brugada.net · What’s standing

Updated 16 July 2026

Four survivors. Probably three. Arguably two.

The previous round claimed four compounds survived its longest test. The files underneath that claim don’t agree with it, and this page is why.

Here is the short version for anyone who doesn’t want the forensics.

A scoreboard file in this project is labelled the “verdict of record”. It says four compounds passed the long test. When I went and read the actual simulation files it was supposedly summarising, they told a different story — and for two compounds, the opposite story.

Tegaserod

The scoreboard calls it a survivor. Every simulation that still exists shows it letting go — including the longest run in the entire project.

Alniditan

The scoreboard calls it the shaky one. Every simulation that still exists shows it holding. The number the scoreboard cites as its failure appears in no file anywhere.

ZINC000006819928

Recorded as three successful runs. There is exactly one. It’s clean — but one is not three, and nobody should round it up.

ZINC000002430405

Isn’t on the survivor list at all. Three runs, all holding. On the surviving evidence it outranks two compounds that are. Nobody has looked at it.

And the fact that undercuts the whole stage: the test was specified as 100 nanoseconds, three times over. No simulation ever reached 100 nanoseconds. The longest is 90.5. Most are about ten.

This is not fraud, and I don’t think it’s even carelessness. An automated cleanup script deleted several finished simulations before anyone had extracted their results — the log records it happening. So those scoreboard numbers may well be real measurements of runs that no longer exist. Unverifiable isn’t the same as invented, and I’m not going to pretend it is. But it does mean the headline result can’t currently be reproduced from the project’s own files.

The picture

Every long-test simulation that still exists, drawn against the time it actually ran rather than the time it was supposed to. Blue holds the grip. Pink lets go. The dashed line is where 100 nanoseconds would be.

Every harvested extended-residence seed, drawn against the time it actually ran One strip per simulation seed for the four candidates that reached the residence stage. Cyan traces hold the D84 clip; magenta traces let go. No seed reached the nominal 100 ns — the longest is tegaserod seed 1 at 90.5 ns, and it escapes. Tegaserod escapes in all three harvested seeds; alniditan holds in all three. CLIP DISTANCE TO D84 · ONE STRIP PER SEED · AGAINST THE TIME IT ACTUALLY RAN Cyan holds the clip; magenta lets go. Band = the GO window, 2.6–3.2 Å. Vertical scale is logarithmic. The stage was specified at 100 ns × 3 seeds — the dotted line is where 100 ns would be. ZINC000004286767 2 hold · 1 escape · n=3 4 8 16 seed1 · 88.9 ns HOLD 4 8 16 seed2 · 13.6 ns ESCAPE 4 8 16 seed3 · 9.7 ns HOLD ZINC000006819928 1 hold · 0 escape · n=1 4 8 16 seed1 · 87.3 ns HOLD Alniditan 3 hold · 0 escape · n=3 4 8 16 seed1 · 87.8 ns HOLD 4 8 16 seed2 · 10.1 ns HOLD 4 8 16 seed3 · 9.9 ns HOLD Tegaserod 0 hold · 3 escape · n=3 4 8 16 seed1 · 90.5 ns ESCAPE 4 8 16 seed3 · 10.2 ns ESCAPE 4 8 16 seed2 · 10.0 ns ESCAPE 100 ns — never reached 0 simulated time (ns)
I

Nothing reaches the dashed line. Tegaserod’s long run — the longest anywhere in this project at 90.5 ns — is the pink one climbing away from the grip and never coming back. Alniditan’s three are all blue. Drawn from the raw frame-by-frame data, not from the scoreboard.

Every seed, every number

Read straight from the per-seed reduction files. The rule is the project’s own: HOLD needs second-half occupancy ≥ 0.70 and a second-half mean ≤ 3.2 Å; ESCAPE is occupancy < 0.30.

CompoundSeedActual ns2nd-half mean (Å)OccupancyVerdict
ZINC000004286767seed188.92.8870.967HOLD
ZINC000004286767seed213.63.8010.214ESCAPE
ZINC000004286767seed39.73.0370.918HOLD
ZINC000006819928seed187.32.7241.000HOLD
ALNIDITANseed187.82.7250.998HOLD
ALNIDITANseed210.12.7541.000HOLD
ALNIDITANseed39.92.7241.000HOLD
TEGASERODseed190.510.5560.000ESCAPE
TEGASERODseed310.28.7170.000ESCAPE
TEGASERODseed210.04.5560.036ESCAPE
ZINC000002430405seed189.62.7641.000HOLD
ZINC000001702464seed189.19.6580.193ESCAPE
ZINC000002430405seed212.42.7920.997HOLD
ZINC000001702464seed212.34.6620.199ESCAPE
ZINC000002430405seed39.92.8091.000HOLD
ZINC000001702464seed39.62.7331.000HOLD

Sixteen runs across six compounds. Six are genuine long runs (87–90 ns); ten are stubs of about ten nanoseconds that were labelled “100ns” anyway.

RESULTS_ONLY_20260711/02_verdicts_reductions/*_seed*_100ns_clip.json

Why the scoreboard can’t be checked

Nothing in the project writes it. Searching every Python and shell script for its filename returns zero hits. The project’s own README says the scoreboards can be regenerated. They cannot.

It states a rule that no code implements. The scoreboard says a compound’s verdict is the majority of its seeds. No code anywhere implements a majority rule. The one classifier that does exist does the opposite: HOLD only if all seeds hold; ESCAPE if any seed escapes. Under the project’s own rule, tegaserod and alniditan are both ESCAPE — and that unimplemented majority rule is precisely what turns them into survivors.

Every seed 1 matches perfectly. Nothing else matches at all. All seven compounds’ seed-1 numbers match their long trajectory to three decimals. Every seed 2, 3 and 4 matches no file — not the long runs, not the short runs, not any truncation of either, and not the archive on the external drive.

One of those seeds is about nine nanoseconds. Alniditan’s seed 4 survives on the drive at roughly 8.8 ns — 44 thinned frames, 139 MB against ~1.3 GB for comparable runs. It is counted as a ≥100 ns success.

One measurement would settle it. Re-run the reduction against tegaserod’s seed-2 long trajectory on the external drive. If it returns 2.777 Å at occupancy 1.0, the scoreboard measured something real and this extract is just incomplete. If it returns 4.556 Å at 0.036, the scoreboard is wrong. Until someone runs it, both stay live and this page says so.

None of these are drugs. Two are catalogue numbers with no safety data of any kind. The two that are real medicines were approved to do something else entirely, and nothing here suggests anyone should take them. This is a list of things that would be worth an afternoon of somebody’s bench time. Nothing more, and nothing less.

What died, and why that’s the good part →

onwards.