Since 8 October 2026 the outlook is built from three readings. Each answers a different question, from different data, on its own clock, and every sentence it writes names that timeframe. Each casts one vote — up, down, or no view — and every reading is shown on the page whether or not it has a view.

1. Trend (past year). Up when the last close is above the 200-day average price and the price is higher than it was 252 trading days ago, leaving out the last 21 (the “12–1” return). Down when both point down. No view when the two disagree. Needs a year of prices; with less it has no view and says so.
2. Results day. The move from the close before a results date to the first close after it, against an ordinary two-session move for that stock (the middle of the last 120 sessions’ two-session moves). A move at least 2× ordinary votes its own way for 60 trading days from the first close after the results; a smaller one does not vote. Results dates come from Twelve Data’s earnings calendar; a stock with none on file has no view.
3. Insiders and short sellers. Directors buying on market in the last 90 days, by more value than they sold, votes up (SEC Form 4 for US stocks, Twelve Data’s filings elsewhere). For ASX stocks, reported short positions of at least 3% of shares that rose by at least 1 point between ASIC’s reports vote down. Both at once is no view. Directors’ sales never vote.

Why these three. They do not share an input: a year of prices, one dated reaction to the company’s own results, and what people with money at stake have done. The year-long trend and the drift after a large results-day reaction are among the most replicated findings in published research; directors’ purchases carry information where their sales mostly do not; a large, rising short position has historically been followed by weaker prices. Sources are in the evidence table.

What no longer votes, and why. Until 8 October 2026 five rules voted: RSI past 30/70, a MACD cross, a 20/50-day average cross, a volume spike and insider filings. Three of them measure nearly the same thing — how the price moved over the last few days or weeks — so “three of four agree” mostly counted one fact three times, and each fired on a single day, so the outlook flickered with small moves. None has a record of predicting the next month on its own. They are still computed and shown as chart figures; they do not vote. Volume participation was evaluated as a fourth reading and left out: it largely restates the last month’s direction. Valuation and dividends never vote: “positive because the P/E is low” is an opinion about value, and they are shown as facts in The company section instead.

Count the votes. More up than down is a Positive outlook; more down is a Negative outlook; a tie, or no reading with a view, is Mixed / no clear outlook. Agreement is said as a count — “2 of the 2 readings with a view point up” — never as a probability that the price will move.

An outlook is labelled High agreement only when all 3 readings have a view and agree. One reading agreeing with itself is 100% agreement on the thinnest possible evidence, and is never labelled strong. Mixed is never labelled anything.

Each part of a stock page names its own timeframe: Last few weeks (trending or range-bound, from ADX), Where the chart is (a pattern on the chart today), and the outlook’s readings (the past year, the last results, the last 90 days). A stock can be range-bound for three weeks inside a year-long fall; the page says both, and they do not contradict each other.

The readings use fixed settings, the same for everyone; a portfolio’s strategy settings change the chart figures (RSI, MACD, the averages), not the readings.

These are shown on every stock page under Chart figures (no vote) and drawn on the chart. Since 8 October 2026 none of them votes in the outlook; they describe the recent price, nothing more. The defaults below are what every new portfolio starts with; a portfolio’s strategy settings can change them, and changing them changes only these figures.

• RSI: 14 days. Below 30 is described as having fallen fast (“oversold”), above 70 as having risen fast (“overbought”).
• MACD: the 12-day and 26-day exponential averages, with a 9-day signal line. The figure says whether the MACD line is above or below its signal line.
• Moving averages: the 20-day and 50-day, compared with each other; the chart also draws the 100-day and 200-day.
• Volume: a day is flagged when its volume is at least 1.5 times the 20-day average volume (an average that includes that day).

The readings that do vote — the past year’s trend, the reaction to the latest results, and insiders and short sellers — use fixed settings, set out above.

Two readings per stock, from its own price history only, never compared to another stock:

• Trend strength from ADX(14): below 20 is choppy, 25 or above is trending (direction from +DI vs −DI), in between is transitioning.
• Volatility state: realised volatility over the last 20 trading days divided by the last 60. A ratio of 1.2 or more is elevated, 0.8 or less is contracting, otherwise normal.

A stock with fewer than 65 bars reads “unknown” on both rather than guessing. Alongside each state the app notes the general historical tendency (trend-following approaches have tended to do better in trending markets, range-bound approaches in choppy ones) and never says which to use now.

Six fixed rules, read off each stock’s own daily price and volume history (dividend-adjusted), never off another stock. When more than one applies, an event (a break, an unusual day) is shown ahead of a position (extended, breaking down), which is shown ahead of a setting (pulled back, consolidating); the rest are listed beside it.

