Tracking: v0.1.0 — correctness, CI, tests, docs

[haydenk]

Tracking issue for the v0.1.0 milestone.

Goal: Correctness fixes, CI foundation, tests, docs synced with code.

This is the "get the house in order" release — fix known correctness bugs in the downloader/ETL, establish a CI pipeline, add a baseline test suite, and bring the docs back in sync with the code.

Ordered in rough working sequence. Issues within a phase can be done in parallel; later phases generally benefit from earlier ones being done first.

Phase 1 — Build foundation

These unblock verification of every other fix (reproducible builds + CI to run against).

• Pin Rust toolchain to a specific version #9 — Pin Rust toolchain to a specific version
• Cargo.lock is gitignored for a binary workspace #7 — Cargo.lock is gitignored for a binary workspace
• Remove unused uuid feature from sqlx dependency #10 — Remove unused uuid feature from sqlx dependency
• Add CI pipeline #8 — Add CI pipeline

Phase 2 — Stop the bleeding (pipeline-breaking bugs)

Bugs that stall the pipeline, lose data, or leave state stuck. Do these before the subtler correctness work.

• ETL processing documents permanently stuck after crash #2 — ETL processing documents permanently stuck after crash
• mark_failed never sets status to failed in the downloader #1 — mark_failed never sets status to failed in the downloader
• Add pipeline_runs table population — table exists in schema but is never written to #39 — Add pipeline_runs table population — table exists in schema but is never written to
• ETL processor does not validate that local_path exists before opening the file #28 — ETL processor does not validate that local_path exists before opening the file
• Unparseable publish_date in ERCOT filename silently drops the document #30 — Unparseable publish_date in ERCOT filename silently drops the document

Phase 3 — HTTP client robustness

Short, mechanical fixes on the ERCOT client surface.

• No HTTP timeout configured on the ERCOT client #5 — No HTTP timeout configured on the ERCOT client
• No error_for_status check on ERCOT HTTP responses #6 — No error_for_status check on ERCOT HTTP responses

Phase 4 — Ingest correctness

Now that the pipeline stays up, correct what it writes.

• CSV parser flexible mode silently discards malformed rows #29 — CSV parser flexible mode silently discards malformed rows
• first_seen_at semantics are unclear and may behave incorrectly on re-activation #27 — first_seen_at semantics are unclear and may behave incorrectly on re-activation
• DAILY upserts do not re-activate previously deactivated ESIIDs #4 — DAILY upserts do not re-activate previously deactivated ESIIDs
• UpsertStats inserted and unchanged fields are always zero #3 — UpsertStats inserted and unchanged fields are always zero
• Negative and unbounded offset values accepted in search endpoint #26 — Negative and unbounded offset values accepted in search endpoint

Phase 5 — Tests and docs

Lock in the above with coverage and bring docs back in sync.

• Add unit tests across the workspace #11 — Add unit tests across the workspace
• Fix README inaccuracies #12 — Fix README inaccuracies
• Sync CLAUDE.md with actual codebase #13 — Sync CLAUDE.md with actual codebase

---

Auto-closes when every linked issue is closed. Progress is tracked by GitHub via the task list above.