SENTINEL Trend

One trailing line: green in an uptrend, red in a downtrend, flips on a confirmed close. Non-repainting, so history never moves. Optional Heikin-Ashi smoothing. Eight alerts, and a panel that grades the line on your own chart.

Version 2.0 (24 September 2026) adds the timeframe board: the same line graded on ten timeframes at once, from 15 minutes to a week, for the coin on your chart, each with its own preset. Every row shows how many long legs it graded, the share that lost after cost, the average gain of a long leg from the flip up to the flip down, the share that reached two percent before flipping back, and the date its record starts. It is a record of what happened, not a forecast: in our study of 148 Bybit perpetuals a coin's calmest past timeframe did not predict its calmest next one, so the board says "top so far" and never marks a timeframe unless its average leg was positive. The panel also compares the open leg with the coin's typical long leg, and reads bitcoin's daily trend as context.

Version 2.1 (24 September 2026) adds one exit rule, the only one that survived our exit study out of sample. When bitcoin's daily trend was down at a long leg's entry and the leg has gained 5 percent, the script draws a trailing level 2 percent under the highest close since entry, and the seventh alert fires when a close falls through it. The take profits, stops and plain trails we tested never earned a place, so in a bitcoin uptrend the line holds to the flip. The rule and both numbers can be changed or switched off in the settings.

Version 2.2 (2 October 2026) comes out of our October study: one minute candles on ten coins, 576 settings on each of fifteen timeframes, then 112 more coins over four years. What decided the result was how long a leg lasts. No line whose legs typically lasted under a day made money after cost, on any coin or setting, so the board now shows the average hold and the flips a month of every timeframe, and the panel says FAST LINE when the line on your chart flips more than sixteen times a month. Below the daily chart the line is drawn dashed while it runs against the coin's own daily trend, daily turns are marked, the panel splits your chart's long record into the stretches with the daily trend up and down, and an eighth alert fires when the daily trend turns down under an open uptrend. Long legs paid more on average while the daily line was also up, in each of four years on the study's 112 coins, though a wider check of 546 perpetuals in October 2026 could not show that gap to be reliable. It is a tilt, not a rule: most legs lost either way. On a chart of one coin divided by bitcoin the panel now says so and stops comparing the coin with bitcoin; in the study that chart was no more reliable than the dollar chart.

Version 2.3 (October 2026) adds a setting for the weekly chart, and changes nothing unless you choose it. A study of 1,377 weekly settings on bitcoin's whole history, checked by two AI model reviewers and then by our own independent check, found that the weekly line flips almost as often as the daily one on bitcoin. The new setting offers a calmer line for bitcoin's weekly chart, 14 by 2.0 instead of 14 by 1.0. Since 2011 it flipped 34 times against 90, with fewer losing long stretches, and the price of that is getting back in later after a bottom, up to 17 weeks later. Whether it earns more than the old line is not established, because the answer changes with the weekday the week starts on, so it stays off unless you pick it. Every other coin keeps 14 by 1.0, because no alternative beat it once settings were scored in a way designed not to reward sitting in cash. The same setting also offers an experimental drawdown filter, long only about one week in twenty on alts, which lost to the old line on risk adjusted return in every coin group we tested. The 2W, monthly, daily and intraday charts, every alert title and the weekly context row are unchanged.

It is context, not an edge. A trend line tells you the regime; it does not tell you a trade has positive expectancy. In our own out-of-sample testing a bare flip did not survive costs, and about two thirds of long legs ended below cost on every timeframe, preset and shield we tried. What mattered was gating it: a higher-timeframe trend filter, a not-overextended check, and real risk sizing. We tested tuning the settings to each coin twice and a live selector once; none beat fixed presets out of sample, so there is no per coin table here. The one exception is the opt in calmer weekly line for bitcoin, chosen on bitcoin's own history and labelled that way. Treat it as decision support, not a forecast. Info only. Not financial advice.