How EDGE reads the tape.
Three independent fundamentals layers. Nine engines, every one live. Twenty-one surfaces in the terminal underneath. The architecture brief for the institutional bias engine at retail price. What it reads. What it does not claim. What the room looks like beyond the leaderboard window.
The architecture in one paragraph.
EDGE is a confluence engine. Three independent fundamentals layers — smart-money positioning from the weekly CFTC report, twenty-five years of calendar memory, and seven live macro engines reading economic prints, central-bank rate-path predictions, real-exchange-rate valuation, the carry differential, cross-asset implied volatility, news-flow geopolitical stress, and G4 central-bank net liquidity — vote on the directional bias for every tracked pair at every timeframe you can hold. Each engine carries two weights. The first is a horizon weight that captures how strongly its mechanism operates at the timeframe in front of you. The second is a mechanism-credibility percentage that captures how directly its documented academic mechanism translates into FX direction. Both multiply against the disagreement penalty on the A++ to FLAT tier ladder. What you see on the leaderboard is the agreement of the engines that have something credible to say at the timeframe you selected. No black box. No “coming soon” engines. No predictions of the next return. Confluence layered with your own discretion is the thesis. Everything below is the proof — and the tour of the terminal that consumes it.
The three layers.
Three layers, each grounded in a different data-source family with a different mechanism on a different horizon. The architectural intent: when all three agree on the same direction, the underlying causes are independent and the joint probability the trade reads the tape correctly is higher than any individual layer's base rate. When they disagree, the tier label drops and you see exactly which layer pulled it down.
Layer 1Positioning · weekly CFTC Commitments of Traders
The Commodity Futures Trading Commission publishes the disaggregated Commitments of Traders report every Friday at 15:30 Eastern Time. Per-currency futures contracts cover the euro, sterling, yen, Swiss franc, Australian dollar, New Zealand dollar, Canadian dollar, Mexican peso, and the US dollar index. The report names three cohorts: commercials (hedgers and producers, the documented smart-money footprint), large speculators, and non-reportables (retail, the documented losing footprint across two generations of academic literature).
EDGE reads this layer through the documented contrarian framing established across decades of academic work on CFTC positioning: when the losing cohort has crowded one side while the hedging cohort has taken the opposite, the binary inequality is the read. The threshold is relative-extreme on a multi-year window, not an absolute contract count, which means the signal is the position of the position. For twenty cross pairs with no direct futures contract, the read is synthesised from per-currency components. Engine credibility 95% — top of the locked table.
Live surface · /cot — per-currency positioning workbench with cohort breakdown and the trailing 156-week extreme band drawn on screen.
Layer 2Calendar · twenty-five-year multi-timeframe seasonality
Calendar effects in FX have institutional drivers: corporate tax estimated-payment cycles, fiscal year-end repatriation flows, options-expiry quarterly cadence, vacation-period thin-liquidity drift, central-bank meeting clusters. Aggregated across twenty-five years of daily price history these drivers leave a fingerprint in the joint distribution of weekly and monthly returns. The engine reads that fingerprint over five separate lookback buckets — 5, 10, 15, 20, and 25 years — and requires multiple buckets to confirm the same direction before the read counts.
The discipline is the multi-bucket threshold. A single coincidence on a four-year cycle cannot win the read. Agreement across decades can. Three confidence tiers sit on top of the bucket layer: standard, strong, and insane — defined by joint thresholds on the bucket winrate and the risk-adjusted return per window. Of every engine in the table, this is the one that retains full weight across the entire tradable horizon from one week to six months.
Live surface · /seasonality — per-year overlays, the window optimiser, the bucket-confirmation matrix.
Layer 3Macro · seven independent fundamentals engines
The macro layer is seven engines, each reading a different data-source family on a different horizon with a different documented mechanism. Some run weekly. Some monthly. One fires only when a Federal Open Market Committee outcome contract on Polymarket clears a $1 million 24-hour volume gate, then falls silent the moment the meeting resolves. The composite is the mean of the available signed scores, with a horizon-weight table ensuring each engine carries the right voice at the right timeframe.
The horizon table is the part most platforms collapse. A carry-trade differential is structural at the 3M-and-beyond horizon and near-zero in the one-week weight slot. A VIX spike has 7-day mean-reversion and near-zero weight at six months. EDGE weights each engine HIGH, MED, or LOW per timeframe based on the academic horizon where the mechanism actually operates — and the engine that disagrees at LOW weight does not get to drag a pair's tier as hard as the engine that disagrees at HIGH weight.
