Case study · built from the source atlas · 2026-08-05

Moon Arena — a habitability board that measures its own ignorance

Eighteen lunar pits scored on four axes, with a second scoring mode whose only purpose is to show how much of the ranking rests on measurements nobody has taken. The interesting output is not the leaderboard. It is the distance between the two leaderboards.

The problem

Every habitability ranking has the same silent defect: a candidate that nobody has studied scores badly for the same reason a candidate that has been studied and found wanting scores badly. The two are indistinguishable in the output, so the ranking quietly rewards whichever objects happened to get telescope time. Moon Arena was built to make that distinction visible rather than to argue it away.

Pipeline

01 AtlasLROC Lunar Pit Atlas, 278 entries listed
02 Curate18 pits captured with per-field provenance
03 Score twiceMeasured-only and unmeasured-charged
04 CompareThe gap between the two is the finding

Live interface

Real working board. Data is inlined into the page, so it runs with no network dependency — same rule the other cases here follow. Switch Reading in the controls and watch the order change.

app/index.html · Moon Arena Open full screen

Score weights (from app.js)

AxisWeightMeasured forSource field
shelter0.3510 of 18overhang_or_cave
access0.3016 of 18depth_m
mouth0.2017 of 18inner_diameter_m.min
thermal0.152 of 18thermal_note

Shelter carries the most weight because an overhang is the entire reason a pit is interesting: metres of regolith overhead is the radiation shield a surface habitat would otherwise have to carry. Thermal carries the least — not because temperature matters least, but because it is measured for two pits, and a heavily-weighted axis that almost nobody has data on would dominate the board through absence rather than evidence.

What it found

The ranking is not robust, and the instability has a direction

Switching from measured axes only to unmeasured charged moves 16 of 18 pits. The two that hold their position are the two at the bottom, and they hold because there is nowhere below them to go. Both fully-measured pits rise. All six two-axis pits fall, by four to five places each.

Under the permissive reading, King 1a — an 11-metre impact-melt pit with a confirmed overhang and no thermal data — leads the board. Under the strict reading, first place goes to Mare Ingenii Pit, and Mare Tranquillitatis Pit climbs from seventh to fourth. Those two are the only pits in the set with published Diviner thermal measurements, and they are also the two the literature treats as the type examples. The board was not tuned to produce that agreement; it falls out of refusing to reward missing data.

What that does not establish

Neither reading is a habitability model. The axes are geometric and thermal proxies drawn from an atlas, not a peer-reviewed suitability function, and the weights are a stated judgement rather than a derived result. A pit with a wide gap is an unstudied pit, not a bad one — Compton Pit carries the widest gap on the board (+34) and might well be excellent. The board can say precisely how much it does not know about Compton. It cannot say whether Compton is good.

Gaps in the source record

Collected during retrieval and carried in the dataset rather than discarded. A board that hides these is claiming a completeness the atlas does not have.

Fields the atlas does not publish

Retrieval problems

Volumes (labelled separately)

278pits listed by the LROC atlas on the retrieval datepagination text on lroc.im-ldi.com/atlases/pits/list
18captured in this dataset — a seeded sample, not the cataloguemoon_pits.json → entries
16confirmed pits; 2 listed as candidatessame file, confirmed flag
2pits measured on all four axescomputed in app.js, shown as coverage 4/4

Sibling case

Mars Arena is the same idea one planet over, and the contrast is the point. Mars Arena scores 1,062 catalogue records by deriving component inputs from priority and type — which lets every row carry a full score, and hides the fact that most rows were never measured on anything. That case documents the shortcut on its own page. Moon Arena is what happens when you refuse it: eighteen rows instead of a thousand, and a column that tells you which of the eighteen you are allowed to trust.