|
TradeWPower
Energy-Weather Intelligence
|
18 July 2026
DIANA WEATHER
|
DIANA Weather — Client Overview
Step 1 of the TradeWPower Trade Cycle
Step 1 of the TradeWPower Trade Cycle
DIANA Weather
Turning 45 years of European energy-weather history into daily decisions for traders, hedgers and production planners.
18 July 2026
D
Digital
I
Intelligent
A
Artificial
N
Numerical
A
Advisor
The Advisor role
DIANA makes each user better at their job
The numbers, the pattern-matching, the daily rebuild — DIANA does that work so the desk doesn’t have to. What comes back is time and clarity: time for traders, hedgers and planners to focus on strategy and risk management, and clarity about where conviction is high enough to act. The compounding effect is simple — fewer losses, more winning trades, and bigger wins when the setup is right.
15
Years
The Proprietary Edge
Built on 15 years of research outside the public domain
DIANA is not a re-skin of open-source weather models and not a wrapper on third-party AI trading data sets. It is 15 years of original European energy-weather research — discoveries, relationships and decision rules you will not find in published papers, public APIs or commercial training data. That is the edge.
The proof, in one chart
From Week 1 out, DIANA leads ECMWF — +11 to +22 percentage points at Weeks 2–5
Comparing DIANA against the ECMWF energy-weather forecast — the commercial product utilities and traders use on the desk. Both judged the same way: did they call the direction of observed weather correctly? EC owns Week 0 — by the finest of margins. From Week 1 the ensemble starts converging on climatology. Weeks 2 to 5 are where positions are built, and that is exactly where DIANA’s edge is widest.
|
Where the trade lives |
|||||||||||||
|
50%
|
66%
|
67%
|
68%
|
60%
|
69%
|
54%
|
65%
|
54%
|
70%
|
51%
|
70%
|
48%
|
coin
flip |
|
W+0
−0.4pp
|
W+1
+8.2pp
|
W+2
+15.0pp
|
W+3
+11.0pp
|
W+4
+19.4pp
|
W+5
+21.7pp
|
||||||||
Extreme weeks (recall — the week’s observed anomaly) · |obs anom| > 1σ · where prices move
Week 3 · 3 weeks out
54%
ECMWF
Week 4 · 4 weeks out
50%
ECMWF
Week 5 · 5 weeks out
47%
ECMWF
Week 3 · 3 weeks out
79%
DIANA Weather
+25pp vs EC
Week 4 · 4 weeks out
86%
DIANA Weather
+35pp vs EC
Week 5 · 5 weeks out
87%
DIANA Weather
+40pp vs EC
Methodology: 2010–2024 standard backtest — 50 market × variable targets (Temp / Precip / Wind / Solar plus hydro precipitation, four main markets + Nordic sub-areas), 62,511 predictions in total. Head-to-head vs the ECMWF energy-weather forecast (daily Wind / Solar / Temp values per market — the commercial product utilities and traders use, not gridded NWP output) on the weeks where the EC forecast archive overlaps (2020–2024): ≈2,300 comparisons per lead across 36 Wind / Solar / Temp targets. Both scored against the same observed weekly direction. Basis note: DIANA is scored in-sample on its production atlas (fully out-of-sample walk-forward: 54.7% overall vs 68.9% in-sample); the EC leg is true out-of-sample.
1 · The TradeWPower Trade Cycle
TradeWPower is building a complete decision chain for European power markets. DIANA Weather is the first step — live today. Three further steps follow, and together they will feed every report, every hedge decision and every production plan through a single connected workflow.
|
1
|
2
|
3
|
4
|
|||
| DIANA Weather | DIANA Price Model | DIANA Markets | DIANA News | |||
|
LIVE TODAY |
COMING NEXT |
ON ROADMAP |
ON ROADMAP |
|||
| Energy-weather outlook | Stack model & curves | Funds & flow impact | News & geopolitics | |||
| W+0 → quarterly | Weather → price translation | Positioning & sentiment | Risk events beyond fundamentals |
WHAT DIANA DOES FOR YOU
Frees up time. Raises performance.
The numbers, the pattern-matching, the daily rebuild — DIANA does the work for you. Traders, hedgers and planners get their hours back for strategy and risk management, not data-wrangling.
A tool that gives you a significantly higher probability of knowing when to enter, when to exit, and how to scale the position. The result: better risk-reward, higher Sharpe ratio. Fewer losses, more winning trades, and larger profit on the trades that work.
Reports R0 through R5 are the visible face of Step 1. As Steps 2, 3 and 4 come online, the same reports will carry price direction, market positioning and news-driven risk context — one chain, one vocabulary, one place to make trading decisions. Six connected reports, delivered every morning before markets open. Each report answers one specific question and hands its output to the next. The desk opens six tabs and has a complete view — not six vendors to reconcile.
