Tested findings
Every green cutout in one place, what held, where, and how long.
How to read this page
Every green cutout in the manual is collected here. Each one is a replay against history, walk-forward, no hindsight, the same rules the desk runs today, and each states the coarse window and the result. What is not published is the selection rule behind a board entry; those stay inside the engine. Two eras matter most: the 2022 grinding bear and the 2020 dislocation, because most tools are only ever tested in the bull market that built them. Where a finding held in both, we say so; where it didn't, we say that too.
Trades: the front door
A trade line, left to right, chapter ↗
Tested: shapes are listed only where the trades the desk takes (the top-ranked ones, after costs) test positive in the current tape, and where the whole book tests no worse with them than without. The number on the group header is that one; the engine page shows beside it the average of every trade that qualifies, which can differ (hedged shorts in a trending tape: slightly negative on average, positive as taken). Trending and Flat: strength runs and hedged weakness runs. Grind: runs and cycle pairs. Dislocation: cycle pairs and the bounce; runs are not offered there. Three more rules sit on top: a run is listed only in a sector whose runs tested positive over the whole window (Energy and Staples runs are not listed; open the name and hedge it by hand if you want the relative view); a run is not listed from a sector that is in a dislocation of its own while the market is not (the sector fund read by the same four-state tape as the index; a run inside a sector that is breaking tested worst of all), and the line under the list names any such sector; and a name is in one trade at a time. The full table, the sector table and the red cells are on the engine page. Type any name under the list and the engine walks its rules for it. Every stock in the TradeMath Universe also has a page of its own with today's read, its record on each shape since 2019, its last runs, the house book's trades in it and its true peers.
Alpha: the number
The three dials, chapter ↗
The chorus has been worth listening to. In our replays, runs the checks agreed on lasted about twice as long as runs they split on, in every tape. On the long side the agreed runs also beat the market by a clearly wider margin; on the short side that pattern has not held, so read bearish agreement as description, not endorsement. On the trade line this is the run clock: agreement says how long, not how much.
How to use it, chapter ↗
The exit that tested best for a long run: leave when Alpha closes back under 60. We made exits compete on identical fresh-run entries. In a trending tape the fixed 21-day hold earned the most in total, but the Alpha-under-60 exit kept most of that edge in less than half the time, with half the worst case, the best return per day held. In the 2022 grinding bear every fixed hold lost, worse the longer you held, and this exit cut the loss by two-thirds. Robust in both weathers, so it is the rule on every long line and the one the tracker flags. There is no payoff window to wait for on a run: the edge accrues while the run persists and stops when the score gives up 60.
True strength: the relative strength chart
Compare mode and the spread lane, chapter ↗
A red net β is a quality caution, not a hedging to-do. In our replays, pairs entered with a large lean were the strongest cohort over the first two weeks and the weakest by a month, with the heaviest tails of any group. A big lean usually means one leg's beta has collapsed or exploded: the pair is lopsided and its payoff is front-loaded. Cancelling the lean with a third leg did not fix that. The tested read is simpler: on a red chip, prefer the front of the payoff window and let the Narrowing rule take you out early rather than late.
Reading a name
Line by line, chapter ↗
"Very Extended means chase risk" turned out to be true in one tape and false in another. We replayed fresh Strength runs by their extension at the start. In a trending tape the already-stretched names carried the edge, Very Extended starts earned about +2% over the market in three weeks while early-move starts earned nothing. In the 2022 grinding bear it was the mirror: stretched starts were the worst cohort and discounted starts the best. So read Extension through the tape: chase strength when the tape is trending; in a grind or dislocation, avoid the stretched names and look where the discount is. The trade line's "why it qualified" says which case you are in.
Hedge pairs
Why pairs, and why not in an up-tape, chapter ↗
In testing, wider score gaps preceded larger forward spreads, and unlike single-name signals the edge persisted over multi-week horizons, because market drift is out of the equation. Freshness matters: pairs with both legs early in their moves carried by far the strongest forward spread, while pairs built from two exhausted moves showed none.
