Add single-process live engine runner

.gitignore

@@ -214,3 +214,5 @@ __marimo__/
 
 # Internal working docs
 .docs/
+# CocoIndex Code (ccc)
+/.cocoindex_code/

README.md

@@ -1,10 +1,10 @@
 # Augur
 
-Structured market anomaly detection for prediction markets. Augur observes Polymarket and Kalshi with adaptive polling, extracts typed signals with calibrated confidence, and attaches investigation prompts drawn from a frozen library. The canonical consumer interface is a JSON schema; deterministic Markdown and a gated, opt-in LLM formatter are built on top of it.
+Augur extracts structured intelligence signals from prediction markets. It observes Polymarket and Kalshi markets, measures consensus velocity, volume, liquidity, order-book pressure, and cross-market divergence, then emits typed events for downstream agents and analysts to investigate. The canonical consumer interface is a JSON schema; deterministic Markdown and a gated, opt-in LLM formatter are secondary renderings built on top of it.
 
 Augur is not a forecaster, an arbitrage engine, or a news writer. It is a deterministic structured-signal pipeline. See `docs/foundations/overview.md` for the full product framing and `docs/foundations/non-goals.md` for what Augur explicitly does not do.
 
-Current version: **0.1.0**. Phase 1-5 scaffolding landed; runnable surfaces are the test suite, the labeling CLI, and the distributed-runtime smoke stack. See `docs/operations/manual-testing.md` for the end-to-end guide.
+Current version: **0.1.0**. The component implementation is substantial, but the live proof loop is not complete. Runnable surfaces are the test suite, the labeling CLI, the single-process engine runner, and the distributed-runtime smoke stack. An active watchlist, backtest runner, calibration runner, and real consumer feed remain follow-up work. See `docs/operations/manual-testing.md` for the current manual testing guide.
 
 ## Documentation
 
@@ -43,11 +43,11 @@ All three workspace packages — `augur-signals`, `augur-labels`, `augur-format`
 Each workspace package exposes extras for opt-in integrations. Install only what a deployment needs:
 
 ```bash
-# LLM secondary formatter (phase 4)
+# LLM secondary formatter
 uv sync --extra llm-local        # augur-format[llm-local] — Ollama client
 uv sync --extra llm-cloud        # augur-format[llm-cloud] — Anthropic SDK
 
-# Distributed runtime (phase 5)
+# Distributed runtime
 uv sync --extra bus-nats         # NATS JetStream adapter
 uv sync --extra bus-redis        # Redis Streams adapter
 uv sync --extra storage-timescale # TimescaleDB via psycopg
@@ -59,24 +59,33 @@ The dev dependency group in the repo root already pulls every extra so CI exerci
 
 ## Runnable Surfaces
 
-### Labeling CLI (phase 2)
+### Labeling CLI
 
 ```bash
 uv run python scripts/label.py --help
 uv run python scripts/label.py candidates
 uv run python scripts/label.py decide <candidate-id>
 ```
 
-### Worker entrypoints (phase 5)
+### Single-process engine runner
+
+```bash
+uv run python scripts/run_engine.py --help
+uv run python scripts/run_engine.py --once
+```
+
+`scripts/run_engine.py` loads `AUGUR_CONFIG_DIR` or `config/`, opens the DuckDB store, runs the existing in-process extraction engine, and writes canonical `SignalContext` JSON to stdout. It fails fast when `config/markets.toml` has no active markets. Active Kalshi markets require `KALSHI_API_KEY`; Polymarket-only watchlists do not.
+
+### Worker entrypoints
 
 ```bash
 uv run python -m augur_signals.workers                 # catalog
 uv run python -m augur_signals.workers.poller --help   # per-kind entrypoints
 ```
 
-The `workers` package exposes bootstrap helpers (`augur_signals.workers.bootstrap`) that every `__main__` module uses for config loading, observability activation, and bus connection. Per-kind transform wiring for feature / detector / manipulation / calibration / dedup / context_format / llm requires a follow-up commit — see `docs/operations/manual-testing.md §3`.
+The `workers` package exposes bootstrap helpers (`augur_signals.workers.bootstrap`) that every `__main__` module uses for config loading, observability activation, and bus connection. Per-kind transform wiring for feature / detector / manipulation / calibration / dedup / context_format / llm requires a follow-up commit. See `docs/operations/manual-testing.md §4`.
 
-### Migration scripts (phase 5)
+### Migration scripts
 