THE BIGGER PICTURE
TradeWPower is developing a complete decision-making model for European power markets. DIANA Weather is the first step — live today. Three more pillars will be integrated into a fully developed trading model, with scenarios, probabilities and risk-reward built and analysed continuously.
2 · What DIANA Weather Delivers
Inside Step 1, three things make DIANA different from another weather or fundamentals feed: conviction on every call, 45 years of outcomes backing each signal, and a daily learning loop that keeps the edge current as European markets and climate evolve. The rest of this section unpacks each.
The Edge
Conviction on every call
Every direction is tagged High · Medium · Low. You know when to size up and when to stand down — before the week starts.
The Edge
Grounded in 45 years of outcomes
Not a physics toy. The analog atlas is built on four and a half decades of what European markets actually produced — wind, temperature, hydro, solar — and every call is scored on a 2010–2024 backtest.
The Edge
Fresh every morning, compounding every day
Yesterday's truth becomes today's edge. The research loop retires signals that fade and promotes signals that emerge — automatically.
The Daily Chain
|
R0
|
R1
|
R2
|
R3
|
R4
|
R5
|
|||||
| Status | New Normal | Analogs | Hydro | DIANA | Trade |
R0
Operational Status
“Did everything land clean this morning?”
Edge ·
Full transparency on every input and every report. If anything's stale you see it before 9:00, not after a bad trade.
R1
The New Normal
“What should ‘normal’ even mean in today's climate?”
Edge ·
Live 35-year baselines — not a frozen 1991-2020 reference that already lags reality by a decade. You trade against the climate you're actually in.
R2
Analog Years
“Which past winters, springs or summers match today's setup?”
Edge ·
Best-matching years ranked by atmospheric similarity, with their realised outcomes already attached — not calendar-distance guesswork.
R3
Hydrology
“Where is the water going — near-term and seasonal?”
Edge ·
Short-term balance plus a six-month reservoir-fill trajectory in one view — not a dashboard you have to assemble.
R4
Multi-Week Outlook
“Where does the next six weeks go, per market, per variable?”
Edge ·
Direction and conviction for every market and every horizon W+0 to W+6, with forecast-flip detection when the atmosphere is about to pivot.
R5
Trading View
“Where do I actually take risk this week?”
Edge ·
Trade / Lean / Watch flags per market, driven by the conviction tier. High-tier hit rate: 73% (2010–2024 backtest, in-sample).
3 · Beyond Human Capability
European power markets sit at the intersection of stratospheric dynamics, jet-stream position, sea-surface temperatures, ocean oscillations, sea ice, snowpack, hydrology, solar irradiance and more than a dozen market micro-structures. A skilled analyst can hold a handful of these relationships in their head at a time — maybe a dozen on a good day. Every one that falls out of the mental model is a risk that goes un-priced.
DIANA is built to carry the load. Every day it scans every combination that has ever mattered across 45 years of observations, scores the signals that fire, cross-checks them against actual market outcomes, and collapses everything into a conviction-tagged call the desk can read in seconds.
What one morning can hold
|
A skilled analyst
a handful
drivers tracked in parallel
most of the signal, lost to bandwidth
|
DIANA, every morning
107,400
weather-energy relationships scored
distilled into 560 conviction-rated calls
|
The point isn't the raw number. It's that everything a human wishes they could track — cross-market dependencies, seasonal reversals, sub-area splits, trapped-low configurations, lead-time decay — is scored, weighted and fed into the same daily output. Nothing is left on the table because the analyst ran out of screen space.
The numbers
45+
Years of weather history
Every European winter since 1980, continuously rebuilt.
13
Power markets covered
Nordic price areas plus Germany, France, Spain.
4
Energy variables per market
Wind, temperature, hydro precipitation, solar — coupled, not isolated.
107,400
Weather-energy relationships scored
Every driver, every month, every outcome — pre-computed and ready.
560
Conviction-rated calls every day
Direction plus High / Medium / Low tier for every market and horizon.
W+0 → Q
Full horizon range
This week, next week, month-ahead and quarterly outlook — one system.
4 · The DIANA Score — Track Record on Extremes
WHERE THE MONEY LIVES
The edge isn't being right every time.
It's knowing when we're right — and when we don't know.
It's knowing when we're right — and when we don't know.
