An algorithmic /NQ futures strategy publishing every trade — winners, losers, and the methodology behind them. Walk-forward validated, ATR-anchored, retrospective only.
Closing the week ending 2026-05-03, the strategy logged 8 trades across 7 sessions for a 50% hit rate and a net of +44.0 points, with profit factor at 1.88 and max drawdown from peak of 25.0 points. Most of the work clustered around Initial Tag entries into prior Imbalance Zones, with a couple of Retest Continuation setups carrying the bulk of the gains after an Anchor Bar confirmed the Initiative Move. The losing half came largely from Pivot Sweep attempts that failed to follow through, which is the honest cost of a 50/50 distribution. Even ledger on count, but winners ran further than losers, and that asymmetry is what kept the week green.
| Session | Trades | Net pts |
|---|---|---|
| 2026-05-03 | 3 | +30.50 |
| 2026-05-02 | 2 | -8.50 |
| 2026-05-01 | 4 | +47.00 |
| 2026-04-30 | 0 | — |
Numbers verified by sha8 hash of source data — see Notes № 02.
Up-IZ formed 09:32 ET, height 7.0 pts. Initial Tag long entry @ 18500.00 at 10:14 ET. Stop 18488.00 (risk 12.0 pts), target 18524.00. Result: winner, +23.5 pts (+1.96R).
Why this one: largest absolute R-multiple of the day. We publish both winners and losers — losses are often where the mechanics show most clearly.
Down-IZ formed 11:05 ET. Retest Continuation short entry @ 18540.00 at 11:48 ET. Stop 18552.00, target 18516.00. Result: loser, -12.5 pts (-1.04R).
We're posting this loser because the failure mode is instructive: the Anchor Bar held on the first test, encouraging the continuation entry, but the second test never reached the target before a Pivot Sweep above stopped us out.
You're going to see specific terms used here over and over. Some of them differ from what's common in trading-education circles. That's intentional.
We use language grounded in market mechanics — auctions, order flow, stop placement, mean reversion. We don't use language that implies we know what large institutions are "doing" or "thinking." We don't. Nobody does. What we can observe is the footprint that order flow leaves on price: where it accelerated, where it skipped, where it reacted.
The core term is the Imbalance Zone (IZ) — a region of price the auction skipped on its way through. Other terms (Initiative Move, Anchor Bar, Pivot Sweep, Retracement Tag) all describe specific, observable, mechanically-grounded patterns of order flow.
We avoid words like smart money, manipulation, and killzone. They sound authoritative but they're load-bearing fiction — they ascribe motive to anonymous prints. We'd rather be precise about what we can see than dramatic about what we can't.
If a term ever feels vague, ask. We'll define it in terms of what's actually happening on the order book.
When price moves fast enough that part of the range gets skipped — bar 1's high sits below bar 3's low (or vice versa), with bar 2 spanning the gap — the middle section was never two-sidedly transacted. That's an Imbalance Zone.
It isn't magic. It's an order-flow exhaustion footprint: one side absorbed, the other side capitulated, and the order book didn't have time to refill at the skipped prices.
Why it matters: skipped prices are unfinished business. The limit orders that would have participated at those levels didn't get the chance. They're still resting. When price drifts back, those orders tend to react — sometimes cleanly, sometimes not.
We mark every Up-IZ (price gapped up) and Down-IZ (price gapped down) on the chart. We don't trade all of them. We trade a specific subset, under specific conditions, filtered by rules that 15+ years of walk-forward testing has validated.
An IZ is the highest-information feature you can extract from a clean chart. Most other "patterns" are derivative.
Markets are auctions. An auction's job is to find the price where buyers and sellers transact in size. When price skips a range — leaves an IZ behind — the auction left work undone there.
Three real forces pull price back:
Not every IZ gets retraced. Some get blown through and never revisited — those are strong regime signals, but bad trades. We're not trying to catch every one. We're trying to identify the subset where the retrace is statistically likely and the reaction at the zone is tradeable.
Price returns to IZs because that's where the auction has unfinished business. Our edge is in filtering which ones matter.
Two trade types on the same zone. Different mechanics.
