astrostock

Forecast Labs

🔬

Eight Pro research labs that stress-test a projection before you trust it: Cycle Profile names the fundamental rhythms a symbol actually follows, the Cycle Tuner shapes one cycle's waveform against an out-of-sample grid, the Wavelet Lab extracts cycles from a time–period heat map, Consensus Zones keeps only the cycle behavior that repeats across independent eras, Swing Odds audits swing targets walk-forward, Seasonal Rhythms builds calendar-conditioned composites with an era-agreement check, Cycle Odds Zones scores recurring-cycle windows across every historical pass, and Indicator Odds compares oscillator readings against the symbol's base rate. They sit beside the Neural Forecast Lab in the Analysis rail.

Where to find them

Open Dashboard → Charts → Analysis. The labs appear as the Cycle Profile, Cycle Tuner, Wavelet Lab, Consensus Zones, Seasonal Rhythms, Cycle Odds Zones, Swing Odds, and Indicator Odds sections. Each one runs on the loaded symbol and applies its result back to the chart.

Cycle Profile

Press Build profile and the lab scans decades of weekly history for long rhythms, then tags each detected period against a catalog of named fundamental cycles: economic ones like the ~41-month Kitchin inventory cycle, the ~9-year Juglar investment cycle, the 8-year gold cycle, and the 520-day cycle; astronomical ones like the 584-day Venus synodic and the ~2-year Mars–Saturn heliocentric cycle; the ~11-year sunspot cycle with its half, third, and quarter harmonics; and the calendar's annual and 4-year Presidential cycles. The one-line Profile readout is the symbol's cyclical portrait — the named rhythms it demonstrably follows.

Every row must earn its name: the cycle is fit on older weeks only and graded on held-back recent weeks, and the badge shows that out-of-sample correlation. Grayed rows were detected but did not hold up — treat them as noise, however famous the name. Check a named cycle tests any catalog entry directly even when the scan did not surface it, for the "I expected the Kitchin cycle here" situation.

Each verified row also routes to the projection engine built for its kind: economic and other regular cycles hand their period to the Cycle Tuner, astronomical matches apply the Astro Composite panel with the planetary pair preloaded, solar matches point to the Sunspot Cycle Lab, and calendar matches point to Seasonal Rhythms or Cycle Odds Zones.

Cycle Tuner

The Tuner shapes ONE cycle's waveform with two honest knobs. Cycle memory decides how much history the wave is fit on: 0 uses everything (the cycle's long-run, permanent form), while 2–8 fits only the last few repetitions (its current form — sharper when behavior has shifted, noisier when it has not). Harmonics decides how detailed the shape may be: 1 is a pure wave, higher values allow skewed crests and pullback shoulders. Guards clamp harmonics automatically when the fit window is too short to support them.

Every variation is verified the same way: the wave is fit strictly before a held-out tail of bars, extrapolated across it, and scored on direction. The memory × harmonics grid shows every variation's held-out correlation at once — click a cell to switch to it. Finding parameters that fit the past is easy; the grid exists so only variations that kept working on unseen bars earn attention.

Apply to chart draws the three-segment overlay — dim historical fit, dashed out-of-sample tracking, solid forward projection. The Scenario fan draws all variations at once as a band anchored at the last close: narrow means the waveform is robust across settings, wide or split means the cycle is fuzzy and no single projection deserves trust. The agreement percentage summarizes how many variations end the horizon on the same side.

Wavelet Lab

Press Compute wavelet map to build a heat map of cycle strength: time runs left to right, period runs up the vertical axis (long periods at the top), and bright bands mark periods that were active at that moment. Dark left edges are warm-up where the math does not yet have enough history. The map never uses future bars, so a bright area describes what was knowable at the time.

Drag across a bright band to extract that cycle. The lab follows the ridge inside your selection, refits period, amplitude, and phase directly on the price data near the selection's end, and adds the cycle to a small table (up to six). The preview underneath superposes the selected cycles into one wave — solid where it overlaps history, dashed ahead of the newest bar.

Set Holdout above zero to verify before applying: the lab recomputes the map on the series with the newest bars removed, re-extracts your selections, projects across the removed span, and reports the direction hit rate and correlation against what actually happened. The held-out bars are never allowed to influence the projection being graded. The same validation view also draws a sign-following simulation line — what following the projection's direction bar by bar across the held-out span would have produced — and an inverted-overlay toggle with a rolling-correlation readout flags spans where price tracked the projection upside-down.

Apply projection to chart adds the Wavelet Projection overlay: a dim historical fit plus a forward segment starting at the last close. Every cycle's period, amplitude, and phase is pinned in the indicator settings, so saved layouts re-render the identical line; the Projection horizon and fit display stay adjustable from the indicator settings afterwards.

Consensus Zones

A composite curve can look convincing while summarizing behavior that only one stretch of history produced. Consensus Zones splits the lookback into equal, non-overlapping eras (two to four), rebuilds the cycle wheel for each era separately, and marks the phase zones where the eras actually agree — by direction, by shape correlation, or both. A zone an era never visited cannot earn consensus, no matter how the interpolated curve looks there.