• Consolidating: ADX(14) below 20, and the last 15 or more closes inside a band no wider than 3× ATR(14) as a share of the price, floored at 4% and capped at 12%. The page reports how long the range has run (up to 120 days) and its bounds.
• Broke out / broke down: within the last 5 trading days, a close above the top (below the bottom) of a range that was Consolidating the day before. Whether volume that day was at least 1.5× its 20-day average is noted, and the record is kept separately for breaks with and without it.
• Pulled back in an uptrend: ADX(14) 25 or more, the 50-day average higher than it was 20 trading days ago, the close not below the 200-day average (where there is one), and the close within 2% of the 50-day average having been 8% or more above it in the last 30 trading days.
• Extended: the close above the 200-day average by more than the stock’s own 90th percentile of that gap, measured over every earlier day that had a 200-day average (at least 120 of them). “Further than it has been on 9 days in 10” is that percentile in words.
• Breaking down: ADX(14) 25 or more with −DI above +DI, the close below the 200-day average for 40 closes in a row or more, and each of the last 3 blocks of 21 trading days topping out lower than the one before.
• Unusual day: volume 3× its 20-day average or more, or a close-to-close move of 2.5 standard deviations or more of the last 60 days’ moves. The app reads no announcements, so it says so and points at them. This is not a detector of manipulation: a price series cannot say why people traded, only that they did.

The record: for each state, every day the stock entered it (the first day it applied, not every day it stayed) is scored by the close 20 and 60 trading days later - higher, lower, or within 1% - next to the base rate of every day in the same period, exactly as the readings backtest scores views. The stock’s own record and the pooled record across every stock people here hold or watch are both shown; a count under 10 says so. A stock with fewer than 65 bars is not read; the 200-day states need a 200-day average with history behind it.

Every sentence this produces is a description of a position on a chart and a count of what followed. None of it says to buy, sell, enter or exit, and an automated test holds that line over every sentence the module can produce.

Alerts and review points built on the states. The daily alert can carry a change of state (measured against the state you were last told about, the way an outlook change is) and unusual days, each behind its own switch on your account page, where outlook changes and your own review points have switches too. A review point is yours: a level below or above what you paid, a trailing point (a fall of N% from the highest close since you set it - the level rises with the price and never falls, and it is told once per fall, re-armed by a new high), or a condition on the state: when its range breaks, when a break falls back into its range (within the five sessions after it), when it becomes Extended, when it stops being Extended, when it closes more than 2% below its 50-day average, when it enters Breaking down, when it closes above its 200-day average - or one of two pairs, becoming and leaving Extended, or a range break and Breaking down, each part told once. One condition can be set for every holding at once, and any stock can have its own or none. Watched stocks can carry a chart-state condition too, with a default of their own; the cost levels and the trailing point are for holdings only. Trendentum applies the condition you wrote and records when it was reached - on every account it shows in Since your last visit and the Sunday digest, and with Pro the daily alert carries it that evening; it never suggests one.

Three measures per holding — 63-trading-day total return (about three months), 126-trading-day total return (about six months), and annualised volatility over the last 60 trading days (lower ranks higher) — each turned into a percentile rank within your own portfolio, then averaged with equal weight. Because the ranks are relative to what you hold, they have no meaning for a stock outside your portfolio, which is why shared pages show the raw figures instead of a rank.

One sentence built from fixed templates — never from a language model — picking whichever of three findings is most notable: concentration (your largest holding as a share of total position value, converted to US dollars only when a portfolio spans more than one currency), sector exposure, and factor exposure (the rankings above, read back as one sentence). Each finding states a fact and, where there is an obvious lever, what pulling it would do. It never says what to do. There is no blended score anywhere in it, and every figure is one click from its arithmetic. Holdings with no usable price are excluded and the panel says so.

One square per holding, coloured by the outlook above: green positive, amber mixed, red negative, blank when the provider returns no price data. The copyable text contains the squares, the date and time, a tally, and this site’s address — never a stock code, a price, a quantity or a return. It is rendered from the colour sequence alone, so a symbol cannot reach it; an automated test holds that property across randomised portfolios.

• Prices: US stocks from Alpaca’s free market-data feed when configured, otherwise Yahoo Finance; every other market (ASX, LSE, TSX, HKEX, SGX and the rest) from Yahoo Finance. Yahoo Finance is an unofficial source and does occasionally fail; a symbol that fails repeatedly is flagged in the app as “no price data” rather than shown as a dash.
• Insider filings: SEC EDGAR, Form 4, for US companies; Twelve Data’s insider-transaction filings for other markets.
• Results dates: Twelve Data’s earnings calendar, refreshed weekly.
• Short positions (ASX): ASIC’s daily aggregated short-position reports, published four business days in arrears.
• Company facts for the deep dive: SEC-reported annual revenue, net income, cash and debt (US only); optional news headlines when a news source is configured.
• Deep dive summaries: written by Claude (Anthropic) from those facts only, instructed to describe and never recommend, with an automatic check for directive language. Available to logged-in users for their own holdings and watchlist; never generated for anonymous visitors.

Signals are recomputed four times a day, at each market close — New York, Sydney and London — plus the Sydney open, and again whenever you open the app and the newest signal is more than 15 minutes old. Because every rule runs on daily bars, a stock's inputs change once a day, at its own market's close; the other passes re-read it on the same bar and move it only if the data did. Every past signal is kept; nothing is overwritten.

• Tell you to buy, sell or hold anything, or when.
• Show a score without the arithmetic behind it one click away.
• Publish a return figure, a track record, a backtest result or a leaderboard. None exists on this site.
• Put a stock code, price or return into anything designed to be shared.
• Substitute invented prices when a data source fails. It says the data is missing.