Assert cliff precision on Nauvis: exact counts clear the shared code, localising #18 - #21
Merged
Merged
Conversation
… localising #18 Answers the question the Vulcanus capture raised: is the cliff port broadly bad, with Nauvis's ~90% recall flattering it, or is Vulcanus specifically broken? | | game | ours | recall | precision | over-placement | | --- | --- | --- | --- | --- | --- | | Nauvis seed 123456 | 282 | 282 | 0.943 | 0.943 | 1.000 | | Nauvis seed 777771 | 52 | 52 | 0.942 | 0.942 | 1.000 | | Vulcanus (3 regions) | - | - | 0.57-0.69 | 0.38-0.57 | 1.14-1.58x | **Nauvis places exactly the game's count at both seeds.** So `crossesCliff`, the 4-tile lattice, the orientation table and the cell enumeration - all of it shared between the planets - demonstrably produce the right number of cliffs. #18 is localised to what Vulcanus feeds them: `makeVulcanusCliffFields` or the band constants (ELEVATION_0 = 70, INTERVAL = 120). The shape of the error agrees. Over-placing while also missing a third of the real cliffs means cells land in the wrong places, not merely too freely; a uniformly over-permissive gate would keep recall high and only depress precision. A wrong interval or offset shifts which elevation crossings exist, which adds and removes cells at once. Adds precision and an over-placement ratio to the Nauvis spec so this class of failure cannot hide behind a recall figure on either planet again. The ratio bound is tight (< 1.1) on purpose: cliff placement is deterministic and the measurement is exactly 1.000. Also corrects the spec's own stale number - it claimed "~89-90%" from the original spike and measures 94.3%. The port improved and nobody re-measured. Verified no duplicate positions in either planet's dump, so set and array counting agree. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018MiZeyBwzc9L1Hu6xiKvVe
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Follow-up to #20. Answers whether the cliff port is broadly bad or Vulcanus is specifically broken.
Nauvis places exactly the game's cliff count at both seeds.
crossesCliff, the 4-tile lattice, the orientation table and the cell enumeration - everything the two planets share - demonstrably produce the right number of cliffs. #18 is therefore localised to what Vulcanus feeds them:makeVulcanusCliffFields, or the band constantsELEVATION_0 = 70/INTERVAL = 120.The shape of the error agrees: over-placing while also missing a third of the real cliffs means cells land in the wrong places, not merely too freely. A uniformly over-permissive gate would keep recall high and only depress precision. A wrong interval or offset shifts which elevation crossings exist, adding and removing cells at once.
Changes
No behaviour change -
src/is untouched.pnpm run verify: 1140 pass, 0 fail.🤖 Generated with Claude Code
https://claude.ai/code/session_018MiZeyBwzc9L1Hu6xiKvVe