Real questions. Full answers.

Margin expanded 290bp in four quarters while shorts grew 20% in a single month — and if you want to act on that, 24 listed expiries run out to January 2029. A data vendor, a US regulator, an options exchange and a central bank, in one call shape. No one of them sells the other three.

Two of the four numbers came from institutions that do not sell data at all — FINRA and the ECB publish, they do not license. That is the part no vendor contract replaces.

The AI-capex story sits in two numbers — 18.2× EBITDA but 56.3× free cash flow, because capex absorbs 63% of operating cash flow at 2.7× depreciation. The question that finishes it lives in another domain entirely: a 1.83% free cash flow yield against a 5.01% ten-year is minus 318 basis points. You are paid in growth or not at all. An equity-only API cannot ask that question.

Capex at 2.7× depreciation cannot hold indefinitely: either depreciation catches up and the multiple compresses, or revenue absorbs it. Either way the spread to the risk-free rate is the number to track — not the EBITDA multiple.

Bull: 28× earnings, +105.9% revenue, 59.5% ROIC, 2.4% capex intensity. Bear: income quality of 0.70 and 150 days of inventory say cash is arriving slower than earnings. The tell: at 2.14 days to cover almost nobody is positioned for the bear case, even though 24 listed expiries make it trivial to express. Four institutions built that argument; the agent didn't get to choose which one flattered the thesis.

For a desk the actionable line is the last one: with 24 expiries and 2.14 days to cover, hedging the bear case costs option premium, not liquidity.

The valuation ranking and the positioning ranking disagree, which is the whole point of joining them: Micron is the cheapest name on the fastest growth and the least shorted at 1.16 days to cover, while shorts grew 12.2% in a month on Intel — the only name in the group with no trailing earnings. One vendor ranked them; a regulator said who was actually leaning.

Positioning is the column that changes the trade. Cheap-and-unshorted (Micron) and expensive-and-shorted (AMD) are different setups, even when the valuation screen ranks them side by side.

Four institutions, one calendar: a data vendor for earnings, a macro provider for prints, a Federal Reserve bank for the realised claims series, and a prediction market for what is already in the price. The street expects durable goods to swing from +1.1% to −0.5% and payrolls to fall by a quarter — while real money pays 54.5% for a hike. One of those two views is wrong, and you can see both in a single response.

The Micron print on 30 September lands the same morning as core PCE. If both surprise in the same direction, the curve moves before the equity does.

Institutions hold 14.47bn shares and shorts hold 298M — a 48-to-1 long-to-short ratio with breadth on a knife edge at 51%. The vendor gave you the crowd; the regulator gave you the absence of a counterweight. You never touched an EDGAR parser.

Read with rung 03: a 48:1 long-to-short ratio on 51% breadth is the profile of a name that gaps on flow, not on fundamentals.

CPI bottomed at 2.33% (Apr 2025), peaked at 4.17% (May 2026) and sits at 3.35%, while unemployment fell from 4.5% to 4.1% and claims dropped to 196k. Two independent agencies agree on the index to the third decimal — and FRED carries realtime_start on every row, so a backtest can ask what was known on the day rather than what is believed today.

For a backtest this matters more than the level: the FRED vintage field reconstructs what an analyst could have known on any given day, with no look-ahead.

The front end sold off nearly twice as hard as the long end, and equity vol gave the whole move back inside a week.

Bear flattening with vol collapsing is the market saying the hike is priced and not feared. Whether that is right is a separate question — see rung 12.

In seven days the market flipped from 62/38 in favour of holding to 55/44 in favour of hiking, on $7.1m of combined volume and $2.3m of resting liquidity — while the 2-year sold off 37bp in the same window. A prediction market and the Treasury curve, two institutions, telling the same story independently. That corroboration is the trade; assembling it by hand is an afternoon.

The move is large relative to liquidity: $2.3m of resting orders absorbed a 17-point swing. Worth knowing before treating the probability as a consensus.

JPMorgan earns 159bp on assets against Citi's 106bp — and the market pays 2.60× book versus 1.06× for that gap. The regulator and the market agree on the ranking. FDIC data needs no vendor contract, no licence and no key; it is simply in the catalog.

The pairing is the point: the regulator's ROA says how the bank earns; the market's P/B says what investors make of it. When the two disagree, that is the diligence question.

A $311.9bn float sitting mostly in short-dated Treasuries earns roughly $12.9bn a year at a 4.14% three-month bill — which is why the front end and stablecoin supply are now the same trade seen from two sides. And with perp funding at 8.04%, the leveraged long pays almost double what the collateral earns. Four vendors: an on-chain analytics firm, a market aggregator, a rates source and an exchange connector. No equity data vendor sells any of this.

It also means the front end is now partly a crypto-demand market — a $30bn change in stablecoin float is a $30bn change in bill demand, visible through these same routes.

The regime is cheap spot vol with expensive tails. VIX and VVIX both sit below where August started, but SKEW is 3% above it and set its high on 11 September — after the VIX spike had passed. A VVIX/VIX ratio of 5.90 is the month's highest: convexity is bid even though the level is not. Put that beside a front end repricing 37bp and prediction markets flipping to a hike, and the classification writes itself — this is not calm, it is unhedged.

Practically: protection here is cheap on level and expensive on skew, so the efficient hedge is a spread rather than an outright put.

Read together: prices re-accelerating into a tightening labour market, a front end repricing hawkishly, speculators structurally short duration, real money paying 54.5% for an October hike — and equity vol asleep at 14.8 while sentiment sits in fear. That is a regime where the surprise is not the hike; it is that nothing is hedged for it. Assembling it from seven institutions is a sprint. Here it is one prompt, ~11 seconds, one key — and every figure still names the source that produced it.

Every figure above carries its request_id, so the packet can be re-run next week and diffed line by line.