Trading Signals
Each indicator your agent reads casts a vote. You decide whose vote you trust. The weighted average is the agent's conviction — the single number that decides whether it acts.
The Signals tab is where you turn the dials on every individual indicator your agent reads. This page explains what those dials do and how the agent's conviction is built from them.
How Indicators Vote
On every market read, every indicator your agent watches asks itself two questions:
- Did I see something worth voting on? — yes or no.
- How strongly? — a score from
0.0(barely) to1.0(extreme).
The "how strongly" part scales naturally to whatever threshold you picked. The score isn't "did it cross the line" — it's "how far past the line did it go."
If an indicator didn't see anything, it sits the vote out. It doesn't count as a zero — it just doesn't show up at the ballot box.
The Weight Slider
For every indicator, you have a weight from 0 to 10. This is the dial you actually turn. It answers one question:
How much should this indicator influence my agent's conviction?
| Weight | Meaning |
|---|---|
| 0 | Show me when it fires but don't let it influence the score. Pure context. |
| 1 | Standard background voice. |
| 2 | Default for indicators in a category you've turned on. |
| 5–10 | "I really trust this. Let it dominate." |
Weight Changes the Vote Count, Not the Score
The most common mistake: thinking weight makes an indicator's score bigger. It doesn't. A weight of 10 doesn't turn a 0.40 score into a 4.0 score.
Think of it as megaphone size:
- The indicator shouts
"0.40!"— its score is unchanged. - You gave it a size-4 megaphone, so the room hears that 0.40 as if four voices were saying it.
- Another indicator shouts
"0.33!"through a size-1 megaphone — one voice. - Your agent's conviction is the room's average, weighted by megaphone size.
No indicator ever screams louder than it actually feels. You're just deciding which voices the agent listens to.
What Conviction Decides
After all the votes are in and the weighted average is computed, the agent runs three checks before any trade can route:
- Can a valid trade even be built? The agent needs current volatility data to calculate sane stop and target levels. If volatility data is briefly unavailable, the read is skipped.
- Did conviction clear your threshold? Your minimum-conviction setting is the floor. Below it, the agent stays on the bench.
- Did enough top-tier indicators fire? Your minimum-primary-count setting requires at least N indicators from your enabled categories to actually vote. A high score from one loud indicator isn't enough — the agent wants corroboration.
Pass all three and the trade routes — either to your approval queue (Signals Only) or straight to the exchange (Trading).
What Else the Agent Sees
The vote tally isn't the only thing the agent considers. For every trade decision, it also gets a per-coin context snapshot — the current state of every indicator category you've enabled on the Strategy tab, whether each one fired this cycle or not.
This is the part most agents are tuned on without realizing it. An indicator can fail to fire but still inform the decision:
- An indicator that fired contributes its weighted vote and shows its raw state in the snapshot.
- An indicator that's enabled but didn't fire sits out the vote but still shows its raw state in the snapshot.
- An indicator that's disabled contributes nothing and isn't shown at all.
So a weight of 0 silences an indicator in the vote tally but keeps its raw state visible to the agent. Useful when you want the indicator to inform reasoning without dragging the score around.
Why This Matters
Two agents with identical signal settings can make different decisions on the same trade — because one has more indicator categories enabled on Strategy, and the agent reads the additional context.
Example. A momentum signal fires on SOL with moderate conviction.
- Agent A only has momentum enabled. The agent sees the signal, sees the pre-validated stop/target options, and enters.
- Agent B has momentum, funding rates, and higher-timeframe trend enabled. The agent sees the same signal — but also sees funding sitting at "elevated" (longs are crowded) and the higher timeframe agreeing with the move. Agent B might enter with smaller size, or skip if the funding warning outweighs the rest.
Same signal, same setups, different decisions — because Agent B saw more.
If you want an indicator to inform your agent's reasoning without letting it dominate the score, enable it on Strategy and set its weight to 0 or 1. The agent still reads the state; the vote just contributes little or nothing.