Choose a cycle source — Annual cycle, Planet position, Pair phase (synodic), or Solar cycle — and press Find consensus. The readout overlays each era's wheel (latest era emphasized) with the agreeing zones shaded, alongside the share of visited phase that agrees and each era's sample count.

Apply filter to panel switches on the consensus filter of the matching panel — Astro Composite for planetary sources, Sunspot Composite for the solar cycle. The panel then paints the agreeing parts of its dashed forward line as solid emphasized segments or, in Consensus only mode, hides the disagreeing parts entirely. The same filter is available directly in each panel's settings; the annual source is a lab readout only.

Swing Odds

Swing Odds shows the statistics behind the Swing Projection Zones indicator. Run swing audit builds the symbol's swing-ratio distribution — each swing's size relative to the one before it — as a histogram with its density curve and detected peaks, plus the spread of swing durations.

Below it, the walk-forward table replays history honestly: every past pivot is re-projected using only the bars that existed when that pivot confirmed, then graded on where the next pivot actually landed — inside any zone (Price), inside the strongest zone (Top), inside the projected duration window (Time), or both (Both). Cov reports how wide the zones were relative to the swing being projected. Wide zones buy hit rate, so always read the two columns together.

A moon-phase split conditions the same distribution on the lunar phase — waxing or waning — read at each swing's start pivot (UTC day). Each phase gets its own distribution and walk-forward rows, so you can see whether swing behavior actually differed between the two halves of the lunar month; when one phase holds too few swings, the lab falls back to the pooled distribution instead of projecting from a thin sample.

Apply zones to chart adds the overlay with Provisional zones switched on: alongside the confirmed projection, dashed amber zones extend from the swing still in progress. Their anchor can move until that swing confirms, so treat them as tentative by construction.

Seasonal Rhythms

Seasonal Rhythms asks one question per mode: what has this symbol typically done at this point of the calendar? Modes cover the annual cycle, the decade pattern, the election cycle, the monthly shape, weekday behavior, and the lunar month. Each composite averages the symbol's own history conditioned on that calendar position and is drawn across recent history and ahead of the newest bar.

The era agreement readout is the honesty check: the lookback is split into independent eras and the composite is rebuilt for each era separately, so you can see whether the displayed shape repeats across eras or comes mostly from one stretch of history. Prefer modes whose eras agree.

An Anchor dates mode composites the same history around a recurring date list instead of a calendar position: use the Chinese New Year preset or paste custom dates one per line, choose the trading-day window before and after each anchor, and read the average path with the latest cycle overlaid. Its era check splits the anchor list itself, and the mode stays inside the lab — it has no chart overlay.

Apply to chart adds the Seasonal Projection overlay with the mode and settings pinned, so saved layouts re-render the identical line. The lab and the overlay are daily-interval tools — switch to a daily chart to use them.

Cycle Odds Zones

Some cycles never produce a clean wave — they matter only inside a specific stretch of their phase. This lab folds the symbol's history over a recurring cycle — US elections, BTC halvings, any fixed period from a chosen start date, or custom dates pasted one per line — and draws every pass as its own ghost curve, so you can see where the passes agree before trusting the average path.

Set an up window and a down window by day offsets from the cycle start, or press Suggest windows to search for stretches with a historical edge. The table reports the record across every completed pass — how many rose versus fell inside the window, the median and mean move, and what the last 3 passes did — so a pattern that died recently cannot hide behind its long-run average.

Paint zones on chart washes every occurrence of the windows — past and future — onto the time axis in green and red, labeled with the historical record ("▲ 9/13"). Future occurrences come from the cycle's schedule, but they never count toward the statistics: the label is always history only.

Indicator Odds

Indicator Odds measures how a familiar oscillator reading has actually resolved on your symbol. Pick an indicator — RSI, Stochastic %K, Bollinger %B, CCI, Williams %R, ROC, or Z-Score — and the lab bins its historical values and reports how often the next bar closed up from each bin, side by side with the symbol's base rate, plus a z readout showing how far each bin's sample sits from that base rate. The current bar is excluded from the sample, so the bin you are standing in never grades itself.

Read the table as a historical tendency, not a forecast: a bin that leaned one way in the past carries no guarantee forward, and thin bins deserve little weight no matter how extreme their rate looks.

Read the labs responsibly

  • Every forward segment is a scenario to compare with price, not a promise. Validation numbers describe your own symbol, settings, and date range.
  • Prefer selections and settings that survive their own checks — the wavelet holdout, the tuner grid, the consensus share, the walk-forward table — over ones that merely look convincing on history.
  • The labs share history with each other and with the Neural Forecast Lab and Forecast Composer; agreement between them is not independent confirmation.
  • Regime changes, news, and gaps break fitted scenarios immediately. Rerun after them.

Plan access

The eight labs, the Wavelet Projection, Seasonal Projection, and Cycle Tuner overlays, the Cycle Odds Zones panel, and the consensus filter require Pro. The Swing Projection Zones indicator itself remains available from Basic; the lab's audit and provisional zones are the Pro layer on top.

Related guides