Live surface · /macro/engines — per-engine current reading, mechanism, and credibility scorecard.
Every engine, every read, every source.
What follows is the engine table — nine engines, every one live and active in the tier math at this moment. Each row carries the layer it belongs to, the mechanism in plain English, the cited source, and the current credibility percentage in the locked rubric. The raw inputs, transforms, gates, and weight tables are not on this page. They live inside /methodology/engines for readers who want the recipe.
Williams COT index on commercials
95% credibilityWeekly CFTC positioning, every Friday at 15:30 Eastern. The mechanism is the contrarian retail footprint documented in the CFTC's own market participant studies and decades of academic work on positioning extremes: when the losing cohort crowds one side and the hedging cohort takes the opposite, the binary inequality is the read. The framing is relative-extreme on a multi-year window, not an absolute contract count — which means the signal is the position of the position, not the position itself. The signal extends to twenty cross pairs without direct futures contracts through clean per-currency synthesis.
CFTC weekly disaggregated Commitments of Traders report (cftc.gov). Per-currency contracts: 6E/EUR, 6B/GBP, 6J/JPY, 6S/CHF, 6A/AUD, 6N/NZD, 6C/CAD, 6M/MXN, DXY/USD.
Multi-timeframe seasonality
80% credibilityTwenty-five years of daily price history aggregated across five lookback buckets (5, 10, 15, 20, 25 years). The discipline is the multi-bucket threshold: a single coincidence on a four-year cycle cannot win the read; agreement across decades can. Three confidence tiers (standard / strong / insane) sit on top of the bucket layer. The engine retains full weight across every tradable horizon from one week to six months — the only engine in the table that holds the entire tradable surface.
25 years of daily price history per pair (Twelve Data EOD with Dukascopy backfill). Per-bucket aggregation and the multi-timeframe ranking.
Macro fundamentals composite (41 indicators)
85% credibilityForty-one FRED series across eight currencies. Growth, inflation, labor, rate paths, sentiment — each composited per currency and differenced into a pair score. The mechanism reads how surprise prints versus consensus shift the rate-path narrative around every meeting; the Federal Reserve's 2025 research on macro-news processing documents the multi-day-to-multi-week persistence of those shifts.
St. Louis Fed FRED API (api.stlouisfed.org). Federal Reserve research 2025 — 'How Markets Process Macro News'.
Rate-event news predictor
95% credibilityPolymarket and Kalshi binary contracts on every FOMC meeting outcome bucket. The engine fires only when the contract is institutionally liquid: an owner-locked $1 million 24-hour volume gate across outcome buckets. Below the gate the engine refuses to vote. After the meeting resolves the contract closes and the engine returns silent — the news-predictor pattern the design is built around.
Polymarket gamma API + Kalshi event API. Live hourly cron writes to prediction_markets. Owner-locked $1M liquidity gate.
REER valuation (real effective exchange rate)
85% credibilityBIS broad REER monthly series. The mechanism is the documented multi-year mean-reversion of real exchange rates: a currency two standard deviations above its long-run trend carries an 18-to-36-month gravitational pull toward fair value. The engine weights low at the short timeframes and higher at the 3M and 6M slots where the academic horizon plays out. Intentionally slow — sets the structural backdrop, never the trigger.
BIS broad REER monthly series (bis.org/statistics/eer.htm). Menkhoff, Sarno, Schmeling, Schrimpf 2017 — 'Currency Value', Review of Financial Studies 30(2).
Carry differential
95% credibilityThe Lustig-Verdelhan term structure of currency carry-trade risk premia. The policy-rate differential between the two legs of a pair drives systematic carry flows from institutional capital at the one-month-to-one-year horizon documented in the AER paper. The differential is a structural tailwind or headwind that compounds underneath any tactical setup. The engine weights low at one-to-two weeks where carry barely compounds and high at 3M and 6M where the academic edge materialises.
Lustig, Verdelhan 2019 — 'Term Structure of Currency Carry Trade Risk Premia', Journal of Finance. Daily policy rates via FRED.
