ChatAnalytica
Evidence beforethe answer.
How to read the analysis and its context.
ChatAnalytica Finance
The vault brings together market information, financial statements, news and context. Each answer describes the records used, their dates and relevant gaps. Counts describe available evidence for that answer, not a guarantee of complete coverage.
Choose the depth
- Light
- Read factsheets and data directly.
- Medium
- Connect evidence and answer your question.
- Ultra
- Explore more context, scenarios and conflicting evidence.
Facts, interpretation and scenarios
Lytica can analyse companies, charts, patterns, news and hypothetical strategies. It separates facts from interpretation, upside evidence from downside evidence, and explains what would change its reading. An AI probability estimate is a subjective judgement; a historical frequency describes a sample and period. Neither guarantees the future.
Chart patterns
When Ultra reads a chart, the pattern is not chosen by the AI: code recognises it, with fixed rules, on daily closing prices. It looks for 14: head and shoulders and its inverse, double and triple tops and bottoms, three triangles, two wedges, two channels and the rectangle. Each case is dated on the day it completes, using only the data available that day.
For each pattern we measure what happened next in our history: in what share of cases the price rose and in what share it fell after 1 and 3 months, with the median, the number of cases and a 90% interval. Next to it there is always what happened on any session of the same assets and years. The table gives the share of cases that rose, with the cases from January 2024 to October 2026. “Cases” counts those at 1 month; at 3 months there are slightly fewer, because the most recent ones have not yet completed three months. It is a historical frequency, not a probability of the future.
| Pattern | Cases | 1 month | 3 months | Uncertainty at 1 month |
|---|---|---|---|---|
| Head and shoulders | 392 | 50.8% | 57.0% | 90% interval: 42.7%–58.5% |
| Inverse head and shoulders | 263 | 47.9% | 51.9% | 90% interval: 41.4%–53.7% |
| Triple top | 706 | 56.4% | 60.2% | 90% interval: 49.8%–62.3% |
| Triple bottom | 681 | 53.0% | 54.1% | 90% interval: 47.7%–58.0% |
| Double top | 3,617 | 56.3% | 60.4% | 90% interval: 50.2%–61.9% |
| Double bottom | 3,329 | 51.9% | 54.7% | 90% interval: 48.0%–56.0% |
| Symmetrical triangle | 289 | 54.0% | 59.6% | 90% interval: 46.7%–61.2% |
| Ascending triangle | 640 | 54.5% | 56.4% | 90% interval: 49.0%–59.8% |
| Descending triangle | 920 | 53.3% | 56.1% | 90% interval: 48.4%–57.7% |
| Rising wedge | 541 | 55.5% | 56.3% | 90% interval: 50.4%–60.7% |
| Falling wedge | 744 | 54.7% | 59.2% | 90% interval: 48.1%–61.3% |
| Ascending channel | 676 | 53.6% | 56.8% | 90% interval: 48.7%–58.5% |
| Descending channel | 632 | 59.7% | 60.0% | 90% interval: 54.5%–63.8% |
| Rectangle | 199 | 57.8% | 60.8% | 90% interval: 52.3%–62.9% |
| Any session of the same assets and years | 487,113 | 53.8% | 56.4% | 90% interval: 49.8%–57.7% |
- How it is recognised: zigzag turning points on closing prices. A turning point counts when the price moves away from its extreme in proportion to its own volatility over the previous 63 sessions. The tolerances of each pattern were set before measuring.
- Out of sample: the only tuning, the zigzag sensitivity, was chosen with the cases from March 2022 to December 2023 and without looking at what happened next. What we publish are the cases since January 2024, measured once.
- Independent spans: cases from the same month move together with the market, so we count the months (quarters at 3 months) and the interval resamples whole months. A pattern is published with 150 or more cases spread over 24 or more months.
- Who is in: the 3,672 vault assets that trade today, losers included; out, those whose history is broken by unverified splits and those with fewer than 300 sessions. Assets that stopped trading are not there, which may inflate the share that rose a little: the any-session reference has the same bias.
- Data and version: closing prices from Yahoo Finance (stocks, ETFs, indices, currencies and commodities) and CoinGecko (crypto). Bank patrones-v1-2026-10-06, reproducible: the same vault and the same date give the same result (fingerprint fd8c723bf2d8).
- What we do not do: no price target, no pattern “measured move”, no entry or exit levels, no “if it breaks…”, no lists of assets showing a pattern. Lines end at the last candle and carry no price.