Initial Tag. Price has gapped (created an IZ), then drifted back to touch the zone for the first time. This is a mean-reversion play — we expect the resting orders at the unfilled prices to react. Risk is tight (just past the far edge of the zone). Win rate is the higher of the two trade types; reward is moderate.
Retest Continuation. Price tagged the zone, reacted off it, and is now coming back a second time. This is a continuation play — the first reaction confirmed the orders were real, and the trapped counter-trend traders from that first reaction are fuel for the next leg. Win rate is lower; payoff is asymmetric.
We trade one Initial Tag per zone. We don't stack multiple entries on the same zone. Walk-forward tests showed that allowing 2+ Initial Tags cut profit factor from 1.74 to 1.52 out-of-sample. Each additional entry shares risk with the earlier ones, and the orders that drove the first reaction are partially filled by the time the later entries fire.
The zone is a one-shot Initial Tag plus an optional Retest Continuation. Anything beyond that is dilution disguised as activity.
A 30-point stop on /NQ doesn't mean the same thing in a 12-ATR session as it does in a 4-ATR session. So we don't use fixed-point stops.
We anchor to the zone. For an Initial Tag, the stop sits a small ATR-derived buffer past the far edge of the IZ — far enough to survive normal noise, close enough that being wrong actually invalidates the thesis. The thesis says: "resting orders at this zone will hold." If price closes through the zone, the orders weren't there. The thesis is wrong. We exit.
We floor the stop at 8 points and cap it at 15. That floor and cap weren't picked by feel — they came out of a parameter sweep across the strategy's full out-of-sample window. Below the floor, normal session noise eats trades before they have a chance. Above the cap, we pay too much for a setup whose statistical reaction window is narrow.
Two practical consequences:
Stops should mean something. Anchoring to the zone makes the stop mean "the thesis is wrong." Anchoring to a point count makes it mean nothing.
Naive backtest: pick parameters that look good across all of history, then quote the resulting profit factor.
Problem: those parameters were chosen with knowledge of the future. You picked them because you saw what happened. The strategy looks clean because hindsight is doing the work.
Walk-forward fixes that. The rule is simple: at every point in time, the strategy may only see data from before that point. Train on a window of past data, test on the next chunk forward, slide the window, repeat. The reported result is a stitched-together series of out-of-sample predictions — predictions made without knowing what came next.
The numbers always shrink. Naive backtest PF on this strategy was ~2.6. Walk-forward, with stop logic matching the live configuration, came in at 1.86 ± 0.21 (mean ± stdev across 12 outer folds). Still a tradable edge. Also ~30% smaller than the naive number — which is the part most published track records don't show you.
If you ever see a strategy quoted with a single PF, no out-of-sample window, no parameter-stability test, assume the honest number is 30–50% lower. Sometimes there's nothing left after the haircut. Sometimes there is. The honest test is the one where the strategy doesn't get to peek.
We publish walk-forward numbers. We mark which window. We tell you when parameters changed and why. The number we report is the number a strategy that didn't know the future would have produced — because that's the only number that matters going forward.
Most retail trading education centers on US cash hours. There's a reason: that's when most discretionary traders are awake, and that's when their preferred patterns feel cleanest. Volume is high, spreads are narrow, the order book is dense.
But the Imbalance Zone mechanism doesn't care what time it is. The auction structure that creates an IZ — fast directional move, skipped prices, resting orders left behind — happens in every session that trades. /NQ trades nearly continuously, Sunday afternoon through Friday afternoon US time, and IZs form throughout.
So we run continuously. No time-of-day filter, no session gating. Every session is in scope.
What does change session-to-session is liquidity texture. Asia-hours volume is thinner; spreads widen; IZs form more often because lower volume produces more frequent fast moves, but the subsequent reactions can be choppier. London open tightens things up. US pre-market and cash hours are the densest book.
The strategy adapts mechanically rather than via session rules. Stops are ATR-anchored, so they widen in higher-volatility sessions and tighten in calmer ones. The same fractional risk translates to different effective dollar risk depending on when the setup fires. We don't need a session filter because volatility-adaptive risk sizing already handles the regime shift.
If a setup looks identical to the eye in two different sessions, the risk math underneath is doing different work. The strategy's job — and ours — is to trust that math, not to overlay a manual opinion about whether 03:00 ET trades "should" count.
