See what dealers are set up to do. Then check if it worked.
Every morning at 8:45 ET we turn the whole options market's positioning into a plain-English plan for SPY, QQQ and IWM: how big the day could be, the lines worth watching, and what changes if they break. Every claim behind that plan runs as a public experiment, locked before the data comes in and graded after. Misses stay up.
Loading…
Loading…
- 8:45 AMThe plan: regime, expected range and the lines to watch
- 9:30–4Live check-ins as positioning shifts, plus the crowd polls
- 4:20 PMHow the plan played, and tomorrow's preview
- SaturdayThe week ahead: Monday's map from Friday's close, big data, earnings and setups, picked by published rules
Daily gameplan
Today's levels, ranges and if/then scenarios for SPY, QQQ and IWM, built from dealer positioning before the open. Every plan gets graded after the close in the Lab.
New to GEX? Here's the whole idea in 60 seconds
- Market makers sell most of the options traded. To stay neutral, they hedge by buying and selling the stock itself, all day, automatically.
- How they're positioned decides which way that hedging pushes. In positive gamma they buy dips and sell rips, which softens moves: calm days. In negative gamma they sell into drops and buy into rallies, which adds fuel: wild days. GEX (gamma exposure) measures which one we're in.
- Levels are strikes where that hedging piles up. Price often reacts near them, so they're worth watching. Our own tests say: treat them as places to pay attention, not walls.
- How to use today's plan: read the one-line summary below (wild or calm?), check the expected range, and know the two key lines above and below price. Wild day: give trades more room, or trade smaller. Calm day: expect chop around the levels.
The one thing to remember: GEX tells you how big today's moves are likely to be. It doesn't tell you which way. We test that in public in the Lab.
Crowd polls 🗳️
Call what's coming: futures, the indexes, the big data drops and this week's earnings. Every poll locks before the event and gets graded after it, in public, just like our own calls. Anonymous, one vote per browser, and you can change it until it locks. For fun, not advice.
0DTE map
Today's same-day-expiry positioning for SPY, QQQ and IWM, written the way we post it. Built before the open. Hit Copy as post for X-ready text.
Pin claim on expiration days: . Values are from contracts expiring today, in shares (dollars of hedging per 1% move ÷ price). Accelerant: the heaviest negative strike near price, where a push can speed up. Wall: the heaviest positive strike above price. Treat these as lines to watch: on negative-gamma days they tend to break rather than hold. Structure, not advice.
This week
The whole week on one page: what kind of week it is, which day matters, where the indexes sit, and the setups worth watching. Educational, not advice.
Watchlist
Names with a catalyst or a notable GEX setup this week, and the level each one is closest to. Tap a card for the full read, or switch to Table to sort.
How this list is built rules, not picks
Any ticker
What the map says for one name, in plain English: the setup, what happens if it breaks either way, and every level explained. Pick a key, or type any ticker with listed options.
| Level | Price | From price | What it does |
|---|
🔥 Near-spot heat by expiration · blue cushions, orange fuels
GEX for the 11 strikes closest to price. Blue is positive gamma (dealers slow price down), orange is negative (dealers speed it up). Numbers are in shares: dollars of hedging per 1% move ÷ price.
Session replay
Watch dealer positioning move through the day: price against the gamma flip, the upside magnet and the downside pivot, snapshot by snapshot, from the positioning we record through the session. Space plays and pauses, arrow keys step.
The Lab 🧪
Every plan and every claim on this site, graded in public. Each test is written down and fingerprinted before its data comes in; misses stay up. Start with the verdict, then open any experiment for the receipts. Since Nov 18 every level comes from the T&G method v2. Earlier results stay as graded on the earlier feed and are never blended with new ones.