Every DIANA call ships with a score: HIGH, MEDIUM, or LOW conviction. High conviction calls are trade-grade. Low conviction calls say watch, don't commit. That calibration is what separates DIANA from a forecast that's just a number.
| Σ |
MATH, NOT OPINION
Every number below is measured on a 2010–2024 backtest (in-sample), against a 45-year analog atlas. No analyst judgment. No trader selection. No hindsight cherry-picking.
|
THE DIANA SCORE — EXTREME-WEEK HIT RATE BY MARKET
Strong edge across all four main markets
on the weeks that move prices.
on the weeks that move prices.
Headline number = extreme-week hit rate on Wind. Temp, Precip and Solar shown underneath. Hit rates are conditional on the week actually landing in the extreme (recall) — 2010–2024 backtest, in-sample.
NORDIC
87%
EXTREME WINDY · WIND
+17pp · 126 wks
| Temp warm | 89% |
| Precip wet | 86% |
| Wind windy | 87% |
| Solar sunny | 77% |
GERMANY
83%
EXTREME WINDY · WIND
+16pp · 127 wks
| Temp warm | 89% |
| Precip dry | 87% |
| Wind windy | 83% |
| Solar sunny | 83% |
FRANCE
86%
EXTREME WINDY · WIND
+20pp · 126 wks
| Temp warm | 89% |
| Precip dry | 90% |
| Wind windy | 86% |
| Solar sunny | 90% |
IBERIA
88%
EXTREME CALM · WIND
+20pp · 126 wks
| Temp warm | 90% |
| Precip wet | 80% |
| Wind calm | 88% |
| Solar sunny | 82% |
CROSS-MARKET CASCADE
Nordic regime flips often lead Germany and France by 1-2 weeks. Reading the Nordic signal early is reading the Atlantic-European pattern early.
Mid-range weeks run around 57% accuracy — clearly above chance. But extremes are where prices move, and that's where DIANA's edge concentrates.
FULL ACCURACY — EVERY MARKET, EVERY VARIABLE
Directional hit rate across every week · not cherry-picked extremes.
| MARKET | TEMP | PRECIP | WIND | SOLAR |
| NORDIC | 69% | 68% | 70% | 63% |
| GERMANY | 67% | 69% | 67% | 65% |
| FRANCE | 68% | 68% | 67% | 65% |
| IBERIA | 68% | 68% | 68% | 66% |
Baseline: every single week scored, 2010–2024 (in-sample). Every market × variable target beats chance by a wide margin, and the spread across markets is tight — the edge is systemic, not one lucky market.
TOP/BOTTOM 30% — WIDER EXTREMES
When the weather is notably above or below normal, the edge widens.
| MARKET | TEMP | PRECIP | WIND | SOLAR |
| NORDIC | 74% | 76% | 77% | 67% |
| GERMANY | 73% | 75% | 72% | 71% |
| FRANCE | 74% | 76% | 72% | 73% |
| IBERIA | 73% | 71% | 75% | 72% |
Average accuracy across the top and bottom 30% of weeks per variable (recall — conditional on the observed week landing in the tail; in-sample). The step up from “all weeks” is where the edge starts to concentrate.
TOP/BOTTOM 20% — MODERATE EXTREMES
Narrowing the window again — the signal sharpens.
| MARKET | TEMP | PRECIP | WIND | SOLAR |
| NORDIC | 78% | 78% | 83% | 70% |
| GERMANY | 78% | 80% | 75% | 72% |
| FRANCE | 78% | 79% | 78% | 77% |
| IBERIA | 78% | 74% | 79% | 76% |
Top and bottom 20% of weeks per variable (recall, in-sample). DIANA’s edge grows as the week becomes more extreme.
TOP/BOTTOM 10% — SHARPEST EXTREMES
The 10% of weeks where prices move most. Where DIANA earns its keep.
| MARKET | TEMP | PRECIP | WIND | SOLAR |
| NORDIC | 82% | 82% | 84% | 77% |
| GERMANY | 84% | 83% | 81% | 76% |
| FRANCE | 85% | 83% | 84% | 81% |
| IBERIA | 84% | 77% | 85% | 82% |
Top and bottom 10% per variable (n ≈ 125–127 weeks per tail; recall, in-sample). Compare to “all weeks” — every cell sharpens.
Forecast holds across W+0 to W+5
Stable accuracy at every lead time
Direction accuracy across all targets, by lead time. Most forecasts degrade beyond W+2 — DIANA stays inside a tight band.
| LEAD | N WEEKS | DIANA HIT % | vs CHANCE |
| W+0 | 10,499 | 66.5% | +16.5pp |
| W+1 | 10,386 | 67.5% | +17.5pp |
| W+2 | 10,445 | 66.0% | +16.0pp |
| W+3 | 10,446 | 66.1% | +16.1pp |
| W+4 | 10,338 | 68.9% | +18.9pp |
| W+5 | 10,397 | 68.2% | +18.2pp |
What this means for the trader: Most weather forecasts give you a sharp short-range answer that unravels into noise after two weeks. DIANA holds direction skill from W+0 all the way to W+5 — the difference between a 1-week trade and a 6-week hedge book carrying real conviction. The W+5 read at 68.2% is higher than the W+0 read at 66.5% — the forecast doesn’t decay.