 ```bash
 uv run python scripts/migrate_to_timescale.py backfill --from labels/snapshots_archive
@@ -120,10 +129,10 @@ augur/
 ├── pyproject.toml          # uv workspace root (v0.1.0)
 ├── uv.lock
 ├── config/                 # TOML configuration
-│   ├── bus.toml            # phase 5 — message bus backend
-│   ├── storage.toml        # phase 5 — DuckDB / TimescaleDB selector
-│   ├── observability.toml  # phase 5 — Prometheus + OTel exporters
-│   ├── llm.toml            # phase 4 — gated LLM formatter
+│   ├── bus.toml            # message bus backend
+│   ├── storage.toml        # DuckDB / TimescaleDB selector
+│   ├── observability.toml  # Prometheus + OTel exporters
+│   ├── llm.toml            # gated LLM formatter
 │   └── ...                 # polling, detectors, dedup, formatters, consumers, labeling, markets, forbidden_tokens
 ├── data/                   # market taxonomy, investigation prompts, calibration state
 ├── labels/                 # newsworthy-event labels (Parquet)
@@ -134,6 +143,7 @@ augur/
 │   ├── export_schemas.py
 │   ├── label.py            # labeling CLI wrapper
 │   ├── lint_detector_now.py
+│   ├── run_engine.py       # single-process live runner
 │   ├── migrate_to_timescale.py    # phase 5 backfill + verify
 │   └── dual_write_sidecar.py      # phase 5 tee replay
 ├── src/

config/detectors.toml

@@ -8,28 +8,28 @@
 # tune against the historical corpus before live deployment.
 
 [price_velocity]
-hazard = 0.004            # 1/250
-fire_probability_threshold = 0.7
-resolution_exclusion_hours = 6
+hazard_rate = 0.004            # 1/250
+fire_threshold = 0.7
+resolution_exclusion_seconds = 21600
 
 [volume_spike]
 ewma_alpha = 0.05
 target_fdr_q = 0.05
 
 [book_imbalance]
-top_levels = 5
+depth_levels = 5
 bullish_threshold = 0.72
 bearish_threshold = 0.28
-persist_snapshots = 3
-min_total_depth_usd = 5000.0
+persistence_snapshots = 3
+minimum_total_depth = 5000.0
 
-[cross_market_divergence]
-window_hours = 4
+[cross_market]
+window_seconds = 14400
 min_historical_correlation = 0.6
-activity_floor_volume_ratio = 1.0
+activity_floor = 1.0
 
 [regime_shift]
 target_alpha = 0.02
-k_sigma = 0.5
-h_sigma = 4.0
-min_dormancy_hours = 6
+k_multiplier = 0.5
+h_multiplier = 4.0
+dormancy_minimum_seconds = 21600

docs/operations/manual-testing.md

@@ -1,6 +1,6 @@
 # Manual Testing Guide
 
-Augur has three runnable surfaces today: the test suite, the labeling CLI, and the distributed-runtime smoke stack. This document enumerates what can be exercised locally and what remains operator-wiring work.
+Augur has four runnable surfaces today: the test suite, the labeling CLI, the single-process engine runner, and the distributed-runtime smoke stack. This document enumerates what can be exercised locally and what remains operator-wiring work.
 
 ## 1. Quality gates and tests
 
@@ -40,7 +40,33 @@ uv run python scripts/label.py coverage                  # per-category coverage
 
 State persists to `labels/queue.json` and promoted rows land as partitioned Parquet under `labels/newsworthy_events/date=YYYY-MM-DD/`.
 
