There is always another pair. Another timeframe, another session, another filter, another indicator to combine with the last one. The search space feels infinite, which makes it very hard to ever declare anything finished — and an unfinished search is how people spend years on a strategy that was answered in the first month.
It is not infinite. Pairs, sessions and timeframes are small, enumerable sets, and you can get to the end of them. Knowing you have reached the end is a result, and it is worth more than most positive findings, because it is the only thing that lets you stop and work on something else.
This module is about finishing: how to map an axis instead of wandering along it, how negatives combine into something positive, and why the most valuable sentence in our notes is usually an instruction not to repeat a test.
Map the axis, do not walk it
The wrong way to explore timeframes is one at a time, stopping when you find one that works. The right way is to run the whole axis and look at the shape.
Here is ours, published as the rejection reason on the M30 version:
M30 with the proven M15 values (MaxHold 480, cooldown 60), London 07-13 server, long only,
2025-26: USDJPY 38 trades PF 1.27 +$16; EURJPY 41 / 0.99 / -$0.4; EURUSD 28 / 0.28 / -$25.
Against M15: 1.69 / 1.57 / 1.44. Timeframe map is now complete: M1 0.59, M5 ~1.0, M15 ~1.5,
M30 ~0.8. M15 is the peak, not a point on a rising curve.Read the sequence: 0.59, 1.0, 1.5, 0.8. That is not four separate results, it is one curve with a peak in the middle, and the shape says something a single test cannot. If the numbers had risen monotonically — 0.59, 1.0, 1.5, 1.9 — the honest conclusion would be "we have not found the top yet, keep going". They do not. The strategy has a natural horizon, M15 sits on it, and the axis is closed.
The last clause is the part to copy: "M15 is the peak, not a point on a rising curve." Having the neighbours on both sides is what converts a good result into a finished question. A peak you have only approached from one direction is not a peak, it is the last thing you tried.
That is also why the losing runs had to be filed. The M1 and M30 rows are not clutter on the catalogue; they are the two data points that make the M15 result mean something.
Find out which dimension owns the failure
When a strategy fails somewhere new, the useful question is which axis the failure belongs to — because that determines whether you have closed one box or the whole column.
Two of our pair rejections read almost identically, and the wording carries the finding:
Proven EURUSD values unchanged. 2025-26: 49 trades, PF 1.49, +$19.7. Out of sample 2023-24:
75 trades, PF 0.75, -$19.9, three of four half-years negative. Fails rule 3 and rule 8. The
edge on this pair is a 2025-26 phenomenon, not a property of the pair."Not a property of the pair" is a claim about which dimension owns the result — in this case neither the pair nor the strategy, but the period. A pair that works in one window and fails in the prior one has not told you anything about pairs at all.
Contrast that with the session work, where three pairs failed together in a new session having all worked in the old one. Three simultaneous failures along one dimension close that dimension; one failure closes one cell.
The practical form of this is that the catalogue itself becomes the map. Our reversion strategy currently has twenty-seven published versions spanning four timeframes, three sessions and eight pairs — eleven of them rejected outright. The search space is not in anyone's head; it is a table you can sort. That is what makes "we have tested this" checkable instead of remembered.
Negatives compose into a theory
Here is the part that makes systematic rejection worth the effort. Individually, a negative result closes a door. Collected, they describe the shape of the room.
We have tested four events that all have a scheduled clock. One works. The three that do not fail for reasons that turned out to be the same reason.
The Tokyo 09:55 fix works — importers and their banks have to transact at a published rate, in a direction set by trade flow, on dates the calendar fixes in advance.
The London 16:00 fix does not survive out of sample: month-end rebalancing has a clock, but the direction depends on which way markets moved that month, so the flow is two-sided and variable.
The Tokyo 15:00 option cut does not work either, and the study says why:
Option expiry has a clock and a participant, but the dealer flow is two-sided (long and short
gamma at different strikes) and strike-dependent, and we have no strike data. Without knowing
where the open interest sits, the cut is just a slightly quieter half hour. This is the same
lesson as the London month-end fix: a clock is necessary, a one-directional obligation is what
pays.And the ECB reference rate fails for the cleanest reason of the four:
The ECB reference rate is published for information, not settled against; no participant is
obliged to trade at 14:15. Same verdict as the Tokyo cut: a clock without an obligation.Four studies, three of them rejections, and out of them a rule that is worth more than any of the individual results: a clock is necessary; a one-directional obligation is what pays.