🔬 The full research · every experiment, every trade, every call tap to open
Four claims every GEX map makes, each tested against a baseline. Every map level is paired with a round-$5 strike that isn't on the map but sits the same distance from the open, and they're compared day by day. The question: does the map know something a nearby round number doesn't? Rules locked before the first run. Re-run after every close as the archive grows.
| Claim | What we measure | Map | Baseline | Side by side | p-value | Verdict |
|---|
Robustness checks were added after the first run to stress-test T1. They're shown for context and never change a verdict. p-value: how likely a gap this big would be by pure chance. Under 0.05 in the direction the claim predicts counts as "shows up". Under 20 cases on either side is too few to say anything.
| Strategy | Live: weeks | Live: total | Simulated: weeks | Simulated total (hindsight) |
|---|
The learner · week by week
| Week of | Negative-gamma rule | Positive-gamma rule | Learner | Always fade | Always break |
|---|
Setup league table · all historycontext only, never used for grading
| Regime | Setup | Trades | Won | Avg / trade | Total |
|---|
Fade a break of the 9:30–10:00 range once the day has used up enough of the range its gamma regime predicts. Four versions were scored on August–September only, and the script picked one by a rule written in advance. Caveat: the October–November check wasn't fully sealed (details below), so live results from Nov 6 on are the real test.
| Version | Design: trades | Won vs needed | Avg / trade | Holdout: trades | Won vs needed | Avg / trade |
|---|
Built on what Experiment 004 found: GEX seems to tell you how much price moves, not where it stops. So v2 uses the regime to choose the trade. On negative-gamma days it goes with a break of the 9:30–10:00 range. On positive-gamma days it fades the break. If knowing the regime doesn't beat "always break out" and "always fade", the map adds nothing.
| Strategy | Trades | Total (after costs) | Avg / trade | Won | Break-even | Total move |
|---|
Playbook v2 · every trade
| Date | Tkr | Regime | Trade | Entry | Target | Stop | Exit | Result | Move (before costs) |
|---|
Map · the earlier feed's open book, frozen at 9:30. Heavy strike: delta-adjusted OI 4× the 3 strikes on each side, in all expirations combined or any single one through Friday. Air pocket: the next two strikes past it are each ≤25% of it. Magnet: the control node.
Magnet pull · At 10:00, if price is 0.3%+ from the magnet, trade toward it. Stop: 5-min close beyond the nearest heavy strike behind.
Air pocket · 9:45–3:30, a 5-min close through a heavy strike into its pocket. Target: next heavy strike past it. Stop: 5-min close back through.
Out at 3:55. Target and stop in the same bar = loss. One trade per rule, per ticker, per day.
Experiment 003 · every trade
| Date | Tkr | Rule | Side | Entry | Target | Stop | Exit | Result | Move (before costs) | R |
|---|
| Strategy | What it does | Right | Hit rate | vs the pass line (60%) |
|---|
Experiment 002 · every call
| Date | Tkr | Move @3:30 | Forced flow | The earlier feed net GEX | Call | Last 30 min | Right? |
|---|
| Strategy | What it does | Right | Hit rate | vs the pass line (60%) |
|---|
Every rule is graded on the same calls with the same grading as the model. "Random pick" is the expected score of choosing EXTEND or ABSORB by coin flip. If the model can't beat the best of these, it isn't adding anything. ABSORB also counts as right when the close barely moves (under 0.10%), so it has a slightly easier test, and the two "always" rules can add up to more than 100%.
| Date | Tkr | Regime | Day range | Close | In range? | Level above | Level below | Played out |
|---|
Every 3:30 call
| Date | Tkr | Move @3:30 | Forced flow | Dealers | Call | Last 30 min | Right? |
|---|
Free model vs the earlier feed
| What we check | Free model (Cboe) | The earlier feed (API, 3:30) |
|---|
Days we missed
The rules, set before the datapass 60% · kill under 55% · 40 days
📜 The rules we hold ourselves tolock first · no verdict before the data · correct for luck · show everything
- Lock it first. Every test gets a card (question, metric, baseline, pass and kill rules, how much data it needs) before its verdict. The card's fingerprint is published. If a locked card changes, our own system refuses to publish.
- No verdict before the data. Each test states its sample size up front. Until it's reached, the result is "collecting", however good or bad it looks.
- Correct for luck. With many tests running, some pass by chance. Every p-value is adjusted for the number of tests (Holm). A pass needs adjusted p < 0.05. Two exceptions, stated: range checks (E007, E008, E013) test calibration, so they're judged on their raw p outside that family; and E003 and E006 have no p-value test, so they pass only if every locked leg holds.