Why pairs, and why not in an up-tape, chapter ↗
The one place the pair fails is a trending tape, and the reason is the short leg. We hedged the same fresh-leader entry four ways, with nothing, the index, its sector, and a laggard stock. Hedging with the index or the sector kept the run's number positive in every tape but a dislocation; hedging with a laggard turned it negative in an up-tape, because laggards get bought in rotations. So the trade line quotes runs plain with an index or sector hedge as a switch, and brings the laggard pair back in flat and grinding tapes, where selling the weak name pays.
How to trade it, chapter ↗
Give it two weeks. The first days of a new pair did little in testing; the spread's gains concentrated in weeks two and three, and by about a month the typical spread had stopped growing and began to give back. An expectation, not a schedule, individual pairs pay early, late, or not at all. The lane, not the calendar, has the final word.
How to trade it, chapter ↗
The exit that tested best: three straight Narrowing days, or day 21, whichever comes first. It fires on winners too, and that is its best case, the tracker says exit with the win when you are ahead. We made sixteen exits compete on the same replayed trades: caps, two Narrowing days, hard stops, trailing give-backs, profit takes, an exit on the legs' own Alpha. Every one landed within a fifth of a percent of this rule. Exits change the shape of the outcome, not the mean, the leg exit halves the worst case at a third of the holding time and is a candidate for grinding tapes once a third such tape is on record. Price stop-losses tested worse than no rule at all, because a spread breathes with noise and a stop locks the breathing in. Leaving the moment a pair drops off the board was far too early; treat that as a review prompt.
How pairs enter and leave the board, chapter ↗
Tested, not taste: pairs retained through a wobble went on to outperform fresh entries over the following weeks. So leaving the board means something, the thesis died: the laggard recovered, the leader broke, or a leg went stale. Even then, treat the departure as a review prompt; the exit remains the three-day Narrowing run on the spread itself.
How pairs enter and leave the board, chapter ↗
Timing matters too. After a sector's violent surge, its higher-beta names tend to give part of it back over roughly three weeks; after a violent washout they tend to rebound over roughly two. The flags name the name, ⚡ COIN headwind ~3wk: sector surged, and whether that helps or hurts depends on which leg it is. Information, not a gate.
Split sectors and compressed sectors, chapter ↗
The pairs that struggled most in our testing shared a profile: a gap that barely qualified, drawn from a compressed sector. The same modest gap inside a split sector held up better. Early evidence, one stretch of history, so the cards do not gate on it; the Market tab's sector cards carry it as the split and quiet tags. When a sector is quiet, ask more of the gap.
The Relative ticket: a name's own basket, chapter ↗
The hedge is insurance in the strict sense. Against the plain directional long across twenty years of tapes: in rising tapes the hedge costs, a strong name's sector and peers are strong too, and shorting them subtracts real drift. In flat tapes it costs a little. When the market breaks the ranking flips hard: in the 2020 crash the hedged forms made money while unhedged leaders took their worst losses, and in the 2022 bear hedges roughly broke even while plain longs bled. Everywhere, hedged worst cases were far smaller. That is why the trade line quotes a long plain and offers the index or sector hedge as a switch with its own tested number: you choose the premium with the number in front of you.
Reading the tape
Trending, chapter ↗
In our twenty-year replays, pair entries made on trending days were the best cohort in every era they appeared in. The one honest footnote: hedged single-name expressions cost the most here, a strong name's sector and peers are strong too, and shorting them subtracts real drift. Trending tape = board at full strength, Relative tab as insurance only.
Flat, chapter ↗
Flat tapes tested as thin pickings: pairs held for the full window made roughly nothing (the 2015–16 replay: +0.1% per trade over two years), and nothing else did much better. Not dangerous: just unrewarding. Expect less from every idea, keep sizing modest, and let the Narrowing exit rule take you out rather than patience.
