Strip: the same positioning, measured in dollars at risk

[mspinola]

The printed positioning reports this page was modelled on draw each position twice, once in contracts and once in dollars. This adds that second reading to the strip as an optional mark, and the measurements that decided its shape are in docs/analysis/2026-08-24-contracts-against-dollars.md with a reproducer script beside it.

What it draws

A new COMPARE control picks the one reference mark each row carries: 6w ago (today's hollow ring, still the default), Dollars, or None. Under Dollars a hollow diamond marks the same 0-100 range index, over the same window and on the same leg, computed on dollar risk (contracts x point value x price x daily volatility) instead of on contracts, with a hairline back to the contract reading. It is a range position like the lollipop, not a percentile; the percentile version of this question belongs to /exposure.

Why risk and not notional

Holding market, leg, window and statistic constant and changing only the unit, across 44 markets:

pair	correlation	p95 gap, index points	different band
contracts vs risk (Raw PF 5/95)	0.917	30.4	11.7% of weeks
contracts vs notional	0.979	14.5	5.4%
share of OI vs risk (NPF 20/80)	0.896	33.9	20.7%

Over a rolling window notional is contracts times a slowly-moving price, so a second mark for it would sit on top of the first. It is one line in the hover instead, because it is the reading those reports plot.

On 2026-08-18, 13 of the 45 priceable markets land in a different band under the two lenses. Silver is the case it was built from: Commercials at the very bottom of their 24-week contract range and at 96 on dollars at risk, because daily volatility fell from 6.7% to 2.7% across the window, so a record short carries a third of the money. Natural Gas runs the other way, 99 against 9.

The gap is not simply "volatility is high": correlated 0.25 with the market's own volatility index at the median and sign-flipping across markets (33 positive, 11 negative), because volatility acts on a position that has a side. It is a state rather than a jitter (lag-1 autocorrelation 0.918).

Design notes

• One reference mark, not a fifth mark. The ring and the diamond ask the same shape of question and at 22px they are one smudge together. The control also lets both be turned off, which the page could not do before.
• The mark takes the row's colour. Every palette slot is already a verdict here (grid_colors builds bull from slot 3, which is also Price, and bear from slot 0, which is also Commercials), so shape carries which comparison it is and colour goes on meaning verdict.
• The connector stays, unlike the prior mark's, because here the gap is the subject and the ink is self-limiting: a market whose lenses agree draws a line of zero length.
• Three bands, not a boolean. Silver's 0-against-96 is both ends of the axis at once, and asking only whether each reading is through a gate scores that as agreement.
• Off by default, and cached per market by the store's newest date: about 4s cold for the universe, 0.3s warm.
• A share of open interest is the same number in contracts and in dollars to 2.2e-16, so there is no dollar version of the NPF basis; the caption states the window and the unit rather than implying only one thing changed.

Testing

464 passed, 17 of them new, plus ruff check src tests clean. Verified in the browser at one and two columns under both models: marks, connector, hover, legend key, caption counts, and no console or server errors.

🤖 Generated with Claude Code