OI/Depth Fragility — Signal Playbook

What it detects

Open positions dwarf displayed exit liquidity: open-interest notional (OI × price) versus the resting book notional at the selected depth. When crowded positioning has to exit through a thin book, the book loses.

How OI/Depth Fragility paints across a heatmap row — the colour language it speaks on the live map.

How VYX surfaces it

VYX evaluates OI/Depth Fragility live on every candle across 300+ Hyperliquid markets and paints it onto the heatmap, so an unusual impact reading stands out at a glance instead of hiding in a watchlist. That is the job it does in the workflow: it is a scan filter and a piece of context, not an automatic entry. Scan the map for where OI/Depth Fragility is firing, focus that market to read the signal against price and the confirmation cues below, and let the next candle or two resolve before acting on it.

How to play it

Use it as scanner context, not an automatic entry:

• Read it as stored energy: nothing here says when — only how violent a forced unwind could be.
• Combine with funding and OI trends — crowded, paying-to-hold positioning on a thin book is the squeeze recipe.
• Expect liquidation cascades to overshoot on high-ratio pairs — forced exits consume the whole visible book.
• On majors the ratio is structurally high; per-symbol scoring (unusual for itself) is the useful lens.

Confirmation

What strengthens the read:

• Funding is stretched or OI is still climbing while depth stays thin.
• Small liquidation runs already move price disproportionately.

Invalidation

What kills it:

• OI unwinds gradually or depth builds — the imbalance resolves without an event.
• The book deepens when tested (passive size steps in).

Risk & honest evidence

Respect the limits:

• A condition, not a timing signal — fragile can stay fragile for weeks.
• Displayed depth understates true absorption; market makers refresh quotes as levels trade.
• Evidence: grounded in market-microstructure research. VYX's own live backtest is still accruing — treat this as scanner context, not a proven edge.