pam ▸ docs/archive/2026-05-02-pam-hardening/diagnostics/overnight-diagnostic/STATE.md

updated 2026-04-16

Pam Operational State — 2026-04-16

TL;DR

Nothing is running. Pam has been silent for 5–6 weeks. Latest process log (data/overnight_overlap_loop.log) last wrote 2026-03-08 08:25 UTC. Background monitor’s monitor.log last wrote 2025-10-02 16:44 UTC — six months ago. No processes matching pam|background_monitor|overnight_overlap|opportunity_hunter|main.py are currently alive.

Processes

Process	Status	Last Activity	PID file
`background_monitor.py`	STOPPED	2025-10-02	`DEPLOYMENT_STATUS.md` claims PID 65284 — dead
`tools/overnight_overlap_loop.py`	STOPPED	2026-03-08 (iteration 96 of 96)	`data/overnight_overlap_loop.pid` = 53818 (stale)
`opportunity_hunter.py`	STOPPED	Crashed on `EOFError: EOF when reading a line` — it’s interactive-only, was backgrounded with no stdin
Main CLI (`main.py run`)	Not running	—	—
Telegram webhook	Not running	—	—
Scheduler	Not running	—	—

Credentials / venues wired

From .env (currently on disk, plaintext): - Kalshi: KALSHI_API_KEY=8ab716c9-08aa-4213-a339-8c42dc3cc21d, private key path secrets/kalshi-live.key (file exists). - Polymarket: POLYMARKET_PRIVATE_KEY=0x8a2…cc806e (live key), POLYMARKET_PROXY_ADDRESS=0x7AA1…FB80, POLYMARKET_SIGNATURE_TYPE=1. - Telegram: bot token + chat id + admin id present. - FORCE_DRY_RUN=0 — dry-run safety toggle is off. - ENV=prod, RAILWAY_ENVIRONMENT=production.

Note: every past start-up line in logs still says "Polymarket private key not provided. API will be limited." — the long-running monitor was launched without loading .env, so all historical scanning ran in reduced mode.

Signal sources feeding the system

Catalog inventory sync (watchers/catalog_inventory.py) — paginates Kalshi /markets and Polymarket CLOB/Gamma into data/unicorn.db.
DB currently holds 477 Kalshi markets and 16,225 Polymarket markets (catalog_market_inventory). Normalized: 477 Kalshi / 7,285 Polymarket.
Kalshi distribution: 196 Entertainment, 81 Elections, 52 Politics, 34 Economics, 29 Companies, etc. — distribution is reasonable, but scale (477) is very small vs. Polymarket.
Overlap watcher (watchers/overlap_watcher.py) — fetches markets then calls find_equivalent_pairs.
Opportunity scanner / RSS signal pipeline — code exists, not active.
Research loop (research/) — has latest_state.json with 22 “deployable_trade” candidates dated earlier; not actively regenerating.

Last known runs

Overnight overlap loop — 97 iterations, all returned matched_pairs: 0. First iter 2026-03-07, last 2026-03-08 07:25 UTC. Across 97 attempts, cumulative Kalshi=22,500 markets / Polymarket=35,258 markets ingested, zero pair matches produced. Also zero ranked_candidates, zero fee_loser_candidates.
Background monitor — 3,010 scans, 0 opportunities (Oct 2025). Started with “Polymarket private key not provided”.
No trades have been executed. arbitrage_trades table presumed empty (not checked live, but results.tsv in arbitrage/ is the only trade-like artifact and it’s an untracked working file).

Git state

On main, up to date with origin/main.
Uncommitted modifications in execution/kalshi_api.py, tests/test_kalshi_api.py, watchers/catalog_inventory.py. repo-snapshot.zip deleted.
Recent feature branches around autoresearch, near-pair review, Kalshi feed restoration.

Bottom line

Pam is off, and even when it was on it has never produced a matched pair in the recent architecture. The matching pipeline is returning 0 pairs across ~58k markets — that is the primary failure. Also: .env has live keys and FORCE_DRY_RUN=0; this is not safe to accidentally start.