What this site used to say, and when it stopped.

A 1.41-million-atom matched simulation of human cardiac Nav1.5 in a POPC bilayer completed its first 0.5 nanoseconds of production sampling on an autonomous workstation. The DEKA selectivity filter remained open at 5.23 angstroms (matching the cryo-EM open state benchmark of 5.46 angstroms) with 12 explicit water molecules continuously hydrating the filter. Residue 104 unhinges into solvent as an unclipped monodentate pivot, confirming that the 31.7 percent current drop in automated patch clamp assays reflects impaired trafficking and intercalated disc assembly rather than physical occlusion of the pore.

A new visual atlas was deployed at /pictures, establishing an open photographic and computational gallery from whole-heart biventricular coordinate meshes (30 mm) down to the atomic selectivity filter (5.2 angstroms). Every panel is grounded in explicit coordinate models and reproducible data.

Every page across the site was synchronized with universal sidebar connectivity and real-time active indicators. An autonomous background watcher on the compute cluster was coupled directly to the production site pipeline, automatically publishing verified molecular dynamics trajectory data to the public record as milestones complete.

Simulations of planar electrical propagation across 2D tissue geometries revealed that abrupt transitions from narrow strands to wide myocardial sheets create severe current sink mismatches. Under reduced sodium current (31.3 percent of normal), propagation fails at sudden geometric expansions where normal tissue conducts. This establishes an existence proof in the model: cellular loss-of-function interacts directly with tissue architecture and fiber coupling.

An unexpected finding in cable simulations where electrical wavefronts propagated despite zero peak sodium conductance was resolved. Current traces proved the wave was sustained by L-type calcium current (ICaL) acting as a slow response. Selective removal of calcium current verified this mechanism, ensuring that future simulations do not mistake calcium slow responses for sodium conduction rescue.

A full audit of cable electrophysiology simulations revealed that an earlier reported threshold for conduction failure was an artifact of stimulus capture rather than true downstream tissue block. When tested with adequate stimulus delivery, the tissue conducted at lower sodium levels than previously claimed. The earlier inference regarding tight RVOT rescue margins was formally withdrawn.

An audit of recovered docking and potential of mean force (PMF) files found that the modeled ligand had three nitrogens rather than the four present in genuine agmatine. In addition, time drift in the sampling prevented equilibrium PMF verification. The candidate was formally gated and removed from being cited as evidence for small molecule rescue.

Reanalysis of 352 human cardiac RNA-seq datasets showed that non-productive splice junction read counts are highly sensitive to low-count filtering and pipeline alignment parameters. Double-edge junction counts dropped from 102 down to 17-18, demonstrating that alternative splicing does not offer a free, unconstrained target for upregulation.

The entire 38-gigabyte project was reorganized into eight numbered home directories with explicit provenance. The primary instructions file was condensed from 114 kilobytes to 6.9 kilobytes while preserving historical audit logs. Every internal reference was verified by automated checkers, resolving 793 broken pointers.

The site navigation was updated to collapse into an off-canvas drawer on mobile and narrow viewports below 1100 pixels, preserving reading room for manuscripts and data tables without introducing client-side frameworks.

The efficiency of the only drug that exists for the interaction route had been derived from a summary of a paper rather than from the paper. Its own table gives a figure two and a half times larger. That would move the conclusion in this project's favour, and it is still not quoted, because the corrected value breaks the model that defines it and the source marks no such comparison as significant. Withdrawing is the honest move where replacing would be the flattering one.

Two magnitudes this project had written down as impossible to recover from published figures were recovered, one from a table that had been there all along and one by measuring the figure properly. A third parameter, which every sample size on the experiments page had been guessed around, was read exactly from a supplementary file because the panel turned out to be drawn as vector rather than as an image.

Every human readable file in the project was read. The defects found were almost all pointers rather than numbers: an index naming files that no longer exist, a tracking record contradicting the thing it tracked, a note warning about a defect that had already been fixed. The habit of correcting a figure and dating the correction works. What it does not catch is a corrected figure sitting above arithmetic that was never recomputed.

The site had been citing version specific identifiers for ten of the papers. Those pin a superseded version. They are now the permanent identifiers, which resolve to whichever version is current and cannot go stale in this way again.

A pass over all 14,035 papers in this project's collection finally asked each one whether it described a route to a treatment. The list gained two entries and a third was rewritten. What makes that worse rather than better is where two of the three came from: they were not hidden in anyone's literature. One had been on the routes page since the site was built while the project's own inventory had stopped listing it. Another had been closed on a measurement that only covered one method of doing it.

The rescaled figures read 34.1 percent against 50, with a gap of about 16 points. Both came from dividing the measured 68.3 percent by two, which assumes the two copies add perfectly. The two copy case was measured directly at 218.4 percent of one, so the divisor is 2.184: the numbers are now 31.3 against 45.8, and the gap is 14.5 points. The retired baseline and the retired comparator travel together, so quoting 31.3 against 50 is a third wrong answer rather than a partial fix.

Six of eleven documents were found to state something untrue rather than merely miscalculate: named tables described as deposited that the archive does not hold, a claim that no expert panel exists for this gene, and a count contradicted two subsections later. The corrected text is served here and the identifier remains the thing to cite.

Publication is an act of record keeping rather than a therapeutic route, which is why it appears in the project inventory and not on the routes page.

An earlier version of this site presented a different hypothesis, that a small molecule could stabilise the protein while it folded, and carried a screen of candidate compounds. The prediction method underneath it failed its own calibration. It is named here rather than deleted, because anyone who read the earlier version deserves to know it was withdrawn and why. Nothing from that campaign should be cited.