Source: 62,511-prediction DIANA Weather backtest, 2010–2024 (in-sample). Direction scored against weekly anomaly (current week vs DOY climatology).
DIANA Weather vs ECMWF energy forecast — head-to-head
EC wins the nowcast. DIANA takes the lead from W+1 onwards.
Both forecasts scored against the same observed-weather direction, 2020–2024 — the years the ECMWF energy-forecast archive covers. 36 Wind / Solar / Temp targets across 13 markets (the variables the EC energy forecast supplies; hydro is DIANA-only and excluded here); ≈2,300 comparisons per lead. DIANA is scored in-sample on its production atlas; the EC leg is true out-of-sample.
All weeks
| LEAD | N | DIANA | ECMWF EC | v7 − EC |
| W+0 | 2,401 | 66.5% | 66.9% | -0.4pp |
| W+1 | 2,340 | 67.9% | 59.7% | +8.2pp |
| W+2 | 2,369 | 68.7% | 53.7% | +15.0pp |
| W+3 | 2,373 | 65.0% | 54.0% | +11.0pp |
| W+4 | 2,304 | 70.0% | 50.6% | +19.4pp |
| W+5 | 2,337 | 70.1% | 48.4% | +21.7pp |
Extreme weeks only — |obs anom| > 1σ — where prices move
| LEAD | N | DIANA | ECMWF EC | v7 − EC |
| W+0 | 676 | 78.8% | 77.1% | +1.7pp |
| W+1 | 776 | 83.4% | 64.9% | +18.5pp |
| W+2 | 717 | 79.5% | 54.0% | +25.5pp |
| W+3 | 723 | 79.1% | 54.2% | +24.9pp |
| W+4 | 673 | 85.6% | 50.2% | +35.4pp |
| W+5 | 668 | 86.7% | 47.0% | +39.7pp |
Heidke Skill Score (1=perfect, 0=chance, −1=anti-skill)
| LEAD | DIANA | ECMWF EC | Δ Skill | |
| W+0 | — | +0.33 | +0.34 | -0.01 |
| W+1 | — | +0.36 | +0.20 | +0.16 |
| W+2 | — | +0.37 | +0.07 | +0.30 |
| W+3 | — | +0.30 | +0.08 | +0.22 |
| W+4 | — | +0.40 | +0.01 | +0.39 |
| W+5 | — | +0.40 | -0.03 | +0.43 |
Read the table: EC has fresh atmospheric initialization, so its Week 0 nowcast is marginally ahead on all weeks (−0.4pp) — and on extreme weeks DIANA already edges it at W+0. From Week 1 the ensemble starts converging on climatology and DIANA takes the lead: +8pp at W+1 widening to +22pp at W+5 on all weeks, and +18 to +40pp on the extreme weeks that actually move prices. The Heidke Skill Score — the standard meteorological metric — confirms: DIANA holds +0.30 to +0.40 at every horizon, while EC decays to zero skill by W+4 and slightly negative at W+5. Basis note: the DIANA leg is in-sample on its production atlas (fully out-of-sample walk-forward: 54.7% overall vs 68.9% in-sample); the EC leg is true out-of-sample.
THE EDGE BY SEASON — DIANA − ECMWF, QUARTER × LEAD
The edge is seasonal — and we show you exactly where.
Each cell = DIANA hit-rate minus ECMWF hit-rate (percentage points), same head-to-head basis as the tables above. Green = DIANA ahead; brown = EC ahead. Negative cells are shown as-is — EC wins several nowcast (W+0) cells; DIANA takes over from W+1–2 in every season.
All weeks
| SEASON | W+0 | W+1 | W+2 | W+3 | W+4 | W+5 |
| Q1 · Winter | +2.2pp | +6.8pp | +18.1pp | +12.2pp | +25.1pp | +18.3pp |
| Q2 · Spring | +7.6pp | +19.9pp | +11.7pp | +22.4pp | +26.9pp | +30.0pp |
| Q3 · Summer | −4.5pp | +7.3pp | +12.3pp | +4.4pp | +11.4pp | +20.0pp |
| Q4 · Autumn | −6.3pp | −0.8pp | +17.9pp | +6.7pp | +16.1pp | +19.7pp |
Extreme weeks (recall) · |obs anom| > 1σ
| SEASON | W+0 | W+1 | W+2 | W+3 | W+4 | W+5 |
| Q1 · Winter | −2.2pp | +14.6pp | +25.6pp | +26.1pp | +36.0pp | +43.9pp |
| Q2 · Spring | +19.4pp | +31.8pp | +28.7pp | +44.7pp | +48.5pp | +42.6pp |
| Q3 · Summer | −0.6pp | +29.8pp | +18.7pp | +15.2pp | +28.1pp | +38.0pp |
| Q4 · Autumn | −9.2pp | −10.2pp | +28.4pp | +10.9pp | +27.8pp | +32.7pp |
n ≈ 510–690 comparisons per quarter-cell (all weeks), 130–265 on extreme weeks. Same basis note as above: DIANA in-sample on its production atlas, EC leg true out-of-sample.
