How Parci works

Parci looks at each asset on four timeframes — 1 hour, 4 hours, 1 day and 1 week — and answers three questions: which phase the market is in and how much the price is likely to move, which way it leans, and how reliable each of these answers has been in the past. Everything is calculated only on closed bars: nothing on the chart changes after the fact.

Phase and projection

The model looks for recurring cycles in the recent price history, removes the long-term drift and fits the strongest cycles. Their sum, continued into the future, is the projection — the dashed line to the right of the last bar.

Growth phase: the projection for the next bar is above the last close.
Decline phase: it is below.

The projection is a hypothesis about the shape of the movement, not a price target. Its value is in the timing of turns, and even that is measured (see below) before you rely on it.

Movement and direction

Two statistical models look a few bars ahead — 5, 10 or 20 — and give a probability, not a promise:

Movement or calm: will the range from high to low over the next bars be wider than usual for the asset, whichever way the price goes. This is where the predictable part is: on most 1h and 4h pairs the model beats the best simple rule.
Direction: the odds that the price closes higher. Across the market it is only slightly better than simple rules, and no single pair has an edge proven — treat it as a weak signal.

In the app each forecast is shown with its probability and signal strength. How often each model was right, pair by pair, is on the track record page.

Turn windows

Where the projection changes direction, Parci marks a turn window: a growth window (▲) where a decline is expected to end, a decline window (▼) where growth is expected to end. A window has a centre and a width of ± a few bars, because the exact bar cannot be known.

With every new bar the model recalculates. If the window moves, you see it as a revision (“moved by 3 bars”), never as a silent rewrite. “In progress now” means the current time is inside the window.

Trading filter

The filter is an oscillator: how far the price is from its short-term average, measured in units of its own recent volatility.

Channel: adaptive upper and lower bounds. Leaving the channel means an unusually strong move.
σ levels: ±1, ±2, ±3 standard deviations. Crossing ±3σ is rare.
Reversal outside the channel: the oscillator turns back while still outside — often the first sign that the move is running out.
Divergence: price makes a new high (low) while the oscillator does not — the move is losing strength. A hidden divergence points the other way: continuation.
Squeeze: the channel narrows; a strong move often follows, direction unknown.

Events are recorded at the close of the bar that confirms them. The time you see is the moment the event became known.

Trend

The trend line follows the price and switches only when the price moves decisively past it. Uptrend, downtrend or sideways. A trend change on the daily chart is one of the few events the quiet notification preset includes.

How we measure quality

Anyone can draw a line into the future. What matters is how often it was right. For every asset and timeframe we run a walk-forward test: the model is calibrated on the past and evaluated on the following period it has not seen, again and again through the history. Then we compare it with simple baselines:

Always up: “the price will be higher in 10 bars”.
Momentum: “the price will keep going the way it went”.
Random dates for turn windows.

On the track record page a pair is marked “Confirmed” only if all of these hold: at least 500 historical forecasts, accuracy at least 3 percentage points above the best baseline, a statistically significant difference (p < 0.05) and, for the projection, turn windows that hit better than random dates. Because we test many pairs at once, the significance is corrected for multiple testing — otherwise a few pairs would look good by pure luck.

Where it says “No proven edge”, the model did not beat the simple baselines on that pair. The app still shows the forecast — with its probability and signal strength, not with a quality badge — so check the track record and treat such forecasts as experiments. Today turn windows hit no better than random dates, and the direction model has no proven edge on any single pair; the movement model is confirmed on most 1h and 4h pairs.

Notifications

Notifications are sent only for assets in your list:

Quiet (default): turn windows on 1d and 1w, and the daily trend change.
Standard: also 4h, channel breakouts, divergences, ±3σ.
Everything: all significant events on the timeframes of your plan.

Notifications arrive in the app’s inbox. In Settings you can also turn them on in your browser — then they show up on screen even when the Parci tab is closed; on iPhone and iPad the app has to be added to the Home Screen first. Quiet hours and a daily limit apply to browser notifications. Delivery to Telegram comes once the Telegram bot is connected (soon). You can mute an asset. The same event is not repeated within three bars.

Limitations

Crypto markets can move sharply on news no model can foresee.
Market data comes from an exchange and can arrive late or with gaps.
A confirmed model can stop working; the track record is updated after each calibration.
Nothing here is a recommendation to buy or sell.

Analytics, not investment advice. Forecasts are uncertain and past accuracy does not guarantee future results. You decide whether to rely on them.