Default Weights
When you create a new agent, every indicator is automatically seeded:
- Indicators in enabled categories → weight 2
- Indicators in disabled categories → weight 1
The seed weights don't lock anything down — they're just the starting point. Once seeded, your per-indicator weights are yours, independent of which categories you toggle later.
The only exceptions: switching strategy presets and toggling category visibility on Strategy both reseed weights to the new defaults. The UI shows you exactly which indicators will be affected and asks for confirmation before the reseed happens.
Indicator Categories
Indicators are grouped into categories — each one covers a different read on the market:
| Category | What It Watches |
|---|---|
| RSI | Overbought and oversold zones, plus divergences. |
| MACD | Bullish and bearish crosses, plus divergences. |
| Stochastic | Overbought, oversold, and crosses. |
| Volume | Surges, dry-ups, on-balance divergences. |
| Volatility | Expansion and contraction phases. |
| Bollinger Bands | Squeezes, band touches, range extremes. |
| Moving Averages | Stack alignment, crosses, position relative to long-term mean. |
| Trend Strength | Trending vs. ranging conditions. |
| Funding | Extreme funding rates and rate flips. |
| Open Interest | Surges and divergences. |
| Relative Strength | Performance vs. peers and rate-of-change. |
| Support & Resistance | Touches and breaks of key levels. |
| Money Flow | Inflows, outflows, and divergences. |
| Cross-Market | Sector divergence, BTC decorrelation, sector rotation. |
| Regime | Trend, volatility, and momentum regime shifts. |
| Price Structure | Fair-value gaps, order blocks, zone clusters. |
| Order Flow | Buy-sell pressure imbalances and divergences. |
Each category contains indicators of two kinds:
- Tunable — has one or more numeric thresholds you can adjust.
- Structural — pure pattern detectors with no thresholds. You can still re-weight them; you just can't re-threshold them.
The Signals Tab
The tab shows a collapsible accordion grouped by category. Each row has:
- Indicator name and description — tap the info button for the full explanation (what it detects, how it fires, examples).
- Weight slider — 0 to 10.
- Threshold inputs — only shown for tunable indicators.
Changes save immediately. Categories you've customized are flagged in the accordion header so you can see at a glance which ones you've touched.
Reset to Defaults
A two-click Reset All button at the top of the section re-seeds every indicator back to the defaults for your current category selection. A confirmation dialog protects you from accidental wipes.
When Switching Strategy Wipes Your Tuning
Switching your strategy preset, or toggling category visibility on the Strategy tab, reseeds every weight and threshold to the new defaults. This overwrites your tuning.
The UI shows a confirmation dialog before any reseed so you can cancel. Behavior-only and persona-only edits never reseed — only an explicit strategy switch or a category toggle does.
The reseed is full-replace. If you've spent time tuning individual weights, switch strategies carefully — you'll lose those edits.
Historical Reads Are Frozen
Every recorded signal stores the exact weights and thresholds that produced it. You can review a six-month-old signal and see "this indicator fired at score 0.40 with weight 4" — even if you've since retuned that weight to 10.
Your agent's performance history reflects what it actually believed at the time, not its current configuration.
TL;DR — The Mental Model
Conviction is your agent's confidence vote. Each indicator that fires casts a ballot (how strongly it sees the setup), the weight slider decides how loud that ballot is, and conviction is the weighted average of every ballot at the table. The single number decides whether the agent acts or stays on the bench.
- Score — how strongly an individual indicator is screaming.
- Weight — megaphone size. Vote count, not score amplifier.
- Conviction — the room's overall agreement, weighted by who you trust.
Safest starting point: leave every indicator at its default weight, run your agent in Signals Only for a few days, then review which indicators consistently showed up on signals you would have approved. Bump those up. Drop weights on the ones that fired on setups you'd have skipped. Tune iteratively — small adjustments give you visibility into what's actually moving conviction.