Compare devlog: v1.1.0 is out
The first public release, a prior that finally knows what kind of body it is looking at, and three scoring terms that had never fired once.
v1.1.0 shipped today - the first public release. A portable executable and a program-folder zip, published as two releases, and the project page now has a Download button that picks the right one.
I use it every session. 1,778 tests, three type-check projects clean. The predictor is the part that keeps moving; everything around it has settled down.
What is in the box
It predicts which species are on a body from the FSS and DSS telemetry alone - a Bayesian ranking over a corpus that has grown to 51,000 observations - with payout ranges and the first-footfall value. It reads the journal live, so scans, organic signals and on-foot analyses merge into one view. There are HUD overlays over the game: a route strip that marks systems nobody has reached, a radar with the plant marks fixed in place, all of it customisable. A system map, a galaxy map, and an encyclopedia with 157 contributed photographs, each labelled with who took it. Statistics - income by source, balance history, carrier upkeep and weeks of runway - measured from your own journals rather than estimated. And optional EDSM lookups and Canonn upload, both off until you switch them on.
The predictor work in this release
The prior now conditions on what kind of body it is looking at - planet class, atmosphere, volcanism family, temperature band - instead of only asking how common a species is across the whole galaxy. I set its weight at 0.4. That is not where the probe peaks; it is the largest value that still agrees with both of the field landings I use as sanity checks, and I would rather agree with the ground than with my own curve.
Bacterium tela finally has a rule I can state in one line: volcanism, or a surface temperature of 300 K or more. 687 bodies out of 687, no exceptions. It took an embarrassingly long time to see something that simple.
And three categorical scoring terms turned out never to have matched anything, ever. Months of a silent penalty applied to every candidate equally, which is exactly why nothing looked wrong. Fixed, and the weights re-swept afterwards.
Found while releasing
The catalogue the app learns from your own on-foot scans was living inside the install folder. It shipped inside builds, and every rebuild wiped it. It lives beside your settings now, where it survives. My test isolation had been leaking in the same direction, and got the same fix.
Next
A prediction archive you can share, and more landings on the neon-plus-volcanism bodies where the model and the ground still disagree.