- Only proven things drive the read. Each claim on the site carries the evidence level of the test behind it. A claim is "tested" only when its card passes, and loses that the day it fails.
- Show everything. Misses stay up. Changes get a changelog line. Our derived data and our code are open for anyone to check.
Method
Where every number comes from, how each call is made and graded, what the model gets wrong, and every change we've made to it. If something on this site can't be traced back to this page, that's a bug.
T&G Labs is the research desk of Trades & Gains. We publish dealer-positioning maps for SPY, QQQ and IWM before the open, post a preview of the next session after the close, and grade every plan and claim in public. Nothing is deleted or edited after the fact, and days we fail to record stay listed as missed.
Who pays for it. Trades & Gains. There are no ads, no paid tiers and no sponsors, and everything here stays free and public.
The site tests ideas. It does not sell trades. A level on this site is a description of where options positioning sits, not a prediction that price will go there.
Every level on this site is computed by us from the full option chain: every contract within 15% of price and 45 days out, with its open interest and implied vol. Nothing is copied from another platform's levels.
The assumption. Dealers are taken to be long the gamma of calls and short the gamma of puts, because customers mostly sell calls and buy puts. It's the standard assumption in GEX work, and it can be wrong on any one strike. We say so rather than hide it.
Vols. Each strike's implied vol comes from its out-of-the-money option (calls above price, puts below) and is used for both sides. In-the-money quotes are thin and noisy. A contract with no quoted vol borrows from the nearest strikes on the same expiry.
GEX. Per contract: sign × gamma × open interest × 100 × price² × 1%. That's the dollars of stock dealers must trade for a 1% move. Positive cushions moves, negative fuels them. Shares = dollars ÷ price.
Gamma flip. We re-price every contract's gamma at prices from 8% below to 8% above today's and take the crossing of zero nearest price. Calm above it, wild below it.
Edge. How far price is from the flip, measured in today's expected moves (price × 30-day at-the-money vol ÷ √252). At 1 or more, the day is firmly in its regime; under 1 it's near the edge, and one ordinary day can switch it.
Hedge load. Net GEX ÷ the stock's average daily dollar volume (20 days): the share of a normal day's trading dealers would need for each 1% move. It makes SPY and a single stock comparable.
Magnet, pivot, walls. Magnet: the biggest positive strike within 5% of price. Pivot: the biggest negative one. Each only counts if it's at least 10% the size of the biggest strike; otherwise the read says there isn't one. Call wall: the strike above price with the most call gamma. Put wall: below price, the most put gamma.
0DTE share and roll-off. Each expiry's share of the gamma within 2% of price: what leaves the board when it expires.
Desk read (v2.1). On top of the map, what a dealer desk watches:
- Near-term levels. Gamma from this week's expiries only, inside one weekly expected move of price. A level is named only if it's a peak (the biggest of its sign among the two strikes on each side) and a real size. Each side gets where a move can speed up (the biggest fuel strike) and where it can stall (the nearest cushion). The 45-day map still sets the regime and the flip.
- Vanna. The stock dealers must buy or sell to stay hedged if implied vol falls 1 point everywhere. When it's positive, a vol drop forces buying: the "vanna tailwind" after fear fades.
- Charm. The stock dealers must trade from one day passing, with nothing else changing. It grows into big expiries.
- Vol. 30-day implied vol against 20-day realized (the premium), 25-delta put skew, and whether front-week vol sits above the month (near-term stress).
- New positions. Day-over-day change in open interest by strike, from our own copy of each morning's chain.
Vanna and charm use Black-Scholes with each contract's own vol and the same dealer assumption as the map. They're estimates of pressure, not forecasts.
Checked twice. Before launch, the same method was run on a second, independent options feed: net GEX agreed within 2–4% and the flip within $1.50 on SPY, QQQ and IWM. We rerun that cross-check whenever the method changes.
