missing. That store lacks id, authority_source, score_source, cost_tier_name and cost_tier_type, which the exposure declares. Making the local writers declare those columns is the plan's T3.0 store-neutral writer work, not this change.
• Dev lane binds Postgres and that path is unchanged, so no .201 lab step applies.
• ruff format/check clean, mypy --strict clean on touched files, pre-commit passed.

Evidence-Ticket: OMN-20329
Evidence-Source: OCC#12231

[github-actions]

✅ Architectural Review — PASSED

Errors: 0
Warnings: 0

What this checks

Rule	Description
ARCH-TOPIC-001	No hardcoded Kafka topic strings in handler code
ARCH-DI-001	No event_bus=None bypass
ARCH-DI-002	No direct Handler instantiation outside workflow_runner/adapters
ARCH-DI-003	No reinvented DI containers
ARCH-TOPIC-002	contract.yaml topics follow `onex.{cmd

---

No architectural violations found.

Static architectural lint — no model inference (OMN-14176).

[github-actions]

✅ Hostile Reviewer — PASSED

Blocking findings (critical/error): 0
Total findings: 12
Models succeeded: qwen3-review,gpt-oss-review

---

Gate semantics

Verdict	Meaning	Blocks merge?
passed	>=2 models succeeded, no critical/error findings	No
blocked	CRITICAL or ERROR findings found	Yes
degraded	Fewer than 2 models succeeded (infra unavailable/timeout)	Yes (OMN-15110)

Powered by omniintelligence.review_pairing.cli_review — node-based adversarial review via HandlerLlmCliSubprocess (OMN-8524)