That rule is now a filter on future work. Any scheduled event can be assessed before a line of code is written by asking whether someone is obliged to trade, and whether their direction is fixed in advance. Most are not, which is why most fixes are not tradeable — and knowing that in advance is the entire return on those three rejections.
The same accumulation happened on the other side of the map. Trend following, pullbacks, breakouts, daily-bias gating and multi-week momentum were each tested and each failed, and the conclusion was drawn across them rather than one at a time:
The map's "continuation" column is now fully closed: trend following, pullbacks,
breakouts, daily-bias gating and multi-week momentum all failed, on their own data.Five rejections that cost real work, and what they bought is a column of the map that nobody has to revisit. You can read the whole thing in the techniques map and the four-technique comparison.
A negative can explain your positive
One more thing rejections do, which is easy to miss: they tell you why your working strategy works.
We trade a dip in the London morning. A popular explanation for that kind of move is the liquidity sweep — price runs the stops below an obvious low, then reverses. It is a satisfying story and we tested it directly:
It also answers a question from the map: the London-morning dip we trade is **not** a sweep
effect. Sweeping the Asian low in that window and buying is worth about nothing; the
RangeRevert entry's money comes from the band distance, the RSI floor and the ADX ceiling,
not from the level being taken out.The strategy was profitable before that study and profitable after; nothing changed in the code. What changed is that one plausible explanation was eliminated, which matters because you will eventually modify a strategy according to your story about why it works. If the story is wrong, the modification is wrong — you would add a sweep-detection filter, remove a third of the trades, and be genuinely surprised when the edge got worse.
Testing your own explanation is not academic. It is protecting the strategy from your next good idea.
Write the instruction, not the verdict
A finished line of research needs a note, and the note has to do more than record the outcome. Compare two ways of writing the same result.
The verdict: liquidity sweeps do not work. That is nearly useless in six months. Sweeps defined how? Tested on what? At which timeframe? A future you with a slightly different definition will simply run it again.
The instruction, which is what we actually wrote:
Rejected as a mechanism. No engine built, nothing filed on the presets page (there is no
run to file). Recorded in the working notes and memory so it is not re-tested with the same
definition.Note the precision of the scope. Not "sweeps do not work" but "not re-tested with the same definition" — and the study elsewhere is explicit that if sweeps carry information, it is below the spread or needs order flow data we do not have. That is a closed door with the conditions for reopening it written on the front.
The same applies to the pair screen. The note does not say "no other pairs work"; it says do not screen more pairs with these values, which leaves the door open for a genuinely different entry while closing it on another afternoon of the same work.
A good rejection note says what was tested, what would have to be different to justify testing it again, and what the result implies about the map. The first line saves you the repeat; the second keeps you honest about what you actually ruled out.
What closing an axis buys you
When the pair, session and timeframe axes for our reversion strategy were all closed, the note ended with a sentence that is the whole point of the exercise: the next lever is engine work, not more screening.
That is a real decision with real consequences. It means no more weeks of running the same entry against a new symbol hoping for a better number. It means the honest options are a different entry, a different mechanism, or accepting what is there — and it was only available because the map was complete enough to say that screening was exhausted rather than merely tiring.
Most abandoned trading projects are not abandoned at a conclusion. They are abandoned at the point where the researcher runs out of enthusiasm, with the search space half-covered and no record of which half. Closing an axis deliberately is the difference between finishing a question and losing interest in it.
What you should have now
The habit of running the whole axis rather than stopping at the first success, so the neighbours tell you whether you are on a peak; the discipline of asking which dimension owns a failure before generalising; the expectation that a set of rejections will compose into a rule worth more than any of them; the willingness to test your own explanation of a working system; and notes that say what would have to change before the question is reopened.
And the reframe the whole course has been circling: a negative result is not a wasted month. It is the only kind of result that permanently reduces the amount of work left to do.
Next module: what the map looks like when the screening is finished — the one mechanism edge we have, the three rejections around it, and an honest account of what we still do not know.