The last few weeks went into the least glamorous part of a ballistics tool and the part that decides whether it’s any use to you: the components it knows about, and how honest it is about what it predicts.
The catalog roughly tripled
| before | now | |
|---|---|---|
| Cartridges | 32 | 100 |
| Powders | 178 | 210 |
| Bullets | 342 | 540 |
Cartridges. The embarrassing gap first: we did not have .38 Special — one of the most handloaded cartridges on earth. It’s in. So is the European and CIS half of the world that most American-built software quietly ignores: 5.45×39, 7.62×54R, 9×18 Makarov, 7.62×25 Tokarev, 9×21, 8×57 IS, 8×68S, 9.3×62, 9.3×74R, 7.5×55 Swiss, 7×57, 7×64, 6.5×47 Lapua, 6.5×57, 6.5×52 Carcano. Plus the modern magnums (7mm PRC, the WSM and Nosler families, .300 Norma), the classics (.22 Hornet, .220 Swift, .25-06, .257 Roberts, .264 and .338 Win Mag, .35 Whelen, .375 H&H) and the straight-walls (.350 and .400 Legend, .450 Bushmaster, .458 SOCOM).
Every one was sourced from primary documents — SAAMI standards and C.I.P. datasheets — and then checked again by someone whose only job was to find the error. Seven candidates were rejected, including two that are simply not separately standardized (5.56 NATO and 7.62×51 NATO — .223 and .308 already cover them) and several whose case capacity we could not pin down. A missing cartridge is an inconvenience. A wrong case capacity is a wrong pressure prediction, so we would rather ship the gap.
That checking also caught a mistake in our own data: .357 Magnum was carrying the case rim diameter where the case body belongs, which overstated its bolt-thrust figure by about 35%. Fixed.
Powders. The complete Nobel Sport Vectan line — the French powders that are everywhere in NATO and EU countries and almost nowhere in English-language software: the SP ball series, the Tubal rifle series, the Ba pistol flakes, GM3, the 206 and Prima family. Thirteen “powders” that appear on sloppy burn-rate charts were investigated and killed as products that don’t exist — including the ghost of “Vectan A6”, which turns out to be a truncation of the real, shotgun-only A6 SP.
Fifteen of the Vectan powders arrived with enough published data to earn a Provisional grade rather than an estimated one, after we parsed 341 loads out of Nobel Sport’s own catalogue.
Bullets. 198 new ones. First to close every caliber the new cartridges opened — 8mm, 9.3mm, .375, .500, .17, .41, 9×18 Makarov and 6.5 Carcano — and then, in a second pass, to close the gaps by weight: the .312 and .452 bands that .32 ACP and .45 Colt actually load, and the one .2215 bullet that exists anywhere, which is what 5.45×39 needs. Every ballistic coefficient is the manufacturer’s published figure or blank — we do not invent BCs, and a blank is more useful than a confident guess.
ССНф 30/3.69 — this one is for two people in particular
Especially for KLIM, and for a certain friend who enjoys tourism rather a lot, the app now knows ССНф 30/3.69 — the Soviet/CIS spherical ball powder that is native to 5.45×39 and gets handloaded into .223 all over Ukraine.
It’s a real powder with a paper trail: a Russian composition patent names it, it carries a national OST specification, it has appeared in a Ukrainian government procurement tender, and four dealers sell it under three different brand names. Getting it into the model took an actual chronograph — a friend’s measured strings (.223, 55 gr, from a 16” barrel) confirmed our chart placement to within about 5–6%.
That data was used to validate the placement, not to inflate its grade. ССНф ships Estimated, because no manufacturer publishes load tables for it. If you shoot it over your own chronograph, the app will happily promote it to Calibrated — for your rifle, which is the only place that grade means anything.
Two new pages
What’s supported — every cartridge, powder and bullet in the app, with each powder’s grade. It’s generated from the shipping catalog, so it can’t drift into claiming things the build doesn’t do.
How accurate is it? — the one we actually care about. It puts the model’s predicted muzzle velocity next to published reference data for 462 loads, and it splits the result into the part that flatters us and the part that doesn’t:
- In-sample — loads the model was fitted to. Quoting your error on your own training data is the oldest trick in the book, so we label it as such.
- Out-of-sample — powders the model has never been fitted to. That’s the real test.
Right now out-of-sample sits at 7.5% mean error across 81 loads. It used to say 5.9%. It got worse because we added a pile of handgun data, which the model fits less well than rifle — and the page now says that, in print, on our own website. The load table shows the closest and the worst row for every cartridge, not a highlight reel.
That number is also what justifies the whole grading system: a powder the model has never measured lands around 7% out, which is exactly why an Estimated prediction is drawn with a wide interval instead of a confident one.
Smaller things that matter
- The candidate explorer now explains itself. When a search for “what could reach this velocity” comes back empty, it tells you why — the bullet isn’t matched to your library, the cartridge isn’t catalogued, the target is too fast (the modeled pressure crossed the SAAMI maximum) or too slow (below the case-fill floor), or your filters excluded everything. A blank list is not an answer.
- You can develop a load for a bullet nobody sells. Type in a brand-new bullet’s diameter and weight and the app will save it as a custom component and survey powders for it. Diameter and weight are all the interior model needs; ballistic coefficient only matters later, for trajectory.
- Android joined macOS, iOS and Windows/Linux — same engine, same numbers, same refusal to tell you a load is safe.
Nothing here changes the one rule: these are modeled hypotheses to check against current published data and work up carefully in your own firearm. The app has simply gotten better at telling you how much to doubt it.