Insights

Open Insights and the market state and your own trading state sit on one screen: what the regime is, where funding is stretched, which carry trades survive their fees, how DEX volume is split, what you paid, what the news says — and a brief written from those blocks. It is a reading surface. Nothing on it places, sizes or proposes a trade.

Most of it is a snapshot of the current moment, with the window you choose setting how far back the history goes. The default is 30 days; anywhere from 1 to 90 is allowed.

What the page shows #

The brief is deterministic on purpose: the same inputs produce the same text, so every sentence in it can be traced back to a block. It states what the blocks say; it does not add a view of its own.

Reading funding APR #

Funding is the periodic payment between longs and shorts that keeps a perpetual near spot. When the rate is positive, longs pay shorts. When it is negative, shorts pay longs.

Venues quote the rate per funding interval — commonly eight hours, but confirm the interval on the venue you are reading. Insights annualises the rate so pairs with different intervals can be compared on one scale:

APR = rate per interval × intervals per year

Three cautions apply to every rate on the page:

• It is a snapshot. Funding resets each interval and can change sign within a day.
• A high positive rate is crowded positioning, not a forecast. It says longs are paying

to stay long; it does not say price must fall, and extremes can persist for days.

• The annual figure assumes the rate holds all year. It never does exactly. Use APR to

compare opportunities, not to predict income.

The carry board, fees included #

Carry means holding a position to collect funding rather than to trade a price move. The board ranks pairs by the funding you would collect over the window, minus the cost of getting in and out.

Fees are the part most funding screens leave out, and they are usually what decides the trade. A rate that looks generous on a small position can be wiped out by the round trip. Set the notional to your real size and the fee to what you actually pay, and the ranking reflects your costs rather than a headline rate.

Even then, carry is not risk-free:

• You still hold the perp, so price and basis can move against you.
• Funding can flip sign while you are in the position, turning income into a cost.
• The board ranks funding net of fees, not safety or liquidity.

Treat a positive ranking as "this would have paid funding after costs over this window", not as an arbitrage.

Your own fee drag #

Fee drag is a different question from the carry board. The board asks what a trade might pay; fee drag reports what your trading already cost: total fees over the window as a share of gross PnL before those fees.

A high figure is not a verdict on its own — a fast mechanic can pay a lot in fees and still be profitable, and a slow one can be ruined by a single bad exit. The useful comparison is the one on this page: average win against the round-trip fee at your usual size. If the win is smaller, the mechanic cannot pay, and no amount of screen time fixes that.

If you pass a notional and a fee rate, the same inputs also drive the carry estimate, so the two blocks describe the same account rather than two different ones.

What "unknown" means #

A block can read unknown when it cannot be computed from available data. Typical reasons:

• no connected venue publishes funding for the pairs in scope,
• the window contains no fills, so there is no fee drag to report,
• the notional or fee input is missing, so a carry estimate would be a guess,
• a venue's data is older than the freshness threshold.

Unknown is not zero, and it is not a failure of the page. Whatever is missing is listed in gaps alongside the summary, with the block it belongs to. Read gaps before quoting a number as if the page were complete — and never read an empty block as "no cost" or "no exposure".

Pulling the same summary #

From an AI client #

get_insights returns the same blocks as one payload. It is read-only and needs mcp:read.

{ "days": 30, "notional_usd": 5000, "fee_bps": 5 }

Blocks that could not be filled come back listed in gaps rather than being dropped or filled with a placeholder. A model reading the response should cite the gaps with the numbers, for the same reason you would.

From the CLI #

jw3 insights --days 30

The same summary over the same window, in the terminal. It reads only — no order, no proposal, no write to the diary — which makes it safe in a morning cron job or a review script.

What it does not do #

• It does not place, size or propose a trade.
• It does not predict regime changes or funding reversals.
• It does not read keys or touch custody.
• It does not fill a missing block with an estimate; it reports the gap.

Read it as a starting state for the day, then check anything you intend to act on against the venue itself.