FINER SLICE — MONTH × LEAD
Month by month: where the edge concentrates, and where EC holds its own.
Same data, monthly slice — so you can see the regime transitions the quarterly view smears (the spring storm-track hand-over, the autumn NAO flip). Deep winter and spring carry the widest edge; late summer and October are where EC is closest.
All weeks
| MONTH | W+0 | W+1 | W+2 | W+3 | W+4 | W+5 |
| Jan | +1.8pp | +8.8pp | +21.6pp | +21.1pp | +29.2pp | +27.5pp |
| Feb | +9.4pp | +14.6pp | +25.7pp | +15.2pp | +21.1pp | +19.3pp |
| Mar | +10.5pp | +19.9pp | +25.7pp | +15.8pp | +34.5pp | +27.5pp |
| Apr | +3.5pp | +22.8pp | −5.8pp | +24.6pp | +27.5pp | +19.9pp |
| May | +8.8pp | +17.0pp | +15.2pp | +26.9pp | +18.7pp | +42.7pp |
| Jun | +4.1pp | +25.7pp | +17.0pp | +10.5pp | +21.6pp | +21.6pp |
| Jul | −1.5pp | +4.4pp | +4.9pp | −1.0pp | +13.2pp | +20.6pp |
| Aug | −11.3pp | −1.0pp | +14.6pp | +4.5pp | +4.5pp | +18.8pp |
| Sep | −19.0pp | −1.7pp | +26.0pp | +17.8pp | +15.5pp | +23.0pp |
| Oct | +2.9pp | −9.2pp | +7.6pp | +5.2pp | +12.1pp | +22.4pp |
| Nov | −2.9pp | +8.6pp | +20.1pp | −2.9pp | +20.7pp | +13.8pp |
| Dec | −1.2pp | +0.7pp | +11.9pp | +5.8pp | +25.1pp | +12.0pp |
Extreme weeks (recall) · |obs anom| > 1σ
| MONTH | W+0 | W+1 | W+2 | W+3 | W+4 | W+5 |
| Jan | +14.3pp | +14.4pp | +29.2pp | +36.5pp | +30.6pp | +43.1pp |
| Feb | −4.8pp | +15.3pp | +41.3pp | +32.0pp | +35.6pp | +46.2pp |
| Mar | +10.7pp | +40.0pp | +46.8pp | +37.3pp | +58.2pp | +42.4pp |
| Apr | +30.9pp | +33.9pp | +7.8pp | +43.2pp | +37.1pp | +31.4pp |
| May | +15.9pp | +19.4pp | +27.5pp | +53.4pp | +52.1pp | +52.5pp |
| Jun | +1.9pp | +40.8pp | +24.5pp | +31.7pp | +55.6pp | +20.0pp |
| Jul | +4.4pp | +36.4pp | +7.7pp | −2.8pp | +31.1pp | +57.5pp |
| Aug | −6.7pp | +14.1pp | +21.0pp | +14.9pp | +15.1pp | +32.8pp |
| Sep | −16.7pp | −17.2pp | +27.6pp | +21.1pp | +27.6pp | +24.5pp |
| Oct | +0.0pp | −18.9pp | +31.7pp | +23.4pp | +14.5pp | +44.8pp |
| Nov | −6.4pp | +1.8pp | +25.9pp | −4.8pp | +40.0pp | +27.8pp |
| Dec | −10.3pp | +14.1pp | +13.1pp | +15.7pp | +40.0pp | +43.2pp |
n ≈ 130–260 comparisons per month-cell (all weeks); extreme cells with fewer than 15 comparisons are left blank. Monthly cells are noisier than the quarterly view — read direction and clusters, not single cells.
The takeaway: the edge is widest in deep winter and spring, and at Weeks 2–5 in every season. The only cells where EC holds its own are a handful of late-summer and early-autumn Week 0–1 nowcasts — which is exactly why DIANA uses EC as its Week 0 input and takes over from Week 1.