Cross-asset implied volatility
75% credibilityVIX from S&P 500 options, MOVE from Treasury options, OVX from crude options. Each encodes the price the market is paying for thirty-day portfolio insurance in its respective asset class. Joint spikes signal cross-asset risk-off flow that moves money out of pro-cyclical currencies (AUD, NZD) and into safe-haven currencies (JPY, CHF). VIX mean-reverts on a 7.2 trading-day horizon, which sets the engine's short-window weight emphasis.
Yahoo Finance — ^VIX, ^MOVE, ^OVX daily close. BIS Quarterly Review Dec 2025 — 'Volatility challenges risk-taking'.
Geopolitical stress (EPU + GPR + GDELT)
70% credibilityThree news-flow indexes that measure policy and geopolitical stress at the source: Baker-Bloom-Davis Economic Policy Uncertainty, Caldara-Iacoviello Geopolitical Risk (2022 AER), GDELT 2.0 machine-coded events. Stress leads the implied-vol reaction by hours-to-days because news drives the option-premium response. The signal is fast and short-lived, which is exactly why the locked rubric caps it at 70% — known limit, named limit.
Caldara, Iacoviello 2022 — 'Measuring Geopolitical Risk', American Economic Review 112(4):1194-1225. Baker-Bloom-Davis (policyuncertainty.com). GDELT Project 2.0.
G4 central-bank net liquidity
80% credibilityFederal Reserve SOMA, ECB asset-purchase holdings, Bank of Japan Rinban schedule, Bank of England Asset Purchase Facility. When the four largest reserve-currency central banks are collectively expanding net of redemptions and quantitative tightening, the system is adding base money and asset prices respond on a one-to-three-month horizon. The engine reads the rate of change, not the absolute level — the level is path-dependent on the post-2008 unwind cycle.
Federal Reserve H.4.1 release, ECB SDMX, Bank of Japan Rinban schedule, Bank of England Bank Return.
What the math underneath actually does.
Every engine produces a signed score in a normalised range. The leaderboard composite is the mean of the available reads. Direction is set with a noise-floor band so flat tape stays flat. Then each engine votes — agree, disagree, or off (no signal, gated, below threshold) — and the pair's tier drops from the A++ ceiling on the A++ → A+ → A → B → C → FLAT ladder for every engine that disagrees or sits out.
The penalty is weighted twice. First by the engine's horizon weight at the timeframe in front of you (HIGH, MED, or LOW). Second by the engine's mechanism credibility in the locked rubric (the percentage you see on every engine row above). Both multiply into the disagreement cost. Penalties round down so fractional results never promote a pair into a better label. The floor is FLAT; no setup sits below that point.
What you see on the leaderboard is the agreement of the engines that have a credible read at the timeframe you selected. The full per-pair vote table, the current engine reading, the horizon weight at your selected timeframe, and the credibility percentage at the moment of the read are all visible inside the per-pair confluence detail at /pair/[symbol]. The methodology surfaces themselves are public and unauthed: the engine list, /methodology/engines for the per-engine recipe, and /methodology/validation for the walk-forward scoreboard. Every claim in this brief can be audited against the source above the email.
The mechanism-credibility rubric.
The 0–100 credibility percentage on each engine is not a popularity score. It is the engine's grade against a five-part rubric adapted from a public messaging discipline used inside government communications work: a claim is credible to the extent its messenger (1) receives the read at the right cadence, (2) presents it in plain language defensibly, (3) cites a source the audience already trusts, (4) translates directly to the decision the audience is trying to make, and (5) holds up across the diverse contexts the audience will encounter.
- Received. Does the upstream publish at a cadence that lets the signal stay fresh inside the user's decision window? Weekly CFTC scores well at 1W–1M; monthly BIS REER scores well at 3M–6M.
- Understood. Can the mechanism be stated in one plain sentence a working trader can verify? Carry differential clears this cleanly; multi-factor models require more setup, hence the lower number.
- Believed. Does the source carry independent academic or regulatory authority? Lustig- Verdelhan 2019 AER for carry. Caldara-Iacoviello 2022 AER for geopolitics. CFTC regulatory mandate for COT. BIS Quarterly for vol-indices.
- Offers a solution. Does the engine's reading map to a per-pair direction the trader can act on? Cross-pair Williams synthesis ships the COT signal to twenty cross pairs without direct futures contracts.
- Brings the result. Does the read hold up across diverse pair-timeframe contexts? Seasonality at multi-bucket agreement is the discipline that filters single-coincidence reads from cross-decade structure.
