The world
A Flow Lenia state is a continuous spatial field with several founder channels. The material moves; the developmental pattern reorganizes it. We call a persistent non-square form an organism here, without implying autonomous reproduction.
A colleague's entrance · five minutes first, receipts underneath
The organism appears before the silhouette. Then it hardens without converging.
We have been using Flow Lenia as a world where a formless seed can become an organic-looking body, then asking a more demanding question than “when does it look alive?”: when does the developing system begin to constrain its own possible futures and the meaning of local interventions?
frozen primary passed
predeclared secondary
open causal atlas
null / unresolved
wrong direction
Evidence snapshot: the synthesis-v4 campaign plus its completed four-run closing programme. The newer impedance, controller-tournament, and causal-passport experiments are deliberately excluded because they were still underway when this page was frozen. No result from those runs appears here.
If you read one paragraph
Flow Lenia is a mass-transport variant of Lenia: continuous fields move material around under local rules, and a simple artificial square can reorganize into a persistent body. We watch the state using past-only information measures, freeze a prediction or local action, and only then open the future. A prospective 64-specimen test found that whole-over-parts predictive organization rises before a sustained visible departure from the square. Later causal experiments found that action meanings stabilize and a standard local push moves older organisms less. The surprise is what did not happen: sibling futures widen with age, and a dedicated fine-shape confirmation was unresolved. The system is not collapsing into one canonical adult; it is becoming a more history-specific individual that is harder to redirect.
01 / Five-minute map
Four things to know before the numbers.
This is the smallest useful map of the world, the measurement, the interventions, and the claim. The technical labels remain in the report because they identify exact analyses; none are prerequisites for reading it.
The question
Does a coordinated whole become causally meaningful before the body is obvious? “Meaningful” means measurable in forecasts and in how matched local pokes alter later futures, not merely visually suggestive.
The instrument
We compare what the whole past predicts with what the strongest independent split predicts. We also launch a fixed book of local actions from the same state and measure the resulting fan of futures.
The answer so far
Organization appears early, then responsiveness falls and identities separate. The evidence supports selective causal commitment. It does not support Φ as a magic steering knob or development as global convergence.
02 / The measurement
Can the whole forecast itself better than its best split?
The same short past is forecast two ways. One model keeps all cross-part relationships. The other cuts the state at its strongest independent bipartition. Their difference is our whole-over-parts coordinate.
The prospective precursor rose by +0.364 before visible form, but the absolute whole−sum level stayed negative: the whole was catching up with its strongest split, not overtaking it. Adding the score to ordinary visible features barely changed held-out event AUC, from 0.8089 to 0.8093.
03 / The discovery arc
The measurements changed before the body did.
Read left to right. Each stage names what was actually measured and its evidence class. The final stage is deliberately a correction: the neat “shape fan closes” ending did not confirm.
A broad action-world already exists.
Future-history spread at q8 was 1.044× its q32 value across 39 organisms and 5,616 futures. Fixed-age atlas; source organisms reused.
Whole-over-parts prediction rises.
Frozen prospective primary, interval +0.231 to +0.502; 54 complete windows, 16 families. Specificity and practical forecast gain remained unresolved.
Action meanings stabilize.
Mean context rewrite fell from 0.02636 before the turn to 0.02343 after it, a 0.00294 decrease. Frozen two-action primary; 19 families and 152 complete panels.
The same push loses leverage.
Mean divergence fell from 0.00881 at age 60 to 0.00274 at age 780. The pre-vs-post-turn family contrast was +0.00385 with interval above zero; same opened cohort, fresh intervention outcomes.
Fine-shape convergence stays unresolved.
The dedicated 12-family confirmation interval was −0.0076 to +0.0038, six families each way. It neither confirmed the original −0.0040 signal nor excluded an effect of that size.
04 / What survived
The strongest results, with their boundaries attached.
No single card carries the story. Together they say that the state gains predictive organization, becomes less rewritable, and becomes more individually specific without collapsing toward a shared adult endpoint.
A pre-form whole-over-parts rise.
+0.364Across a fresh 64-specimen cohort, the event-near whole−sum signal exceeded its own earlier baseline in 39 of 54 complete windows. The whole−sum level itself remained negative, and the held-out forecast secondary was effectively null.
organism-birth-signal-v1/prospective-results.json
Narrower transition timing, prospectively.