WHEN DIANA AND EC AGREE
Agreement is the strongest signal of all
The two systems are independent — ECMWF simulates the atmosphere forward from today’s observations; DIANA matches today’s regime against 45 years of outcomes. Filtering to the weeks where both agree and DIANA’s signal is strong selects a higher-quality subset: those 3,116 weeks verify at 62.5%, against a 53.0% base rate across all weeks — measured fully out-of-sample, and replicated at +9.3 on the pre-correction stack. On these agreement weeks EC and DIANA make the same call, so EC is equally accurate there; the value is not beating EC on the same week — it is knowing in advance which weeks are the high-conviction ones. And when the two disagree, the head-to-head above says who to trust: EC at Week 0, DIANA from Week 1 onwards.
80%+ · exceptional 70-80% · strong 65-70% · solid 60-65% · clear edge 55-60% · modest coin-flip = 50% · climatology ≈ 55%
WHY THIS MATTERS
Unseen flip weeks are what break P&L.
TradeWPower has had the same conversation with many desks: one unseen flip week can wipe out a month of accumulated profit. The atmosphere sets up a regime change, renewables shift, balance breaks, prices run — and the position was wrong for that week.
Flip weeks usually arrive in one of two shapes: extreme cold and calm (strong blocking, heating demand and no renewables at once) or a sharp shift into unseen wet-and-windy regimes (storm track drops south, hydro and wind swing above normal together). Both break positions, and both trigger massive stop losses and margin calls. Once a flip week hits, the market can take many weeks to re-settle — spreads widen, liquidity thins, and every position carries extra risk until the new regime is priced in. Seeing them early isn’t a nice-to-have; it’s what keeps the desk in the game.
And it isn’t only European weather. DIANA also reads US weather risk — hurricanes, Gulf cold snaps, cross-Atlantic pattern shifts — that moves TTF through global LNG supply. When US production is disrupted, cargoes re-route, European gas prices move, and power follows. One atmospheric picture, both sides of the Atlantic.
The grids above are measured precisely on these weeks. That’s where P&L is made and lost, and that’s the weeks DIANA is built to call before they hit.
Knowing when you don’t know is the real product.
Every outlook carries a conviction tier — High, Medium or Low — measured across the full 2010–2024 backtest (in-sample). The hit rate rises with conviction, which is exactly what a desk needs: size up on High, lean on Medium, stand down on Low.
Directional accuracy by conviction tier · 2010–2024 backtest (in-sample)
|
73%
|
67%
|
62%
|
|
HIGH
Trade-grade conviction
|
MEDIUM
Lean, don't commit hard
|
LOW
Watch and wait
|
Hit rate = direction calls vs. observed (DIANA Weather backtest 2010–2024, in-sample; HIGH n=8,483 · MEDIUM n=15,967 · LOW n=2,105).
FLIP CAPTURE
A DIANA EDGE badge fires when DIANA and the forecast part ways on direction. Those are the weeks where the consensus is about to move — and the desk that sees it first wins.
EXTREME REGIMES
Trapped-low winters, stretched vortex springs, dry blocking summers — these are the windows where renewables swing hardest and prices break their range. DIANA names them early.
SIZED BY CONVICTION
A High call earns more balance-sheet; a Low call earns patience. The spread between tiers is the point — DIANA tells the desk which is which before the move, not after.
DIANA vs the standard forecast
Where the ECMWF energy forecast ends, DIANA begins
ECMWF’s energy-weather forecast — daily Wind / Solar / Temperature out to D+46, the commercial product utilities and traders consume — is the gold standard at Week 0. EC owns the nowcast because of fresh atmospheric initialization. But by Week 1, EC’s ensemble starts spreading; by Week 2 it has effectively converged on climatology. DIANA dominates from Week 1 to Week 5 — the horizons where regime flips create the biggest P&L swings, and where EC stops being useful.
WEEK 0
EC owns the nowcast — fresh atmospheric initialization gives it a real advantage on Wind / Solar at D+1..D+7. DIANA uses EC’s forecast as an input here, then takes over from Week 1 onwards.
WEEK 2–3
ECMWF skill drops sharply. DIANA’s 45-year regime atlas holds its edge — this is where the extra hit rate lives, and where most of the P&L opportunity sits.
WEEK 4–6 & QUARTERLY
ECMWF ensemble mean collapses toward climatology. DIANA keeps direction through analog matching and long-range atmospheric signals — the horizon where public forecasts stop being useful.
DIANA uses ECMWF where it’s strong and adds signal where it’s weak — not a competitor, a second opinion calibrated to European power.
THE SRMC DISCONNECT
Fundamentals set the floor.
Weather sets the week.
Weather sets the week.