Grinding bear, chapter ↗
This is the one tape where the hedge-pair book tested weakest: in the 2022 replay, pairs held flat for the window lost about −0.8% per trade. Two things earned their keep here. The three-day Narrowing exit halved the bleed and brought the book to roughly breakeven, it matters double in a grind. And hedged single-name expressions roughly broke even while plain longs lost −2.5%. Posture: fewer, more skeptical ideas; exit on the rule without negotiation; lean on the Relative tab.
Dislocation, chapter ↗
In the 2020 replay, market-neutral pairs made about +1.8% per trade through the crash: market-neutrality doing exactly its job, and hedged single-name expressions were the best of all forms (+3 to +5%) while unhedged leaders took their worst losses (−3.4% with tails past −40%). Speed matters: dislocations resolve fast, so the exit rule and the front of the payoff window carry the edge. The board keeps running; the Relative tab moves from insurance to advantage.
Dislocation, chapter ↗
The trap inside a dislocation is the recovery, and which kind you get decides which pairs pay. In the V-shaped rebound of Mar–Jun 2020 the weakest names snapped back to their peers: pairs whose legs were true peers (◆) lost about 1.4% per trade with the exit and won one time in three, while pairs between unrelated names in the same sector made +0.9%. Long the strong true peer, short its weakest ones, on fixed holds, lost 6–8%. But in the long grind of 2008–09 the same ◆ pairs were the best subset (+1.0% vs −0.7%, half the tail): the leader kept leading its own peer. So the rebound risk is real and the shape of the recovery is not knowable at entry. What is knowable: ◆ pairs carry about half the swing in every era, the board never shorts Deep Discount names, and the three-day Narrowing exit cut the worst case by a third across the board in 2020. Across the five dislocations we could replay (2008, 2011, 2015, 2018, 2020), the one pattern with any consistency: in the rebounding half, ◆ pairs tested worst in two of three episodes with enough of them (2020 −1.0%, 2011 −4.0% with no winners; 2008 the exception). Don't add to shorts after the worst day; take the exit when it fires; treat ◆ as a tighter hedge, not a stronger thesis, and read it with extra care once the tape says rebounding.
Dislocation, chapter ↗
What the desk lists in a dislocation. Since Bernoulli, three shapes and only three, all hedged, none a plain long. The cycle pair, a fresh riser against a stale leader, is offered here because the hand-off inside a sector kept paying while the market fell. The dislocation bounce, the weakest tenth of a sector over five sessions, held against the sector fund, out when the five-day move is back to zero or on day 5, is the desk's answer to the snap-back described above: instead of being caught short the name that squeezes, it owns it, for days not weeks. It tested positive in every dislocation we could replay, and only clears its costs there. The sector pair (2.5) is the third: a sector fund whose own tape is Trending against a sector fund whose own tape is a Grind or a Dislocation, dollar for dollar, two at most, out when either sector's tape leaves its state or on day 21. A dislocation is when sectors disagree most, and the trade lives on that disagreement; the exit when the two tapes re-agree is what makes it pay, and held to the cap it tested negative. It improved every replayed dislocation (2018, 2020, 2022) and touched nothing else. Runs are not offered in a dislocation at all. Open positions keep their exits.
The one thing the tape changes in the score, chapter ↗
Why so narrow: we replayed the alternatives. A blanket "trust bullish reads less when the tape is bad" removed the entries that actually worked, in the 2022 grind the marginal names (Alpha 60–70) did better than the strong ones. Cutting stretched names everywhere outside a trend was wrong too: in a Flat tape the stretched names were the good ones (+0.8 to +1.6% over the market in three weeks). Only one cohort was reliably bad, Very Extended bullish entries in a grind or a dislocation, about −3% vs the market in three weeks in both 2022 and 2020 while everything else sat near flat. That is the one the engine now trims.
Returns are not the number
The worked example: the desk's rules against buy-and-hold, chapter ↗
What this table does not say, and the desk will not claim: that the book beats the index in a bull year. In 2019, 2023 and 2024 the index simply outran it, and there are stretches, a summer rotation, for one, where the book loses while the index rises. A strategy that promises to beat buy-and-hold every year is selling you its beta. This one promises most of the return with a fraction of the pain, every rule in the open, and a live record you can check on the engine page.