CandencePattern

One row per symbol, showing the last time its trend flipped sides. Click any row for that symbol's full history.

• “Today, per symbol” is one row per symbol, read off the state table and the other panels at the last close: each system's daily side and how many sessions it has held (GNG, Custom, and any other system in a column of its own: GNG+MQ, MenthorQ, or Legacy, the TrendSpider alerts of 18 to 27 Aug, shown only as when it was last seen), whether the sides current at the last close agree, the newest intraday read of each system and whether it sits with that system's own daily side, where the scale-in clock stands, and the mark since entry (or since the exit, after a NoGo). Everything is dated against the last close, never "today": a flip is “at the Sep 23 close”, and a row with nothing new since the previous close says nothing new in its last column. The Custom side is a heartbeat: its batch fires at every close a symbol stands in Strong Go or Strong NoGo, every symbol at once, from about a minute after the close (16:00 New York, 13:00 on a half day), its rows landing over up to half a minute. So the side is current only if it fired at the last close. After a close, a Custom side that was strong at the close before reads in turn: no Custom heartbeat seen for the Sep 23 close yet until the first batch row lands, then the batch for the Sep 23 close is still arriving for five minutes, and only then, if its own row has not come, left strong GO at the Sep 23 close, which is a change; the first two are not. When closes that proved nothing sit between its last row and the first close that proves its silence, the exit is said as a span, left strong GO between the Sep 22 and Sep 24 closes, with the reason, and is a change at that first proving close. Before the batch lands, every Custom side reads that way. Which batch row counts is decided close by close. Where the rows name their alert (the batch script's batch field lists the symbols each alert checked, sent from 25 Sep once its five alerts are recreated), the proof is per batch: only a row whose alert lists the symbol counts; while none has, the line says no heartbeat from an alert listing it seen, and the close proves nothing about it. Where only rows naming no alert landed (rows from before the field), the proof is per close and the reading says (per close): it assumes one batch row from any symbol vouches for all five of the batch's alerts, and that the symbol is on one of them; an alert that failed reads the same way there. A row delivered more than five minutes after the close's first batch row is not routine (every batch row in the 24 Sep export landed within half a minute of the first), but if one is, the symbol reads as having left, which is new, until the row lands and it reads as current again. A batch row received more than 15 minutes after its close (every one in the 25 Sep export came within 88 seconds of it) is late: it still says its own symbol's side, but it proves nothing about any other symbol at that close, either way, so a symbol it would have proven reads no heartbeat landed in time, and the line above the table counts such rows. A heartbeat the gateway refused for its token says heartbeat refused at the Sep 23 close: the gateway did not accept its token instead, but only when the refusal came from one of TradingView's own webhook addresses (the address the gateway judged it by, which it records on every rejected row; on rows stamped by gateway version 2026-09-25 or later the gateway also records whose address list, in the configuration it held when it judged the request, holds that address, and that decides, except where it could compare nothing (no list, or its configuration could not be read), where the page checks the address against TradingView's list as published on 23 Sep, as it does on every row stamped by an earlier version; on a row recorded before 24 Sep it takes the last address of the stored chain, and not on a chain so long it was cut) and arrived between the close and five minutes after the batch's first row. Anyone can post a refused row, so any other refusal is ignored, and the close then reads as silence, which here means left strong GO and new: a real refusal from an address TradingView adds later reads that way until the address is added in /admin. An intraday bar labelled 1D (an alert on a script older than 22 Sep, left on an intraday chart) that reached the gateway at a close before 24 Sep could take the heartbeat's place in the ledger; the page then reads the heartbeat it displaced as the daily row, and only if that row is not loaded does the close say the Sep 23 close is unknown, not left or quiet. It is not a change, and the fix is the alert. Since 24 Sep the gateway keeps the two apart, so both rows are stored. The batch reports only STRONG states, so leaving strong GO is not the same as leaving GO. A symbol that comes back after a close the batch ran without it has entered strong GO or NOGO again at that close, and its sessions and mark count from there; a symbol's first Custom heartbeat on the page is said as the first one seen. It never says buy, sell, add now or reduce: this page is a record, not advice. Our own daily proxy studies of the add-on and trim (P-142, P-143, P-145, Aug 2026) measured our stand-ins for them and are not a verdict on GoNoGo. The one thing it states about size is a reading of the measured scale-in schedule (P-163): a countdown to the next add, or that an add point has been reached.
• An old date is normal. The indicator changes sides only about 5 to 10 times a year per symbol (recent median about 7), so a three-week-old row means that trend is still running.
• Add-ons and trims show as a badge under Since flip, not as their own row — the trend did not change sides.
• “no flip yet” means only in-trend events have arrived for that symbol so far, never a Go or NoGo.
• Patterns, the panel of that name, lists chart-pattern events from a pattern detector: a pattern forming (watch) or its break (breakout), the newest per symbol, frame and pattern, with a breakout linked to the watch before it when that row is on the page. They are display only: never a stance, never a tier and never a trade. This table, the Today panel and every scoring panel leave them out, and the Observe gate counts them apart; a symbol's drill-down lists them among its events, each naming its pattern, stage and detector. They are measured after a week of rows, first against what GNG and Custom said at that close, then by how far price followed through.
• There is no alert for Neutral, so a symbol that has gone quiet still shows its last flip. For the indicator's reading right now, check the chart.
• Export CSV saves the rows this page loaded, in a file named for the instant the page read the ledger, in UTC, never the time of the click: for example candence-signals-2026-09-25T20-20-05Z.csv. The intraday checker and agreement.py read that instant as when the export was taken. A page left open exports what it loaded, under that instant; press Refresh first for newer rows.