Limits. Open interest updates once a day, overnight, so during the session the map moves with price and implied vol, not with new positions. Data is about 15 minutes delayed.
| Data | Source | When it's pulled (ET) | Notes |
|---|---|---|---|
| Option chains: open interest and implied vol for every contract within 15% of price, 45 days out. GEX, flip, magnet, pivot, walls, edge and hedge load are computed from them with the T&G method | Licensed options data (15-minute delayed) | 8:45 AM, live check-ins every 2 minutes during the session, 4:20 PM, evening preview | Open interest updates once a day, overnight, so intraday the map moves with price and implied vol, not with new positions. |
| Option chains: implied vol for the expected range | Cboe delayed quotes | 8:45 AM, 4:20 PM | About 15 minutes delayed. |
| Prices, daily high / low / close for grading | Yahoo Finance | 4:20 PM | Blank or bad bars are skipped, never filled in. |
| VIX | Yahoo Finance | 8:45 AM | Prior close. |
| Leveraged ETF sizes | Fund sites, entered by hand | Updated periodically | Estimates. They shrink after big down days, which overstates forced flow until updated. |
| Earnings and macro calendar | TipRanks, Earnings Whispers, Yardeni QuickTakes | Weekly | Entered by hand each weekend. |
Upside magnet is the strike with the most positive GEX within 5% of price. Downside pivot is the strike with the most negative GEX within 5% of price. Control node is the single largest strike-and-expiry cell this week. (Our playbook, Experiment 003, treats the control node as the magnet.)
Gamma flip is the price where total dealer gamma changes sign. We find it by re-pricing every contract's gamma at prices from 8% below to 8% above today's, each with its own implied vol and time left, and taking the crossing nearest price. Calm above it, wild below it.
Today's range is the at-the-money implied move for the session. Options price roughly a two-in-three chance of closing inside it.
Regime is the sign of net GEX: negative means dealer hedging tends to add to moves, positive means it tends to cushion them.
Rules first. The pass and kill rules for each experiment are published before the first call and never moved. Experiment 001 passes if big-flow EXTEND calls hit 60%+ over 40 trading days, and dies under 55% or if EXTEND does no better than ABSORB.
No hindsight. A gameplan only counts if it was built before 9:30 AM. A plan built later is still published, marked late, and never graded. Levels from the 3:30 map are graded on the next session, not the same one.
Baselines. Every call is also scored against simple rules (always EXTEND, always ABSORB, random pick) on the same days. The model has to beat them to mean anything.
Misses and gaps stay up. Wrong calls are never removed. Days we failed to record are listed and never backfilled. Imported history (the old Google Sheet) is shown for reference but not counted.
Suspect data is flagged. A gamma flip more than 10% from price is marked suspect and left out of flip stats.
Our free dealer model is naive. It assumes customers always sell calls and buy puts, which reads short gamma almost every day and pushes Experiment 001 toward EXTEND. Experiment 002 tests the same call with the earlier feed's dealer read instead.
Backtests flatter. The Experiment 002 backtest uses today's leveraged fund sizes for past days and can't account for anything the earlier feed changed in its own method over time. That's why only live calls count.
Small sample. Until 40 trading days are logged, every hit rate here is noise. Treat early numbers as a progress bar, not a result.
Delays. Option chains are about 15 minutes old, and live levels refresh every 2 minutes. Neither is a live feed. Open interest only updates overnight.
Two regime sources. Before late September, backtests read the regime from the earlier feed's first capture after 9:30; after that, from its open book. Rules fit on the older data are applied to the newer source.
ABSORB is easier to score than EXTEND. A 3:30 ABSORB call also counts as right when the close barely moves. Baselines are graded the same way, so comparisons are fair, but EXTEND vs ABSORB hit rates aren't apples to apples.
GEX is an estimate everywhere. No one sees dealer books directly. Every GEX number, ours and everyone else's, rests on assumptions about who holds which side.
📐 Every experiment's rules, as lockedExperiment 004 · learner · 005 · playbook v1 and v2
Map levels are the magnet plus the 5 strikes with the most delta-adjusted open interest within 3% of the open, from the earlier feed's open book. Baseline is every other round-$5 strike within 3% of the open, because map levels are mostly round numbers and round numbers attract orders on their own.