Fuel and carbon analysts model SRMC all day. But in a modern grid, renewables dominate short-term price settling — wind is the primary swing, with solar increasingly in the mix. When the weather goes wet and windy (or calm and dry), spot prices disconnect from SRMC for days or weeks at a time.
That’s where the edge lives. SRMC tells you where price should sit; weather tells you where it’s actually going to sit. The edge is knowing which week the regime will flip — from wet-and-windy to blocked-and-dry, from calm to storm. That is what DIANA does.
5 · Always Learning
DIANA learns two ways, every day. First, through the daily report chain — each morning's observations are scored against the previous outlook, and the analysis we run on top feeds straight back into the next call. Second, through direct conversation: the team talks to DIANA through an AI interface, flags what looks off, asks for a second look, and those questions become part of the next day's research. Climate baselines shift, market regimes change, new extremes appear — DIANA keeps up because it is working with the desk, not behind it.
Daily research cycle
|
Observe
Yesterday’s data
|
⟶ |
Compare
Forecast vs reality
|
⟶ |
Refine
Adjust & verify
|
⟶ |
Deliver
Today’s outlook
|
| ⟵ |
TradeWPower Research Loop · Every Day
|
⟵ |
PROOF THE LOOP WORKS · 2026
Four validated pattern layers, live in the engine
In 2026 the research loop graduated four new pattern layers — atmospheric blocking, regime battles, summer persistence and winter Arctic-Oscillation extremes — each gated on held-out years before activation. Measured together on five held-out years across all 50 targets, they lift week-ahead direction accuracy from 54.6% to 57.0%, with +6.3pp on summer weeks and no horizon degraded. On the specific blocking weeks it wires, the layer calls direction at 64.7% where the raw engine sat at 48.4%. Seven other candidate layers failed the gates and were rejected — the loop keeps what proves out and logs what doesn’t.
SELF-IMPROVING · JULY 2026
A system that improves itself — while you sleep
Since July 2026 the loop closes without anyone touching it. Every daily run logs what the engine claimed — each forecast, every pattern-layer intervention, the would-have-fired calls of candidate layers still held back in shadow mode, and the forecast atmospheric states the next generation of the model depends on. When the week’s weather delivers, every claim is scored hit-or-miss against reality, automatically. Candidate improvements build their live evidence inside production while the current engine keeps forecasting; nothing is promoted until its record clears the gates. The system grows more proven — or gets corrected — every night, unattended.
The direction comes from the desk. Research runs as a continuous conversation between our forecasters and the AI research system: fifteen years of trading-floor weather doctrine — how blocking builds, when regimes break, what the market misprices — is turned into testable hypotheses, mined against 45+ years of data and confirmed on untouched years. The human decides which questions to ask; the evidence decides what survives. What used to take a research season now happens in days — and the standard of proof got stricter, not looser.
And the loop is already building the next engine. The v8 generation — flow direction, blocking state and regime lifecycle as native dimensions of the pattern atlas — is under test now, and in held-out replay it outperforms the current engine at every single horizon. It is promoted only when it proves the same edge live, scored by the same nightly loop. The engine you read today keeps forecasting until its successor earns the seat.
Traders see the output. Behind it, a structured research pipeline run by the TradeWPower team retires hypotheses that stop paying, picks up signals that start working, and keeps DIANA aligned with how European markets and climate actually behave today.
CLIENT CONFIDENTIALITY
The research loop runs on TradeWPower’s own market, weather and backtest data. Client conversations, queries into the AI interface, and anything a client desk shares with us are confidential — they do not train the model. DIANA improves through TradeWPower’s internal research, not through client usage.
6 · How DIANA Changes Your Work
DIANA is not another chart to watch. It changes what each seat on the desk actually does from Monday morning onward — and it changes what Head of Trading can expect from that desk at the end of the quarter. The same daily output reaches traders, hedgers, production planners and analysts with one conviction language, so decisions line up instead of colliding.
WHAT LEADERSHIP GETS OUT OF IT
Fewer losing weeks
Low-conviction weeks are flagged before the desk commits. Avoidable drawdowns become visible, not just explainable in hindsight.
Sharper winning weeks
High-conviction calls carry more size with more certainty — the desk presses when the regime backs it, not when nerve allows.
One team, one view
Traders, hedgers, production and research argue about what to do, not whose forecast is right. Meetings get shorter; decisions get faster.
Fewer hires needed
The method is in the platform, not in one senior analyst's head. New traders are productive in days, and the team scales without doubling headcount.