The columns:
System — which alert sent this row. Notes — anything worth knowing about the row. What the other system says about this symbol right now: ✓✓ both agree, or a line naming each side when they differ; and no flip yet when the row is an in-trend event rather than a real flip. Blank when there is nothing to add — one system alone is not a cross-check.
Trend — the side, plus a coloured dot for the strength wording the Custom alerts carry. Since flip — the latest in-trend event. Pattern: the newest chart-pattern event on that symbol and frame, recent ones only, in a dashed tag. Display only, never a stance. Tap or click it to open the Patterns panel at that event. Alert price — the price when the alert fired.
Move size — how hard the stock moved after the signal, measured in that symbol's own typical daily range, so a 3% move is read as violent for GLD and unremarkable for TSLA. Result — a Go closed by the next NoGo, or on a NoGo row the move avoided until the next Go (negative is good there). Until the opposite signal arrives it reads open, marked at the latest daily close. Age — trading sessions closed since the row, counted to the last close when the page loaded; last close for a daily row at that close, this session for an intraday row that arrived inside the session in progress, since last close for one that arrived after the last close, outside a session. Amber from 5 sessions (a trading week); the drill-down's Since column uses the same words and turns amber from 21 sessions (about a trading month).
Timeframe and the exact chart bar are on the Signal time cell's tooltip; hover any tag, badge or heading for a fuller explanation.

What the marks mean
GO the trend is bullish  ·  NOGO the trend is bearish
▲ Turned Go / ▼ Turned NoGo — the trend changed sides. The only event the automatic indicator itself calls a signal.
● Add-on point — the trend paused, held, then pushed on again. A point to add to something you already hold, not a new entry.
▼ Trim warning — price ran too far and is losing steam. A bearish warning even while the trend is still GO.
⚡ Breakout — a quiet market started moving. A separate system; never an entry on its own.
Strength dot — how strong the trend is: Strong Go Weak Go Neutral Weak No-Go Strong No-Go. Only Custom rows have one — the automatic indicator does not report strength, so blank there is correct, not missing.
These marks and colours match what the indicator draws on a chart, if you have one open.

⚠ These are two different indicators — do not average them together.
GNG is an automatic indicator: a ready-made trend model, captured exactly as it fires. Custom is a rule built here from RSI, EMA and ADX.
They report different kinds of thing. GNG marks an event — the few times a year a trend actually turns. Custom reports a daily condition — it fires again every day the condition holds. Averaging a rare event together with a daily reading produces a number that means nothing, which is why each is scored separately above.
Reading them against each other is useful, and that is what Notes does. But both look at moving averages, so when they agree it may be for the same reason rather than two independent confirmations.