T1 · Regime and range. Is the day's high-to-low range, compared with its own 20-day normal, bigger on days the open book shows negative net GEX? Permutation test.
T2 · Levels hold. After 9:45, the first time a 5-minute bar trades through a level: did the next six closes stay within 0.10% of it? Map levels vs baseline strikes.
T3 · Magnet pull. Did the close end nearer the magnet than the open was? Compared with the same test on baseline strikes. Days opening within 0.1% of the magnet are skipped.
T4 · 0DTE pin. On days with same-day expirations, did price close within 0.15% of the largest same-day GEX strike? Compared with baseline strikes within 2% of the open.
Verdict. "Shows up" needs a p-value under 0.05 in the direction the claim predicts. Under 20 cases on either side reads "too few". Nothing here is tuned; a change would be a new dated version.
What it does. Every weekend it looks at every past session, nothing from the coming week, and picks one setup per gamma regime: skip, break out of the 9:30–10:00 range, or fade the break once the day has used at least U of its expected range (U: any, 0.4, 0.6, 0.8). That version is frozen before Monday and graded on the week.
How it picks. Best average % per trade in training, with at least 20 training trades. It starts at "fade at U ≥ 0.6" in both regimes. A fade threshold moves at most one notch a week, and switching setup type needs an average at least 0.02 points better.
How it's graded. Each week only by the version frozen before it. Past weeks reconstructed on Nov 8 are labeled "simulated"; weeks frozen before they happened are "live". Every version and its reasons stay published.
The split. Sessions before Oct 1 are the design set. Oct 1 to Nov 5 is the holdout (not fully sealed: see the Nov 8 note below). After Nov 5 is live. Four versions of one fade rule are scored on the design set only; the script picks one by a rule written in advance (highest average per trade, 20+ trades), then scores that one pick on the holdout, once.
Range used. U = (high so far − low so far) ÷ (20-day median range × R), where R is the average range multiple for that day's gamma regime, measured on the design set only.
The trade. After 10:00 and up to 3:30, the first 5-minute close outside the 9:30–10:00 range that meets the version's U condition. Fade it: target the range midpoint, stop on a 5-minute close one range-width past the break, out at 3:55.
Versions. A: U ≥ 0.6. B: U ≥ 0.8. C: U ≥ 0.6 on positive-gamma days only. D: U ≥ 0.4.
Pass. On the holdout, the pick must win at least 5 points more often than its break-even rate, average more than zero per trade, and beat "always fade" per trade. Anything less is a fail, published as one.
Why v2. Experiment 004's first run suggested GEX predicts how far price travels, not where it stops. v2 uses the regime to choose the trade instead of treating levels as walls. v1 keeps running with its own record.
Regime. Sign of net GEX in the earlier feed's open book, frozen at 9:30. Before late September, when no open book exists, the earlier feed's first capture after 9:30. Both are known before 10:00.
Opening range. High and low of the 5-minute bars from 9:30 to 10:00. The trigger is the first 5-minute close outside it, after 10:00 and up to 3:30.
Negative gamma: break out. Trade with the break. Target: one more range-width beyond it. Stop: a 5-minute close back past the middle of the range.
Positive gamma: fade. Trade against the break. Target: the middle of the range. Stop: a 5-minute close one range-width past the break.
Baselines. "Always break out" and "always fade" on every day, same trigger and exits. One trade per ticker per day, out at 3:55, scored in % move before costs.
The map. The earlier feed's open reference book for each session, frozen at 9:30, so the rules only ever see what was knowable at the open. Strike weight is delta-adjusted open interest, read for all expirations combined and for each expiration through that Friday on its own. A heavy strike has at least 4 times the average weight of the 3 strikes on each side, in either view. An air pocket is where the next two strikes past a heavy strike are each a quarter of it or less. The magnet is the control node. Only strikes within 5% of the open are used.
Magnet pull. At 10:00 AM, if price is at least 0.3% from the magnet, trade toward it. Target: the magnet. Stop: a 5-minute close beyond the nearest heavy strike behind price.