Top of the table at 95%: positioning, carry, the rate-event predictor. Floor of the live table at 70%: geopolitical stress, where the short-lived nature of news-flow shocks documented in Caldara-Iacoviello 2022 caps the directness — a known limit, named so the user can weight it themselves. The percentage drives how much each engine's disagreement actually costs the pair on the ladder.
The terminal underneath the leaderboard.
The brief above describes the bias engine. The terminal that consumes it is larger. A ranked leaderboard you can read in one screen. A confluence detail view that explains every tier label. A CFTC workbench, a seasonality workbench, a macro engines deck, a screener, a setups feed. Underneath the bias surfaces sits the lifecycle stack: position sizing, equity-curve simulation, prop-account tracking, the journal that reconciles every closed position against the brief that opened it. Underneath that, the audit layer: the recipe for every engine, and the public scoreboard for the walk-forward validation result. Twenty-one surfaces. One source of truth.
- /leaderboard
The ranked bias terminal. Every tracked pair, every timeframe, the tier label and direction and confluence count in one screen.
- /pair/[symbol]
Per-pair confluence detail. The engine vote table, the layer agreement, the citation underneath every read, the timeframe stack.
- /cot
The COT workbench. Per-currency positioning by cohort, the 156-week extreme band drawn on screen, smart-money vs retail at a glance.
- /seasonality
Twenty-five years of per-year price overlays. The window optimiser stress-tests any [start, end] window on any pair across the five bucket layer.
- /macro/engines
Seven live macro engines in one deck. Current read, mechanism, credibility scorecard, last-refresh timestamp per engine.
- /screener
The universe filtered. Tier, direction, confluence count, engine agreement — every cut of the leaderboard you want to make.
- /setups
Today's A++ and A+ candidates in priority order. The setup feed for traders who run a setup-driven book.
- /tools/position-size
Account-anchored size from tier and stop. Account currency, risk percentage, target pair, stop distance — out comes lot size that protects the account.
- /tools/monte-carlo
Equity-curve simulator. Feed in a setup mix and the runner returns a thousand simulated paths with drawdown distributions and ruin probabilities.
- /prop-tracker
Funded accounts against drawdown limits. FTMO, MyForexFunds, FundedNext, all of them — daily DD, max DD, profit target, distance to breach.
- /journal
Auto-imported trades from MyFxBook plus CSV. Per-trade tagging by tier and engine confluence. Every closed position reconciled against the brief that opened it.
- /alerts
Watchlists and tier-change alerts. When EURJPY crosses from A to A++ at the 1M slot, the inbox knows before the screen does.
- /methodology/engines
The per-engine deep page. Every input, every transform, every weight, every gate. The recipe lives here for the curious.
- /methodology/validation
The walk-forward audit. Bailey-López de Prado 2014 Deflated Sharpe Ratio scoreboard, machine-readable, public. The honest constraint with the receipts.
The architecture beyond the leaderboard.
The bias engine is the foundation. The terminal is the floor it supports. When the leaderboard hands you an A++ short on USDJPY at the one-month horizon, the next questions are mechanical, not conceptual.
How big? Plug the account, the risk percentage, the stop distance, and the pair into the position calculator. The calculator handles every account currency and contract specification so the lot size lands inside the rule the funded-account agreement actually wrote.
What does this look like across a thousand simulated equity curves? Feed the setup mix into the Monte Carlo runner. Out comes the drawdown distribution, the time-to-ruin probability, the median path and the worst tail percentile. Decisions made against a thousand simulated futures sit better than decisions made against one observed past.
Which of the funded prop accounts can carry the trade? The prop-firm tracker holds every active funded account against its daily drawdown cap, max drawdown floor, and profit target, with the breach distance updated in real time. The account that has the headroom takes the trade.
And after the position closes — the journal reconciles the closed trade against the brief that opened it. Tier, engine confluence, timeframe, entry rationale, after-action. Imported automatically from MyFxBook or uploaded via CSV. The journal is the institutional after-action that turns ninety-five closed trades into the data you can stress-test the next setup against.
Trigger-layer indicators — the regime indicator that converts confluence into entry timing — ship as Pine-script for TradingView. Distribution happens where the chart is already open; the terminal stays the source of truth on the brief that fed the trigger.
