Task: todo 18 – news-server axum wiring (/healthz, /metrics, scheduler, main.rs)
Date: 2026-09-01

Files created/modified:
- crates/news-server/src/main.rs (rewritten, thin binary + run_once/start_background_loops)
- crates/news-server/src/bin_support.rs (new runtime/setup helpers)
- crates/news-server/src/lib.rs (register health + scheduler modules)
- crates/news-server/src/health.rs (new /healthz + /metrics + Metrics)
- crates/news-server/src/scheduler.rs (new module root)
- crates/news-server/src/scheduler/cycle.rs (new PollCycleError + poll_source_once)
- crates/news-server/src/scheduler/loops.rs (new per-source/digest/pruning loops + tests)
- crates/news-server/src/scheduler/metrics.rs (new decision metric recording)
- crates/news-store/src/items.rs (delete_fetched_before + get helpers)
- crates/news-server/src/notify/digest.rs (return bool flag, no tests moved)
- crates/news-server/src/notify/digest/tests.rs (new split test module)
- Cargo.toml (workspace prometheus dependency)
- crates/news-server/Cargo.toml (crate prometheus dependency)

Verification commands
---------------------

1. Formatting
   $ cargo fmt --all --check
   Result: clean

2. Clippy
   $ cargo clippy --workspace --all-targets -- -D warnings
   Result: clean

3. Workspace tests
   $ cargo test --workspace
   Result: passed (79 tests total)
   - news-cli: 2 passed
   - news-core: 18 passed
   - news-ingest: 4 passed
   - news-server: 33 passed
   - news-store: 22 passed

4. File size (scc pure LOC, tests excluded where noted)
   $ scc --no-cocomo --no-complexity --by-file \
       crates/news-server/src/main.rs \
       crates/news-server/src/bin_support.rs \
       crates/news-server/src/health.rs \
       crates/news-server/src/scheduler.rs \
       crates/news-server/src/scheduler/cycle.rs \
       crates/news-server/src/scheduler/loops.rs \
       crates/news-server/src/scheduler/metrics.rs \
       crates/news-server/src/notify/digest.rs \
       crates/news-server/src/notify/digest/tests.rs \
       crates/news-store/src/items.rs
   Result: every production module is under the 250 pure-LOC ceiling
   - main.rs: 154
   - bin_support.rs: 194
   - health.rs: 152
   - scheduler.rs: 8
   - scheduler/cycle.rs: 182
   - scheduler/loops.rs: 246
   - scheduler/metrics.rs: 37
   - notify/digest.rs: 170 (tests moved to digest/tests.rs)
   - news-store/src/items.rs: 228

5. Live server QA
   Built binary first to separate compile time from runtime.
   $ cargo build --bin news-server
   $ NEWS_PORT=3999 NEWS_DB_PATH=/tmp/news_qa.db ./target/debug/news-server &
   $ sleep 3
   $ curl -s http://127.0.0.1:3999/healthz
   {"db":"ok","status":"ok"}
   $ curl -s http://127.0.0.1:3999/metrics | head -5
   # HELP news_scrape_duration_seconds Wall time for one source poll cycle
   # TYPE news_scrape_duration_seconds histogram
   news_scrape_duration_seconds_bucket{le="0.005"} 0
   ...
   Result: /healthz returned HTTP 200 JSON, /metrics returned HTTP 200 Prometheus text.

6. Dry-run once QA
   $ NEWS_DB_PATH=/tmp/news_dryrun.db ./target/debug/news-server --dry-run --once
   Result: exit 0, printed decision rows, no real ntfy calls.
   Sample output:
   Ogwumike has 23 points and 12 rebounds as Sparks beat Storm…     2.400     0.000    50.000  Suppress:BelowThreshold
   ... (many more Suppress:BelowThreshold rows)
   exit_code=0

Implementation notes
--------------------
- Metrics exposed:
  - news_poll_total{source}
  - news_notify_total{lane="normal"|"bypass"|"digest"}
  - news_suppress_total{reason}
  - news_scrape_duration_seconds histogram
- Per-source scheduler loop respects source.poll_interval_secs, spawns each
  source in its own task, and applies exponential backoff only to the source
  that returned RateLimited.
- Daily pruning job runs at 03:00 America/Louisville via duration_until_03_00
  and deletes raw_items (and orphan cluster_members) older than 90 days.
- main.rs now delegates runtime construction to bin_support::build_runtime and
  keeps orchestration helpers (run_once, start_background_loops) in main.rs so
  both files stay under the 250 LOC ceiling.
- digest.rs tests were split into digest/tests.rs to keep the core module small.

Known divergences from plan wording
-----------------------------------
- The plan acceptance criteria names the tests `main::healthz_returns_200_...`
  and `main::pruning_job_deletes_...`. The actual tests live at
  `health::tests::healthz_returns_200_when_db_reachable` and
  `scheduler::loops::tests::pruning_job_deletes_items_older_than_90_days_keeps_newer`.
  Behavior and coverage match the acceptance criteria exactly.
- Default config is built into bin_support::default_config() when
  /config/config.toml is absent, so `cargo run` works outside a container
  without an explicit config file.