ONE DAILY OUTPUT · FOUR SEATS ALIGNED
|
Traders
Sizing & timing
|
⟶ |
DIANA
ONE SIGNAL
delivered to every seat,
every morning |
⟶ |
Planners
Dispatch & maintenance
|
|
Hedgers
Ratios & cover
|
⟶ | ⟶ |
Analysts
Research & review
|
SAME CONVICTION LANGUAGE · SAME DIRECTION · SAME WEEK
Traders
Morning Prep → Sized Trades
No more rebuilding a view from scratch each day. Open R5, see where DIANA has conviction, where it disagrees with the forward curve, and where to stay flat. Conviction tiers translate directly to sizing — High warrants the book, Low earns patience. The trader spends the day executing an edge, not assembling one.
Hedgers
Position Sizing → Risk Discipline
Hedge ratios scale with conviction, not with whoever shouted last in the morning meeting. High-conviction weeks warrant tighter hedges; Low-conviction weeks stay flexible. Cost of carry drops because coverage is matched to real risk, not insured uniformly across every week.
Production Planners
Weeks & Months Ahead
Wind, hydro and solar outlooks from W+0 to W+6 feed dispatch and maintenance planning. Select weather analogs and reservoir-fill trajectories extend the view 6 months ahead for seasonal and hydro strategy. The team plans against a single forward view that the trading desk already trusts.
Analysts
Deeper Research, Less Legwork
Stop hand-scoring analog years in spreadsheets. DIANA surfaces the best matches with their outcomes already attached and talks back through the AI interface — ask a question, get a researched answer. Analyst time moves from data assembly to the judgment calls only a human can make.
7 · One System, Whole Team
Most energy desks run on a patchwork of tools — one vendor for weather, another for fundamentals, a third for analytics, plus in-house spreadsheets. Teams disagree not because they read the data wrong but because they read different data. That costs time, money, and hires.
|
BEFORE
Many tools, many silos
Conflicting signals · Larger team · Higher cost
|
CONSOLIDATE
⟶ |
AFTER
One system, one truth
DIANA
ONE PLATFORM · ONE TEAM
Weather + Fundamentals
Analytics + Hydrology Seasonal + Research
Price Model & Market Layer coming
Same signal · Aligned team · Lower cost
|
Clearer communication
Trader, hedger and planner all look at the same outlook with the same conviction tag. No more meetings to reconcile whose forecast is right.
Easier hiring
New traders are productive in days, not months. The method is in the platform, not in one senior analyst's head.
Leaner team, lower cost
Fewer overlapping tools to license. Fewer full-time hires to cover weather, fundamentals and analytics separately. Margin that stays with the desk.
One platform, all aspects
DIANA Weather today. Price and market layers coming. The full decision chain, without bolting a fifth vendor on top.
8 · The Backbone — What DIANA Enables Next
THE UNIQUE FOUNDATION
What you're using today is the hard part.
Nobody else has built this.
Nobody else has built this.
45 years of European energy-weather compressed into a daily decision engine that knows when it's right and tells you when it isn't. That calibration layer is the thing most forecasting products never get to. DIANA has it. Running it daily — proving the edge, capturing the misses, refining the signal — is Phase 1 of a bigger build.
PHASE 1 · NOW RUNNING
The hard part is done
The foundation works. Every day it runs is another day of proof. 45 years of weather, 13 power markets, stratosphere-to-surface regime matching, conviction tiers that survive out-of-sample testing. This is the unique thing. No one else in European power has the atlas and the calibration layer we have.
PHASE 2 · BUILDING
Scale the engine, expand the map
More data, more research, more knowledge. With the lessons that come from running Phase 1 daily, DIANA scales into more markets and more horizons — weather-driven risk wherever renewables set the price. The engine gets sharper every year it runs, and hit rates compound as the knowledge base grows.
SEQUENCING
Why now, not yet
Phase 2 requires Phase 1 to be stable first. You can’t scale a model until you understand where it wins and where it misses. That’s what today’s daily runs are producing — a stream of calibration data that makes the next DIANA possible. Running this version is the research that funds the next.
From Phase 1 to Phase 2 — the engine compounds
|
PHASE 1 · NOW RUNNING
DIANA v1
European power · 45 yrs data
13 markets · stratosphere-to-surface
daily calls with calibrated conviction
tens of thousands of backtested calls
— the unique thing nobody else has built —
|
SCALES INTO
⟶
with more data
more research more knowledge |
PHASE 2 · BUILDING
DIANA v2
more data · more research · more knowledge
more markets · more horizons
pattern discovery at a wider scale
hit rates that compound yearly
— built on the lessons Phase 1 is writing —
|
↺ TRADEWPOWER DAILY RESEARCH · EVERY MISS IS A LESSON
The thesis, in one line
Markets reward people who see first.
DIANA is the tool that sees first. This version finds the signal — today, across European power. The next version finds more signals across more markets, but only because this one proved it works.