Every contract across the energy complex — ICE Europe, ICE US Energy Division (IFED), NYMEX/CME, and Nodal Exchange — decomposed in your browser into dated factor legs with per-leg product codes. Outrights, strips, calendar spreads, CSOs, bullet options, basis swaps, heat-rate futures, refining cracks — every structure handled. v3 extends the universe to 430+ factors across 470+ contracts, applies a single hierarchy across all of it (gas delivery points regionalized, power zones mapped to their marginal gas hub, crude organized by region × grade × application), and layers in open interest and volume per contract.

Drive it below — hit "Load sample" for a mixed portfolio spanning every sector, paste your own CSV, or open the Playbook tab: five strategy archetypes (locational, term-structure, crack, heat-rate, cross-venue) plus the new ATLAS vs VECTOR case study — a $5.07B offset book against a $2.66B conviction book, and why the bigger one is the safer one. Full coverage breakdown, what's under the hood, and limitations are below the dashboard.

Every sector below is decomposed by the same engine driving the dashboard above — 430+ atomic risk factors across 470+ contracts, spanning ICE Europe, ICE US Energy Division (IFED), NYMEX/CME, Nodal, and Nasdaq. Natural gas is organized by delivery-point region, power by ISO → Hub → Zone with each region mapped to its marginal gas hub, crude by region × grade × refining application, and refined products by exchange geography with the full East–West and crack spread suite. Open interest and daily volume ride along per contract (snapshot as of June 30, 2026 — CFTC COT, exchange releases, and clearly-flagged estimates).

A calendar spread isn't one exposure — it's a pair of dated legs whose basis can dominate the P&L. A power Cal-strip doesn't scale evenly across months because peak-hour counts differ. A 3:2:1 crack's constituents move on different ICE codes. If your risk system papers over any of that, every downstream report carries an invisible error.

This project is what happens when you take that observation seriously: a decomposition library — ice_energy_decomp — that expands every contract type into dated factor legs with per-leg ICE codes, and a browser dashboard whose calculation engine is checked cell-for-cell against the Python reference. The point isn't the dashboard itself. The point is that the decomposition is what every downstream calculation (margin, P&L, scenario, value-at-risk) needs to start from.

The Playbook tab ships five worked example portfolios, each chosen to exercise a different part of the decomposition engine — and each loadable into Portfolio Analysis with one click:

• Locational arbitrage — long the Waha discount, short the HSC premium, with the ERCOT West-vs-North power echo. Demonstrates basis-leg netting: both basis swaps carry Henry Hub legs that cancel automatically, leaving pure location risk.
• Term-structure arbitrage — gas winter/shoulder calendar spreads, a Brent Dec-Dec roll, and a TTF quarter-vs-strip. Nearly zero net factor exposure; the whole book lives in the tenor dimension, which only the per-month expansion can see.
• Crude / crack arbitrage — long 3:2:1 vs short 2:1:1 (a pure gasoline-vs-distillate tilt), plus sour-grade discounts and the Murban–Dubai quality spread. Demonstrates fractional coefficients and gal→bbl / mt→bbl unit conversion.
• Heat-rate arbitrage — a listed spark future, the same structure replicated manually from power + gas legs, and a winter/duck-curve pair. The cross-commodity matcher reconciles all three against the power↔gas association map.
• Diversified low-net / high-gross — seven economically-flat cross-venue pairs (NYMEX↔ICE Henry, WTI, Brent; ICE↔Nodal PJM; ICE↔CME basis). Net delta ≈ 0, but five clearing houses each demand full margin — the venue table makes the margin tax visible.

Each playbook includes the thesis, the position CSV, a step-by-step account of how the engine decomposes it, a tab-by-tab guide to reading the result, and what can go wrong.

To put the machine to work at institutional scale, the Playbook tab now ships two case-study portfolios built from the same five archetypes and analyzed side-by-side in a full research note. ATLAS is a $5.07B-gross book engineered from offsets — cross-venue pairs, calendar spreads, crack-vs-crack tilts, spark replications, basis pairs — that collapses to $133M of net directional exposure (2.6%). VECTOR is a $2.66B-gross conviction book carrying $866M net (32.5%). Same instruments, same liquidity standard (≥97% of gross in liquid tiers on both).

The result: the book with 1.9× the gross notional loses 8× less in its worst stress scenario (−$12.6M vs −$102M) — yet posts twice the initial margin ($465M vs $228M across five clearing houses), because CCPs margin positions, not economics. Exposure in energy is a function of net, not gross; but the market prices gross anyway. The full analysis — decomposition mechanics, sector nets, tenor structure, heat-rate matching, scenario grid, and the margin table — is in the PDF.