-## 3. Distributed-runtime smoke stack
+## 3. Single-process engine runner
+
+The monolith runner drives the existing in-process engine against configured active markets and writes canonical `SignalContext` JSON lines to stdout.
+
+```bash
+uv run python scripts/run_engine.py --help
+uv run python scripts/run_engine.py --once
+```
+
+Runtime contract:
+
+- `AUGUR_CONFIG_DIR` overrides the default `config/` directory.
+- `config/markets.toml` must contain at least one active market.
+- Polymarket-only watchlists run without platform credentials.
+- Active Kalshi markets require `KALSHI_API_KEY`.
+- DuckDB storage is opened from `config/storage.toml`.
+- Output is deterministic canonical JSON from `augur_format.deterministic.json_feed`.
+
+Current repository state still has only an inactive placeholder watchlist, so `uv run python scripts/run_engine.py --once` fails fast with:
+
+```text
+run_engine failed: config/markets.toml has no active markets
+```
+
+Populate `config/markets.toml` before using the runner for live capture.
+
+## 4. Distributed-runtime smoke stack
 
 The phase 5 compose stack brings up every external dependency the workers need: NATS JetStream, Redis, TimescaleDB, Prometheus, and (optionally) an OTel collector. Workers run as separate host processes so each one is inspectable.
 
@@ -128,7 +154,7 @@ bus = build_event_bus(cfg.bus)                        # nats or redis
 await bus.connect()
 ```
 
-## 4. Migration scripts
+## 5. Migration scripts
 
 Both scripts are fully runnable against the smoke stack once TimescaleDB is initialized.
 
@@ -162,7 +188,7 @@ uv run python scripts/dual_write_sidecar.py \
 
 Requires the engine to publish to `augur.writes` — this path is not wired in the monolith yet, so the sidecar is smoke-testable against handcrafted fixtures for now.
 
-## 5. Container build and Kubernetes
+## 6. Container build and Kubernetes
 
 ### Build the image
 
@@ -191,24 +217,24 @@ kubectl -n augur create secret generic augur-secrets \
     --dry-run=client -o yaml | kubectl apply -f -
 ```
 
-## 6. Observability
+## 7. Observability
 
 - Prometheus: `http://localhost:9090` after compose is up. Scrapes `host.docker.internal:9091..9097`.
 - NATS admin: `http://localhost:8222/varz`.
 - Redis CLI: `redis-cli -h localhost ping`.
 - TimescaleDB: `psql $AUGUR_TIMESCALE_URL -c 'select * from timescaledb_information.hypertables'`.
 - OTel collector: spans print to the container stdout (`docker compose logs otel-collector`).
 
-## 7. Tear down
+## 8. Tear down
 
 ```bash
 docker compose -f ops/docker/compose.yaml down -v
 unset AUGUR_CONFIG_DIR AUGUR_TIMESCALE_URL AUGUR_REPLICA_ID
 ```
 
-## 8. Known gaps
+## 9. Known gaps
 
-- No end-to-end monolith launcher (`python -m augur_signals.engine` has no `__main__`). The engine is driveable from Python and from `tests/signals/test_engine_integration.py`, but no production script.
+- The monolith runner exists as `scripts/run_engine.py`, but the checked-in watchlist still has no active markets.
 - `scripts/backtest.py` and `scripts/calibrate.py` are stubs that raise `NotImplementedError`.
-- Worker entrypoints for feature / detector / manipulation / calibration / dedup / context_format / llm require the bus message-schema work described in §3 above.
+- Worker entrypoints for feature / detector / manipulation / calibration / dedup / context_format / llm require the bus message-schema work described in §4 above.
 - Live failover tests against a real NATS or Redis cluster are operator-owned; CI uses dependency-injected fakes.

pyproject.toml

@@ -1,7 +1,7 @@
 [project]
 name = "augur"
 version = "0.1.0"
-description = "Structured market anomaly detection for prediction markets"
+description = "Structured intelligence signals from prediction markets"
 readme = "README.md"
 requires-python = ">=3.12"
 license = { file = "LICENSE" }