What this means in practice: the leaderboard is the window. The terminal is the room. Most platforms stop at the window. EDGE built the room.
What this document deliberately does not claim.
EDGE is a confluence display, not a predictive model. The tier label on any pair at any timeframe is the agreement tally of independent engines. It is not a prediction of that pair's next return distribution. We do not make per-pair directional calls. We do not publish backtested performance numbers. We do not claim any engine cleared a multiple-testing-corrected statistical-significance bar on standalone walk-forward validation.
We ran exactly that test in 2026. Twenty-eight standalone signals across seven G7 pairs and four trading-day horizons (5, 10, 21, 63) through walk-forward partitioning plus a Deflated Sharpe Ratio threshold (Bailey-López de Prado 2014) at DSR > 0.5. Of one hundred and twelve trials, zero cleared the bar. The audited scoreboard for that run lives at /methodology/validation — public, machine-readable, no auth, with every trial preserved in raw form.
That result does not invalidate confluence. It clarifies what confluence is. Confluence is a structured way to display the agreement of multiple independent indicators without claiming any single one will predict the next return. The trader brings the discretionary read on top: the entry trigger, the stop placement, the sizing. The terminal hands them nine independent reads, organised, with mechanism and citation visible. What they do with those reads is on them.
Stated plainly: no single signal in the tested set cleared the deflated-Sharpe bar on standalone walk-forward. The architectural point of EDGE is that confluence layered with discretion is a different game than betting a single signal, and the terminal publishes the multi-engine read so the operator can do the layering themselves. The honest constraint is the brand. Anything else would have been a lie at the cover.
Citation manifest.
Every engine mechanism above ties back to at least one of the following sources. Academic citations carry author, year, journal, and volume where applicable. Data-source citations carry the public endpoint. There is no “internal research suggests” anywhere in this document.
- academicLustig H., Verdelhan A. (2019). 'Term Structure of Currency Carry Trade Risk Premia.' Journal of Finance.
- academicMenkhoff L., Sarno L., Schmeling M., Schrimpf A. (2017). 'Currency Value.' Review of Financial Studies 30(2):416-441.
- academicCaldara D., Iacoviello M. (2022). 'Measuring Geopolitical Risk.' American Economic Review 112(4):1194-1225.
- academicDu W., Tepper A., Verdelhan A. (2018). 'Deviations from Covered Interest Rate Parity.' Journal of Finance 73(3):915-957.
- academicAsness C., Moskowitz T., Pedersen L. (2013). 'Value and Momentum Everywhere.' Journal of Finance 68(3):929-985.
- academicBailey D., López de Prado M. (2014). 'The Deflated Sharpe Ratio: Correcting for Selection Bias, Backtest Overfitting and Non-Normality.' Journal of Portfolio Management.
- researchFederal Reserve System (2025). 'How Markets Process Macro News.' FEDS Notes / working paper series.
- researchIMF (October 2025). 'Risk and Resilience in Global FX.' Global Financial Stability Report chapter.
- researchBIS Quarterly Review (December 2025). 'Volatility challenges risk-taking.'
- dataU.S. Commodity Futures Trading Commission. Weekly disaggregated Commitments of Traders report. cftc.gov.
- dataSt. Louis Fed FRED API — 41 macro series across G8 currencies. api.stlouisfed.org.
- dataBank for International Settlements — broad REER monthly series. bis.org/statistics/eer.htm.
- dataPolymarket gamma API + Kalshi event API — FOMC outcome-bucket prediction markets, hourly snapshot.
- dataFederal Reserve H.4.1 release, ECB SDMX, Bank of Japan Rinban schedule, Bank of England Bank Return.
- dataYahoo Finance — ^VIX, ^MOVE, ^OVX, ^DXY daily close.
- dataBaker, Bloom, Davis Economic Policy Uncertainty — policyuncertainty.com.
- dataGDELT Project 2.0 — Global Database of Events, Language, Tone.
Closing.
This brief is the public proof document. The live surfaces that consume it are listed above and indexed across the three layers — bias, lifecycle, audit — every one reachable from this page without an account. Corrections, questions, or a request for the raw config behind any engine: email derek@edgeterminal.app. Every part of this document — the engine list, the credibility rubric, the citation manifest, the walk-forward scoreboard, the surface tour — is owner-locked and dated. Future revisions will preserve the structure with a version stamp at the top.