The 0-100 score
Every asset with a sufficient price series receives a score from 0 to 100. It is computed by code, not by the AI: same data, same score. Formula p1, in force since 17 September 2026. There are three scores by horizon; if you do not ask for one, the 3-6 month score is used.
| Component | 3-6 months | ≈ 1 month | ≈ 12 months | What it measures |
|---|---|---|---|---|
| Trend | 20 % | 20 % | 10 % | Price versus its 50- and 200-session averages, average crossover, slope of the 200-session average and position in the 52-week range. |
| Momentum | 25 % | 20 % | 15 % | Returns over 1, 3, 6 and 12 months (the latter excluding the latest month), compared with its group. |
| Technical indicators | 10 % | 30 % | — | Conventional readings of eight oscillators: RSI, MACD, stochastic, ADX, CCI, momentum, Williams %R and awesome oscillator. |
| Analyst consensus | 15 % | 5 % | 25 % | Third-party data: mean recommendation, upside to the median target and 3-month revision, compared with its group. Stocks with 5 or more analysts only. |
| Financials | 15 % | — | 30 % | Revenue and earnings growth, margins and return on equity, and valuation (forward P/E, price/sales, EV/EBITDA), compared with its group. Stocks only. |
| News | 5 % | 15 % | — | Balance of Bullish and Bearish readings of the news from the past 14 days; with few stories it weighs less. 3 or more are required. |
| Contained risk | 10 % | 10 % | 20 % | Annualised volatility and largest drawdown over the past year, compared with its group: less risk, higher score. |
- “Compared with its group” is a percentile: stocks with stocks, ETFs with ETFs, crypto with crypto; commodities, indices and currencies together.
- No data is not zero: a component without data counts as neutral (50), neither adding nor subtracting, so that missing data gives no advantage. If what is available is below 60 % of the weight, the asset stays “unscored” with its reason.
- A “top” includes: stocks above 300 M USD of market cap (by default, those above 10,000 M; mid caps, small caps or all can be requested), price of 1 or more, traded value of 1 M or more per day (20-session median), no anomalous jumps in the series and 75 % of the weight available; ETFs above 1,000 M in assets that are neither leveraged nor inverse; crypto assets among the 100 largest, excluding stablecoins and wrapped tokens. The rest have a score on their factsheet but take no position.
- Record: each asset’s score and the top 20 of each group are stored once a day in an append-only record; it is never rewritten or deleted. It started on 17 September 2026.
The score orders assets with today’s data: it does not predict, it is identical for every user, it uses no personal data and it is not an instruction to buy or sell. Analyst consensus is third-party data, not an opinion of this service. A high score does not guarantee a rise and past returns do not anticipate future ones.
Bullish, Neutral and Bearish
Labels summarize the possible effect of news on the asset and time horizon, with a reason. Neutral may mean mixed effects or uncertainty; insufficient samples are labelled accordingly. Social sentiment describes observed posts, not the entire market.
Dates and coverage
Quotes may be delayed. News shows publication dates and financial statements their reporting periods. Gaps are never converted to zeros. The estimated liquidation map models zones and is separate from observed liquidations.
How high could it go
If you ask how high an asset could go in 1, 3, 5 or 10 years (or the years you say, up to 10), code computes three scenarios, bear, base and bull, each with its price, the change from today, the change per year and its assumptions in one line. If the company makes money: the earnings per share forecast by the analyst consensus for each fiscal year (Nasdaq, Zacks data) as far as it goes and, after that, growth that fades every year towards that of the economy; times its sector P/E in our vault (lower quartile, median and upper quartile), with the dilution or buybacks of its shares over its last fiscal years (SEC EDGAR). If it does not make money but has sales: its sales per share, with the growth of its last fiscal years fading, times its sector price-to-sales. If it has neither earnings nor sales (crypto, a fund, a commodity): only its statistical band, the Lytica Forecast band for the first year and with no trend after that; half of the paths end between the bear and the bull case.
Next to the base case, what would have to happen: how much earnings or sales would have to grow every year, against what they grew in its last fiscal years. It is an estimate, identical for everyone; it is neither a recommendation nor a guarantee. Every estimate shown is recorded, and the list of the last 12 months is public.
Estimates historyYour usage
The beta is free. View tokens used and your available allowance percentage in chat, with each reset time. Light does not consume AI analysis allowance.
Scope
Analysis is informational, may contain errors and is not tailored to your personal circumstances. It is not investment advice and does not execute trades. Avoid including personal financial details in your questions.
Attributions
Required content attribution is retained alongside the content. Our privacy information identifies services that process your data.
Privacy