Air pocket. Between 9:45 AM and 3:30 PM, a 5-minute close through a heavy strike into its pocket. Trade through the pocket. Target: the next heavy strike past it. Stop: a 5-minute close back through the heavy strike.
Scoring. Out at 3:55 PM if nothing hits. If the target and the stop land in the same 5-minute bar, it's a loss. The headline is each trade's % move, added up. R (result divided by the distance from entry to the stop level) is shown too, but when an entry lands right at its stop level the risk is near zero and R explodes, so we also show the median R and the total without the three biggest trades. Per-trade moves and equity charts are before that cost; every total and win rate is after it. Prices are the earlier feed 5-minute bars.
No tuning. The thresholds (4×, 25%, 0.3%, 10:00) were set before any backtest. Changing them creates a new dated version; the old version's record stays up.
🧾 Trading costs and the independent audit0.02% per trade · what the Nov 8 review found
Every playbook, Experiment 005 and learner trade is charged 0.02% round trip, covering the bid-ask spread and some slippage on SPY, QQQ and IWM. Totals and win rates are shown after that cost; per-trade ledgers show the move before it. Wins are assumed to fill exactly at the target, which is slightly generous.
Before showing this publicly we had a separate reviewer, with no stake in the results, go through every experiment looking for ways the backtests could peek at information they shouldn't have, or count things wrong. What it found, and what we did:
Experiment 004's baseline favoured the map. Map levels sit near price; the comparison strikes were spread farther out. Fixed in v2: each level is paired with a round strike the same distance from the open, compared day by day. The regime test now counts one case per day, because SPY, QQQ and IWM move together.
Experiment 005's Oct–Nov check wasn't fully sealed. We'd already seen "always fade" results over that period before locking it. It's now labeled that way; live trades from Nov 6 are the clean test. Its pick is saved and can't change.
The learner's simulated weeks use hindsight. Its starting rule and settings were chosen after seeing that period. Only live weeks, starting Nov 9, count.
No trading costs in earlier versions. Now charged on every trade.
A 3:30 call could be re-run or logged late, and a missed 4:20 run left days ungraded. Now a call is only accepted between 3:25 and 3:45, is never replaced, and missed days are graded on the next run.
Still being checked: whether the earlier feed's stored history matches exactly what was available live (we compare our same-day copies against fresh downloads), and that 5-minute bar times mark the start of each bar.
🤝 Use today's read anywhereembed card and JSON feed, free
Drop-in card. One line shows today's read (wild or calm, each index's two lines, today's range) on any page, and it updates itself every morning:
<iframe src="https://tg-labs.netlify.app/embed.html" width="100%" height="330" style="border:0"></iframe>
Options: ?only=SPY for one index, ?theme=light for light backgrounds.
Data feed. The same read as JSON, open to any website: tg-labs.netlify.app/feed/today.json. Updated at 8:45 AM ET on trading days and on Sunday mornings (preview). Fields only ever get added, so anything built on it keeps working.
Credit line when you use it: "T&G Labs read." Educational, not advice.
🗄️ Retired: how the 3:30 call was madeclosed out Nov 10, still graded in the Lab
Forced flow is how much leveraged ETFs must buy or sell into the close to keep their leverage: fund size × leverage × (leverage − 1) × the day's move. QQQ also counts chip and single-stock funds that trade through QQQ names, using each one's own move.
The call. If dealers are short gamma, we call EXTEND: the last 30 minutes move with the forced flow. If they're long gamma, we call ABSORB: the move stalls or reverses. Flow under $50M is too small to call.
Graded at the close. EXTEND is right if the 3:30-to-close move goes the same way as the flow. ABSORB is right if it goes the other way or moves less than 0.10%.
📝 Changelogevery change to how calls are made or graded
| Date | Change | Effect on past results |
|---|
Built and run by Trades & Gains (Ben The Bull). Questions, corrections or a bug in the numbers: reach us on X. If we got something wrong, we'll say so in the Lab.
Educational only. Nothing on this site is investment advice or a recommendation to buy or sell anything. Options and leveraged products carry substantial risk.
Learn GEX
A five-lesson first week: each one takes about five minutes and uses today's real numbers on this site. Then every term on this site, in plain English.