Some directional moves start clean. The bar immediately before the move was opposing — a bullish move follows a red candle, a bearish move follows a green one — and that opposing bar absorbed flow right up until the break.
We call the last opposing-direction bar before an Initiative Move an Anchor Bar.
Why it's interesting: that bar is the level where the order book almost held, then didn't. Traders who sold into the red bar before the breakout are now underwater. Traders who bought the green bar before a breakdown are wrong. Their stops, plus the limit orders left resting at that level, become a magnet on any retracement.
When price returns to test an Anchor Bar, two things happen:
Anchor Bars aren't standalone signals. They're context — they tell you where to expect a reaction if one comes, not whether one will. We use them as confluence with Imbalance Zones, not as triggers in isolation.
Stops cluster at obvious levels: yesterday's high, the prior swing low, round numbers, the overnight extreme. Anyone who can read a chart can see them. Anyone running a stop runs it just past one.
A Pivot Sweep is what happens when price briefly trades through one of those levels, triggers the cluster of stops parked just beyond it, and reverses.
The mechanism is mechanical, not conspiratorial. When stops trigger, they become market orders in the direction of the sweep — supplying flow to whoever is providing liquidity at that price. The liquidity-provider's incentive is to absorb, not chase. Once the stop cluster is exhausted, there's no remaining flow to support continuation. Price reverses.
What identifies a real Pivot Sweep:
What it isn't: every penetration of a prior high. Plenty of moves break a pivot and keep going — those are continuation, not sweeps. The distinguishing feature is the close-back-inside.
When a Pivot Sweep happens into an Imbalance Zone, the setup compounds: the sweep cleans out one side's stops, the IZ provides resting orders for the reversal, and the resulting reaction tends to be the cleanest move of the session.
Most published track records are highlight reels. Wins shown, losses skipped, drawdowns minimized in the chart, edge cases waved away.
We publish the misses on purpose:
Three reasons.
One — survivorship bias is the default failure mode of trading content. If you only ever see wins, you'll calibrate your expectations to a fictional version of the strategy, then size up at the wrong moment, then blow up.
Two — knowing what failure looks like is part of the edge. A real strategy fails in characteristic ways. If you can recognize the failure modes while they're happening, you'll risk-manage differently than if every loss feels like an unexpected betrayal.
Three — it keeps us honest. The discipline of publishing every result, indexed and timestamped on a permanent track-record page, makes parameter-tweaking after a bad week harder to hide. The receipts are public.
A track record that only ever shows you wins is selling a story. A track record that publishes its full distribution is showing you a process. We're after the second.
We publish per-trade, daily, and weekly recaps. Numbers without context are noise — here's how to read them.
Per-trade entries show: setup type (Initial Tag or Retest Continuation), zone direction, entry, stop, exit, R-multiple. R is the trade's profit or loss in units of risk taken. A +2.0R win made twice what it would have lost if the stop hit. R is the apples-to-apples unit; raw point counts vary with stop size.
Daily recaps show: trade count, hit rate, average winner R, average loser R, day's profit factor, equity-curve update. Don't anchor to hit rate alone. A 35% hit-rate strategy with +3R wins and −1R losses is positive-expectancy. A 70% hit-rate strategy with +0.5R wins and −2R losses isn't.
Weekly roundups show: rolling 4-week PF, max drawdown from peak, trade count, and a walk-forward comparison (live PF vs. simulation's expectation for the same period). That last comparison is the key signal. Sustained live underperformance vs. simulation means something has drifted — execution, parameter set, regime — and we investigate.
What to ignore in any single window: day-level results, week-level results, sometimes even month-level. The strategy's expectancy emerges over hundreds of trades. Drawing conclusions from 5 is statistical noise. So is celebrating a 20-trade win streak.
We publish short-window results because they're the only honest way to keep the receipts current. We caption them with that caveat every time.
A strategy is defined as much by what it rejects as what it takes.
Wide-zone setups in high-volatility regimes. When the IZ is wide enough that an ATR-anchored stop exceeds the 15-point cap, the trade is rejected. Below the corresponding position size, risk-adjusted expectancy isn't economic.
