Ask your commodity
data a question.

A research assistant that runs the statistics for you. Ask whether crop conditions actually predict yield, or where managed money sits versus its own history — it computes the answer across 25 years of public data and draws the chart.

Try the demo →
No account. No email. The weekly reports are free too.
CFTCUSDAEIA NOAAERA5DROUGHT MONITOR
The problem

Everything you need is public. That's exactly why it's unusable.

Six agencies, six schedules CFTC on Friday, EIA on Thursday, WASDE monthly, NASS weekly in season. Assembling one comparable history is a project, not an afternoon.
Your broker shows price, not why Charts and quotes are everywhere. The positioning, the conditions, and the balance sheet that move them are not.
Your AI can't reach the numbers It follows the news well and confidently invents the data. Ask it for a correlation and you get a plausible number from nowhere.
What it does

A real question, and the real answer.

This is an actual exchange against the live data — not a mockup. Every number below is reproducible from the same public sources listed at the bottom of this page.

chat.getsoftsignal.com
Does corn condition at silking actually predict the US yield, or is that folklore?

It predicts it, and it's the strongest relationship in our registry. Pairing the USDA national poor + very poor share during the July silking window against the WASDE US corn yield, 2008–2026:

r = −0.90 in year-over-year differences (n = 18, p < 0.0001). In levels it's r = −0.73. Differences are the more honest read — yield trends upward with genetics, so levels partly measure that trend.

2009 · condition −2.5 pp · yield +10.8 bu/ac 2010 · condition +1.5 pp · yield −11.9 bu/ac 2011 · condition +2.2 pp · yield −5.6 bu/ac 2013 · condition −26.9 pp · yield +35.0 bu/ac 2014 · condition −4.5 pp · yield +12.9 bu/ac 2015 · condition +3.5 pp · yield −2.6 bu/ac 2016 · condition −3.6 pp · yield +6.2 bu/ac 2017 · condition +5.6 pp · yield +2.0 bu/ac 2018 · condition −2.8 pp · yield −0.2 bu/ac 2019 · condition +4.3 pp · yield −8.9 bu/ac 2020 · condition −5.0 pp · yield +3.9 bu/ac 2021 · condition +1.8 pp · yield +5.3 bu/ac 2022 · condition +2.6 pp · yield −3.3 bu/ac 2023 · condition +2.4 pp · yield +3.9 bu/ac 2024 · condition −4.8 pp · yield +2.0 bu/ac 2025 · condition −3.5 pp · yield +7.2 bu/ac 2026 · condition +3.5 pp · yield −3.5 bu/ac 2012 drought · condition +25.2 pp · yield −24.1 bu/ac 2012 drought Δ poor + very poor at silking (percentage points) Δ yield (bu/ac) +40 0 −30 −30 0 +30
Each point is one crop year, 2009–2026. Worse conditions at silking, lower yield — with 2012 as the extreme case: conditions 25 points worse, yield down 24 bu/ac.
Year-over-year change in corn poor+very poor condition at silking vs change in US corn yield
YearΔ poor+very poor (pp)Δ yield (bu/ac)
2009−2.5010.80
20101.50−11.90
20112.15−5.60
201225.20−24.10
2013−26.8535.00
2014−4.5012.90
20153.50−2.60
2016−3.556.20
20175.602.00
2018−2.80−0.20
20194.25−8.90
2020−5.003.90
20211.755.30
20222.60−3.30
20232.403.90
2024−4.752.00
2025−3.507.20
20263.50−3.50
Two things to hold against this. The overlap is capped at 19 years by the WASDE yield history, not by the condition data. And the same measure one month earlier — June, before yield is set — returns nothing on the same data and the same method (r = +0.27, p = 0.29). That negative control is what separates this from a number found by looking hard enough.

Ask it your own question →

What you can ask for

Four things, on any series we carry.

Correlate

Any two series, levels or differences, with the sample size and p-value attached. It will tell you when the answer is nothing.

Chart

Every result comes with the plot. Ask for a different window, a different pairing, or a seasonal composite and it redraws.

Rank & compare

Where a number sits against its own history — this week's managed-money net as a percentile since 2006, not just its level.

Connect your own AI

The same data over MCP and REST, unlimited. Point Claude, ChatGPT, or your own scripts at it and skip our interface entirely.

Pricing

The reports are free. The assistant is the product.

Every weekly report and supplement stays free to read, with no account and no email. You're paying for the layer that answers questions the reports don't.

Reports
$0
No account · no email
  • Energy · Grain · Coffee · Livestock, weekly
  • ENSO, WASDE, positioning and weather supplements
  • Read them yourself, or point your AI at any page
  • Quote and cite freely with attribution
Read the reports →
Using this for client work? Commercial is $249/mo — commercial-use license, citation rights, and 100 queries a day. Subscribe →
No trade signals, no proprietary models, no market calls. The data and the statistics are the product — you bring the interpretation.
Sources & reliability

The data is public. Keeping it honest is the product.

Every number traces to a public, authoritative source. Those feeds move on their own schedules, change formats, and occasionally go quiet — so every series publishes how often it should update, when it actually last did, and a live on-time / behind / stale flag. See live status →

CFTC Disaggregated COTWeekly positioning — producers, swap dealers, managed money. Fridays. Why it matters →
USDA NASS & WASDECrop progress and condition weekly in season; supply & demand balance sheets monthly. Crop calendar →
EIANatural gas storage, Henry Hub spot, dry gas production, power burn, LNG exports.
NOAA / GFS16-day forecast, population-weighted HDD/CDD.
ERA5 reanalysis30-year climate normals on the WMO 1991–2020 baseline.
US Drought MonitorWeekly D0–D4 severity by crop region.
Questions

Frequently asked

Are the reports really free?

Yes — all four weeklies and every supplement, no account and no email address. The subscription buys the research assistant and programmatic access to the data underneath, not the reports themselves.

What stops it from making numbers up?

The assistant can't see raw data and can't do arithmetic on it. It calls a statistics engine that computes server-side and returns the result plus its caveats, which it has to report verbatim. If a relationship isn't there, the answer comes back null — as the June control in the example above does.

What does a query cost me?

Individual includes 25 a day. A full statistical analysis draws more than a simple lookup, so a day spent entirely on heavy correlations lands closer to ten. The counter resets at midnight UTC.

Do I need to set up MCP?

No. The hosted assistant runs in your browser and needs nothing installed. MCP is there if you'd rather drive the same data from Claude, ChatGPT, or your own code — it's included, unlimited, on both tiers.

Do I need a background in futures?

No. The reports explain what each number is as they go, and the assistant answers in plain language. Knowing your way around a balance sheet helps, but nothing here assumes a trading desk.

Can I share reports with others?

Freely — they're public. Your API key and assistant login are single-user. Redistribution, meaning serving our data to another product's users, needs a separate agreement on any tier.