What was tried
Most of this did not work, and that is the record.
A dated account of every idea pursued, and how each one ended. Most were refuted outright, and most of those were caught by this project testing its own favourite ideas harder rather than by anyone else.
The shape of it in one line
Block the channel, then rescue it from a distance, then stabilise it as it folds, then grip the broken site with a small molecule, then grip it with a rigid one, then discover that the idea underneath all of that does not hold at this position, and that the deciding question is one afternoon in somebody else’s laboratory.
The timeline
What was believedThe structure being used showed the region of interest well enough to reason about.
What was doneChecked which structures actually resolve the relevant part of the protein.
What happenedThey did not. The whole project was rebuilt on a different structure, the one that resolves the position in question. Every measurement since descends from that fix.
This one was caught by the project itself.
What was believedThe variant breaks a buried contact, the protein misfolds, and less of it reaches the cell surface.
What was doneMeasured the contact geometry, ran the first simulations, and opened two lines of attack: an existing drug that might stabilise the protein, and a second genetic change that might cancel the first.
What happenedThe geometry held up. Both lines looked plausible. This is the most optimistic the project ever was.
What was believedA drug binding the well-studied part of the channel might rescue a fault in the distant part.
What was doneMeasured how mechanically connected the two regions are, using a free method that runs on a laptop.
What happenedThey are barely connected at all: the signal between them was indistinguishable from noise, and thousands of times weaker than a genuinely coupled pair. The two regions are about 53 angstroms apart. A costly simulation booked to test this was cancelled before it ran.
What was believedIf the drug stabilises the channel, the simulation should show the protein becoming more rigid.
What was doneRan the test. Then questioned the test rather than the drug.
What happenedThe result was null, and the conclusion was that the measurement was on the wrong axis rather than that the drug was useless. Worth recording because that reasoning could have been motivated, and it was later overtaken anyway.
This one was caught by the project itself.
What was believedThe existing drug will stabilise the broken region directly.
What was doneTested it against the actual site, on three independent measures, alongside several alternatives.
What happenedIt failed on all three. That closed the most attractive idea in the project, the one with an existing safety record and a published precedent in the same gene. A small molecule found in the same screen looked excellent instead, and became the new lead the same evening.
What was believedAmong the millions of purchasable compounds, some will grip the site the way the missing piece did.
What was doneTwo independent exhaustive searches over hundreds of candidate molecules.
What happenedZero. Neither search found a single purchasable compound that could make the required double-sided grip. Only purpose-designed molecules could, and none of those can be bought.
What was believedThe new lead compound holds the site. It scored beautifully in the cheap test.
What was doneRepeated the test three times in a full, realistic model rather than the simplified one.
What happenedIt let go. Occupancy fell from 0.89 in the cheap test to 0.10 across three proper runs, and a free-energy calculation put its preferred position well away from the site entirely. The cheap test was retired for having produced a false positive, provably.
This one was caught by the project itself.
What was believedAn approved, cheap, widely used drug holds the site. Recorded that morning as the first approved-drug hit.
What was doneLet the simulation run to full length instead of reading it mid-run.
What happenedBy the second half of the run the compound had left completely. Occupancy zero. The claim was retracted the same day it was made, and every document leading with it was corrected.
This one was caught by the project itself.
What was believedThe purpose-designed molecule that survived everything so far is the lead.
What was doneRan it three times instead of once.
What happenedIt held in one run of three. Both other runs escaped. Demoted from lead to proof of concept, with the explicit instruction not to present it as a top pick.
This one was caught by the project itself.
What was believedThe compounds that survived the short test will survive a long one.
What was doneRan them five times longer.
What happenedOne confirmed survivor escaped anyway. Length caught what replication could not. Separately, every one of 20,340 project files was audited, and the honest count was published: zero compounds had cleared all six stages, because the final stage had not been built yet.
This one was caught by the project itself.
What was believedThe ranking of the surviving compounds is settled.
What was doneRe-ran the free-energy calculation with a method where the simulations exchange information, after noticing the original windows were barely communicating.
What happenedThe ranking inverted. The compound labelled the winner became second, and the one labelled a mere candidate became the deepest. The labels had come from a calculation that had not converged.
This one was caught by the project itself.
What was believedA specialist might reply in a couple of weeks, if at all. Two separate predictions put the odds of a useful reply at somewhere between a quarter and a half.
What was doneSent it anyway, with the negative results left in rather than trimmed out, and against written advice to use a different channel.
What happenedA reply came the same morning. The clinician forwarded it to a laboratory that had already run the key experiment, and that group checked the structural claim themselves, corrected it, contributed a contact nobody here had noticed, and offered to talk about what came next. Both odds predictions were wrong and the channel advice was wrong.
What was believedPredicting how much a change destabilises the protein identifies which variants break it. Six weeks of screening rested on this.
What was doneTested the method against four variants already known to break this channel, chosen in advance so the answer key could not be adjusted afterwards.
What happenedIt got one right out of four. Two of the misses ranked as the least damaging substitutions possible at their own positions. A second method would have reversed the conclusion entirely, which is the strongest argument for testing against reality rather than against another prediction. The premise underneath the whole compound campaign does not hold at this position, and a document written weeks earlier had already named this exact condition as the thing that would collapse the framing, with the instruction to test it first.
This one was caught by the project itself.
What was believedRaising output from the healthy copy sidesteps the unresolved mechanism, so it should work regardless.
What was doneMeasured the raw material it depends on across 827 human heart samples, having stated in advance that a few percent would kill the route.
What happened0.0045 percent. Around 300 times less than in the tissue where the same strategy already works in children. The ceiling is a 1.06-fold increase where roughly 1.5 was needed. The route was attractive because it was mechanism-independent, and it is dead because the substrate is absent. Those are separate reasons.
The hazard that closed it, rather than the disappointment
The compound idea did not merely fail to work. If the broken copy interferes with the healthy one at the cell surface, then helping more broken protein get there could make the problem worse rather than better. That is a reason to stop, not a reason to try harder, and it is independent of whether any particular molecule gripped anything.
What the pattern says
The corrections got deeper over time. The early ones are operational: a wrong lookup key, a full disk, a time limit that wiped nine jobs at once because nobody had turned on checkpointing. The middle ones are results: a compound retracted the same day it was announced, then the project ’s own lead demoted, then a ranking inverted by a better method.
The last one is the premise. The framework that justified six weeks of screening was tested against an answer key chosen in advance and got one call right in four. Nothing downstream of it survives that.
What survived
Not much, and that is a real answer rather than a failure. Two editing designs stand on their own and are described on the routes page. A measurement of a public database, which has nothing to do with this variant, turned out to be the most widely useful thing here and has its own page. Two costed experiments are written up in full, either of which would settle something.
And a large number of routes no longer need walking. For a project run by one person, deleting the wrong paths faster than you pursue them is the only way to reach anything at all.
Why publish the failures
Because the alternative is a record nobody can check. Every result here that went against me is written up with the number that killed it, and the papers carry the same discipline: each one names, in advance, the result that would prove it wrong.
If you spot something in this history that is still wrong, I would rather know: ethan@brugada.net.