Thirteen tabs each answer one question well. This answers the one actually being asked: is anything wrong. Every tile is a verdict, not a metric, and opens the tab that can act on it.

In-house tick feed, bar aggregator and alert emitter

Feed connects and drops, warm-start, reloads, every alert submitted or rejected — newest first, this process only

One alarm when no alert from any source (TradingView or in-house) has arrived for this many minutes in market hours, and a note when alerts resume. Not in the first minutes after the open, not on holidays, not while nothing is expected to signal.

A strategy that stopped firing leaves no record anywhere else: the webhook stays up, the heartbeat stays green and the desk reads a quiet day. This judges each strategy against its own learned cadence instead.

Every in-house signal is a function of a bar that may never have been built. The feed reads green whenever any symbol is ticking, which is exactly when one name has silently had no bars for an hour and its strategy has stopped being able to fire.

Kite pushes every order state change down a WebSocket. It shortens the wait between an entry filling and its stop-loss resting at the broker, and catches a stop firing or an RMS rejection without waiting for the next reconcile. Read-only — it never places an order.

Every live order and the state the broker last reported: pending, open, partial, filled, rejected or cancelled. A rejected entry frees its slot at once; a refused exit puts the position back. "Reconcile had to fix" is what the 5-minute safety net corrected that this missed — it should stay at 0.

Each order has its own tag. When Kite does not answer, its order book is searched for that tag before anything is resent, so a slow network never doubles a position. Refusals (margin, RMS, circuit, invalid) are never retried.

One socket per live account with a valid Kite session. An account with no session simply polls, as it did before.

NSE holidays and Muhurat sessions. Every "is the market open today?" check reads this — the uptime report, auto-square-off, the reconcile window, the paper trade monitor and the entry gate.

Removing a seeded date keeps it removed across restarts.

Every change to a setting that steers trading — risk rules, sleeve sizing, in-house strategy parameters — with what moved and who moved it. Open the history to see earlier versions or put one back.

One row per change made through the admin API — who, what, when, and whether it was accepted. Reads are not recorded; nor is the trading webhook, which has its own trail under Trades.

Derived from the broker's own answers. An account whose broker stops responding refuses new entries only — exits, stop-loss arming, kill-switch flattens and auto-square-off are never gated on it, because during an outage those are the calls that matter most.

On an ex-date a split or bonus moves the price without anything happening to the company, so the day's indicators and any stop carried from yesterday are priced in the wrong money. A confirmed action refuses entries on that symbol for the ex-date; a suspected one only alerts.

Accounts, risk limits, sleeve sizing, strategy parameters and the calendar exist nowhere but this database — trade history is recoverable from the broker, this is not. Snapshotted daily. Credentials are never included; every account needs its Kite login re-entered after a restore.

Nifty 100 daily candles from 2010 and 1-minute candles as far back as Zerodha has them (5- and 15-minute are built from the 1-minute). Fills itself after the close, never during the session, and stops at the database size budget. Prices are as Zerodha sends them — not adjusted for splits or bonuses.

Replays every in-house strategy over the archived candles (2015 onward) with the same code the live engine runs, and saves its signals as backtest evidence — kept apart from real alerts and shown only where you pick Evidence → Backtest (Strategy Performance, Strategies Edge). Runs outside market hours only, picks up where it stopped, and starts a strategy over by itself when its settings change.

Which NSE industry each stock is in, for the sector-exposure limit. A stock with no sector is blocked by that check, so keep this current: import NSE's Nifty 500 list about monthly (niftyindices.com → Indices → Nifty 500 → ind_nifty500list.csv), and set ETFs or anything else missing by hand. It is also the in-house ticker's default list of stocks.

The list the entry gate consults. NSE republishes it in full, so an import replaces the frameworks it contains — a symbol that has dropped off must stop being blocked. Dry-run first to see what changes.

Every address that has posted to /trigger_AlgoExecution since this process started. Run a session, then paste the suggested value into WEBHOOK_IP_ALLOWLIST — that beats a published range list, which ages out without telling you.

Re-submits through the real /trigger_AlgoExecution path — same risk engine, sizing, duplicate guard and order queue as a live alert. A replayed entry that already got through is refused as a duplicate; a missed exit is the one this recovers. Paste from TradingView's alert log, or give a stored trigger id.

One-off jobs. Each is safe to run twice.

Cancels pending orders, market-exits every open position and stops every account trading, on all accounts at once and concurrently. For when the broker is degraded, a strategy is misfiring everywhere, or you have stopped making decisions and want out. Every exit is a Zerodha order — if Kite is down, so is this, and the result will say so account by account.