Setups around major scheduled events. FOMC, NFP, CPI, PPI release windows produce liquidity vacuums and binary-outcome volatility that overwhelms structural setups. Our event calendar blocks new entries 30 minutes before through 30 minutes after each release. The expectation on those days might still be positive, but the variance and tail risk aren't worth wearing.
Setups during data-pipeline gaps. If tick capture has a hole, the simulation can't validate what happened. Trades that depend on bars inside a gap are skipped. Better to miss a real trade than take one based on phantom data.
Setups during contract rollover. The days around quarterly rollover have volume migrating between front-month and next-quarter contracts. Structure becomes ambiguous on whichever leg you're looking at, and IZs can form from rollover artifacts rather than real flow. We pause new entries on rollover and resume the day after.
If a rule rejects a setup that turned out to be a winner, that's expected. The rules are filters tuned for the distribution of outcomes, not for any single trade. A filter that never rejects a winner is a filter that isn't filtering.
Markets go through regimes — periods where underlying volatility, correlation, and order-flow texture all shift together. A strategy that's edge-positive in one regime can be edge-flat or edge-negative in another.
The right defense isn't "find a strategy that works in every regime." That strategy doesn't exist. The right defense is knowing how robust your parameters are across regimes, then accepting that periods of underperformance are part of the cost.
We test parameter stability by running the strategy across multiple non-overlapping historical windows and comparing results. If the optimal parameter shifts dramatically between windows — say, the best stop floor is 4 in one period and 12 in another — the strategy is overfit. Whatever value you pick will be the wrong one for the next regime.
Stable parameters look different. A sweep across stop floors {0, 4, 6, 8, 10, 12} produced a smooth curve with a wide plateau around 8 — meaning small variations in the parameter produced small variations in result. That's robustness. We're not standing on the edge of a cliff.
When live performance diverges from walk-forward expectation, the first question is regime, or bug? Regime is rarer than people assume. Most divergence we've investigated has turned out to be a data-pipeline issue, execution slippage, or a stop-logic mismatch between training and live — never the romantic "the market changed" narrative. We always check the boring causes first.
We can't tell you when the next regime change will happen. We can tell you the parameter set is robust within the regimes we've tested, and that the moment live drift exceeds threshold, we investigate before we tweak.
This is a systematic strategy. Entries, exits, stops, and sizing are rule-based and automated. In production, the system runs bar-by-bar without human approval.
So what's the role of discretion?
Strategy design. Choosing which mechanism to test, which features to engineer, which parameter spaces to sweep — human judgment. The data tells you whether an idea works; it doesn't tell you which ideas to try.
Filter design. The decision to filter FOMC, contract rollover, low-liquidity windows — pre-trade discretionary calls, validated by data but originally proposed by judgment.
Anomaly response. When the system's behavior diverges from expectation, a human decides whether to investigate, pause, or override. The system can't pause itself for "this looks weird" reasons it can't formalize.
Strategy retirement. When walk-forward expectation degrades over a sustained period and the cause isn't a pipeline issue, a human decides whether to retrain, retire, or reframe.
What discretion is not: skipping a system signal because it "feels wrong." Adding a trade because the chart "looks good." Tightening a stop because we're nervous. Widening one because we're hopeful. Every in-the-moment override of a systematic rule is, in expectation, a small subtraction from edge — because the rules were built to capture that edge.
The discipline is asymmetric: discretion in the lab, systems in the seat. The hardest part of running a strategy isn't designing it. It's not interfering with it.
Trading content often markets breadth: "we trade ES, NQ, RTY, YM, CL, GC — edge everywhere." The implication is that more instruments = more robust.
The reality is that most "multi-instrument" strategies are one strategy applied to multiple instruments without per-instrument validation. Same parameters, same filters, same setups. If the underlying mechanism transfers cleanly, that can work — but you're betting it does, and that bet is rarely tested as carefully as the original single-instrument result was.
We focus on /NQ for this strategy specifically because:
If we extend to other instruments later, it'll be after running the full validation pipeline per instrument, with separate parameter sets and separate published track records. We'd rather run one strategy on one instrument honestly than six strategies on six instruments with shared assumptions.
Depth over breadth, until the data justifies otherwise.