The engine and its browser counterpart share their interface: 430+ factors, 470+ contracts, nine contract structures (OUTRIGHT, STRIP, CAL_SPREAD, OPTION, CSO, plus basis, cracks, heat-rate futures, and spreads). Each contract maps to a weighted sum of factors; positions expand through the structure handler into dated monthly legs scaled by the contract's unit conventions. Contracts that represent economically equivalent exposure across exchanges (e.g., NYMEX HH ↔ ICE HH LD1, NYMEX WTI ↔ ICE WTI) keep distinct factor codes so inter-exchange basis surfaces explicitly in the cross-exchange offset view rather than silently netting.

• Power MW→MWh. A NERC- and EU-aware calendar counts on-peak / off-peak / base hours per delivery month — verified against independent sources (PJM peak July'26 = 368h, Christmas-Friday correctly drops a peak day, peak + off-peak ties to total).
• Options. Black-76 delta for bullet options on positive forwards; delta_override for non-standard structures including CSOs.
• Strips and spreads. Calendar strips (CAL27, 2026Q3, WIN26, custom ranges) expand to monthly legs; calendar spreads net the front and back tenors.
• Cracks. 3:2:1, 5:3:2, and other refining cracks decompose into their crude / gasoline / distillate constituents with industry-standard ratios.

Parity testing. The Python reference engine and the JavaScript browser engine are checked cell-for-cell — 32/32 (factor, month) cells match exactly on the full reference portfolio. Phase 1 tests: 24/24. Phase 2 tests: 23/23. The dashboard is not a separate, drift-prone reimplementation; it shares the same logic that's checked against the source. The v3 data layer adds its own integrity gate: every decomposition leg must reference an existing factor, every hierarchy child an existing node, and every playbook CSV must parse and expand without error before the build ships.

Two defects the v3 verification pass caught and fixed. First, the v2 data merge dropped the NERC/EU delivery calendar, so MW→MWh block-hour scaling could fail for power positions. Second, Nodal zone contracts were authored with a delivery-scaling alias the engine didn't recognize, silently understating every Nodal MW position by the month's block-hour count (~350×). Both are corrected in the v3 merge — power exposures now scale correctly across ICE, CME, and Nodal venues.

This is a research and education tool. It is not investment advice, not a trading system, and not a settlement-grade risk system. Concretely:

• The seed portfolios are illustrative, not real positions. They exist to demonstrate the decomposition and to give you something to drive the UI with on first load.
• $-PnL requires user-provided marks. There is no live ICE/EEX/NYMEX price feed wired in — pasted marks are the source of truth for any P&L number you see.
• Open interest and volume are a snapshot, not a feed. Figures are as of June 30, 2026, sourced from CFTC Commitments of Traders (June 9 positions), ICE/CME/Nodal press releases, and — where marked with an asterisk or "est." — order-of-magnitude estimates anchored to the latest verified data. Thin hubs below the CFTC 20-trader reporting threshold show tier only.
• Some ICE product codes are flagged unverified. In the Contract Reference tab, a small number of contracts show an UNVERIFIED chip — those are best-effort mappings I've been unable to confirm against ICE's product master. Treat the verified-flagged codes as load-bearing; treat the unverified ones as placeholders.
• Black-76 assumes positive forwards. Calendar spread options (CSO) and basis options where the underlying can go negative require a normal/Bachelier model — currently those positions use delta_override as a workaround. Don't rely on the Black-76 delta for those structures.
• Power calendars cover US (5×16) and EU (5×12 peak / 7×24 base) conventions. Other regional conventions (UK 7-day with bank-holiday handling, Nordic, Iberia variants) need explicit configuration.
• Crack ratios use industry-standard conversion factors. 3:2:1, 5:3:2, etc. follow USGC convention. Verify against your firm's spec before applying to a real book.
• Persistence is browser-local only. Portfolios and preferences are saved in localStorage — no cloud sync, no server backup. Clearing site data deletes your portfolios. Don't use this as a system of record.
• Scenario shocks are deterministic factor moves, not Monte-Carlo. The Risk panel applies named factor-level shocks and reports the linear P&L impact; it doesn't propagate correlated tails or run distributional VaR.
• Margin and clearing logic is out of scope. The dashboard returns exposures and P&L; it does not compute SPAN, VaR-based margin, or ICE's actual margin algorithm. Don't use these outputs to anticipate margin calls.

The roadmap that keeps this useful as it grows:

• Live market data integration. v3 ships an embedded OI/volume snapshot (as of June 30, 2026); the next step is a refresh pipeline off the exchanges' daily volume-and-OI files, then a feed-driven mark layer for P&L.
• Verified ICE product codes. Close the gap on the unverified-flag set against ICE's product master.
• Normal-model option pricing. Replace the Black-76 / delta_override workaround with a proper Bachelier model for CSOs and basis options.
• Server-side persistence (optional). A save-portfolio endpoint with auth, so portfolios survive a browser clear and can be shared across devices.
• Additional regional conventions. UK power with bank-holiday handling, Nordic system price, Iberia, Japan JEPX.
• Export. One-click PDF / Excel of any tab's state so analyses can be circulated outside the browser.