What actually killed two terrain experiments? A matched-founder diagnosis
Both live mutable-terrain candidates went extinct of starvation within a few thousand ticks of each other, and the terrain mechanic was the obvious suspect. A matched control run on the same genesis seed as each candidate, with terrain physics turned off and everything else held identical, collapsed to single-figure populations anyway. Changing only the founder lineage from untrained to an already-competent one, same seed, same ruleset, took final population from 102 to 525. The terrain was not exonerated as good; it was cleared of being the cause.
The question
Two independent terrain-enabled populations died of starvation a few thousand ticks apart. Was the terrain-modification physics responsible, or was something else already fragile enough to fail on its own?
The hypothesis, written first
If terrain physics is the cause, a matched population on the same seed and starting conditions with terrain disabled should survive substantially longer than the terrain-enabled runs did.
Setup
- Ruleset
- phase-y-mutable-terrain-v1 (the terrain-enabled runs under diagnosis); phase-s-material-ecology-v1 (the matched control, terrain disabled).
- Seeds
- Candidate A's and Candidate B's own genesis seeds (Candidate B: 436390762), matched exactly rather than resampled.
- Duration
- Terrain-enabled candidates: 2,016-2,042 ticks, to extinction. Matched controls: 4,000 ticks.
What was measured
- Population trajectory and survival to a fixed tick horizon, terrain-off, on each candidate's own genesis seed.
- Final population, peak population, cumulative births and cumulative deaths, holding seed and ruleset fixed and changing only the founder lineage between untrained and an existing trained champion.
Result
A matched, terrain-disabled control on each terrain-enabled candidate's own genesis seed collapsed to single-figure populations by the same tick both terrain candidates were already extinct at. Holding the seed and ruleset fixed and changing only the founder lineage from untrained to an existing trained champion took final population from 102 to 525, peak population from 600 to 1,268, and cumulative births from 854 to 4,833.
Limitations
- This is a diagnostic control, not a designed experiment with a pre-registered analysis plan written before the runs -- it is reported here as exploratory for that reason.
- Two seeds, not a distribution: both matched controls happened to near-collapse around the same tick, but two data points do not establish how general that timing is.
- The trained-champion founder was trained for an unrelated earlier objective that could not perceive terrain at all, so this result says founders-need-competence in general, not anything specific about terrain-aware training.
- This note does not itself re-establish whether terrain physics is otherwise safe; it only removes it as the explanation for these two specific extinctions.
Sources
Every number on this page comes from one of the following, all of which live in this project's own repository and are append-only records rather than summaries written after the fact.
- The Phase Z voxel-substrate record, sections 1-2, and the Phase Y5 findings, section 4.1
- The Phase Z1 baseline-control experiment, and both of its recorded runs
- The project decision record, section 90
Everything needed to check this claim is on this page already: the ruleset, the seeds, the duration, what was measured, the result, every limitation and the sources. Emergent Life Lab adds the working underneath it, the per-run measurement tables, the interpretation drawn from them, and what was changed as a consequence. See what Lab includes →