ρ −0.248Higher whole−sum organization forecast a smaller range of transition times under the same eight local actions, after controlling for checkpoint, visible shape, size, topology, and raw field change. Interval −0.420 to −0.045; 63 specimens.
causal-repertoire-contraction-v1/results.json
Older organisms are harder to redirect.
+0.00385The family-mean divergence caused by the standard push was larger before each organism's past-only structural turn than after it. Interval +0.00185 to +0.00545; 15 of 16 family contrasts positive. The intervention outcomes were fresh, but the cohort had been opened by sibling experiments.
steerability-age-profile-v1/results.json
Look-alike states become rarer.
6,144 → 1,665At a fixed early tolerance, cross-history pairs with similar present morphology collapsed with age, Spearman ρ −0.988. The experiment's matched-ratio primary was null, so this is a pre-named secondary over the same sibling corpus—not an independent confirmation.
history-disambiguation-curve-v1/results.json
05 / The correction that improved the story
Harder to move does not mean closer together.
We expected mature sibling futures to occupy a narrower basin. The frozen test resolved decisively in the other direction. That did not flatten the programme; it told us what kind of commitment we were actually seeing.
06 / The campaign, not just the winners
95 experiment folders sit behind this snapshot.
The denominator matters. Only ten folders contain a predeclared prediction that resolved in its stated direction; most of the campaign is exploratory work, nulls, reversals, or mechanically stopped runs. The story above survives because those outcomes are part of it.
10243850
10prediction met24null / unresolved3wrong direction8abort / mechanical stop50open / descriptive
07 / Why this matters
A causal route into organismhood.
The work is interesting because it turns “this looks like an organism” into interventions with competing predictions. We can ask when the whole begins to govern what local actions mean, what becomes harder to change, and which degrees of freedom remain open.
What we would share
A Flow Lenia organism becomes organized first as a constraint on possible futures and action meanings. With age it becomes harder to redirect and more identifiable by its own history, even while siblings diverge.
The GARD connection
GARD supplied the provocation: look for collective predictive organization before familiar organism-level behavior. We use a documented Flow Lenia whole-over-parts analogue; we are not claiming to reproduce an undisclosed estimator from another system.
What remains open
We have not shown autonomous reproduction, a universal scalar of organismhood, or confirmed fine-shape convergence. The action-world and same-state spatial-ablation tests provide real causal handles; none of the evidence summarized here establishes Φ-directed pushes as a general steering policy.
08 / Tiny glossary
The recurring language, in ordinary words.
Keep these six terms and the rest of the dossier becomes legible.
- Flow Lenia
- A mass-transport variant of Lenia: continuous spatial fields reorganize under local update rules while material is transported rather than created as ordinary additive growth.
- q60 / age 60
- Developmental time in simulation steps or pulses. Early, turn, and late are organism-relative when the experiment uses each organism's own past-only structural turn.
- whole−sum / Φ
- The predictive information in the coordinated whole minus the strongest prediction available after splitting it into independent parts. Here it is a partial coordinate, not a consciousness claim.
- action-world
- The fan of futures produced by applying the same fixed book of local pokes to matched copies of one state.
- future-blind / frozen
- The selection rule, action, analysis, or pass condition was fixed from past data before the relevant future outcomes were opened.
- family
- The usual statistical unit: organisms from a shared founder lineage are averaged within family before uncertainty is calculated, so 64 organisms often means 16 independent family units.
Source-bound evidence ledger · claims used on this page
Each result SHA-256 was recomputed byte-for-byte before this page was written. A matching digest establishes file identity; the status and claim boundary were then checked against the named JSON fields. The final row binds the audited v4 campaign census, which is context rather than a scientific result. No unfinished experiment appears below.
| Result artifact | SHA-256 | What it supports here | Status / boundary |
|---|---|---|---|
organism-birth-signal-v1/prospective-results.json | c6e3c6634d04de86ee89f5d11dec423e936befe81e022240a82dbdbc85daf963 | +0.364 precursor; interval; 54 complete windows; null held-out AUC increment | fresh prospective primary; specificity unresolved |
causal-repertoire-contraction-v1/results.json | 9fe83df68e0e2cef8ba487673e4bf1005cacc5bb1dbbf84c963b90d8d2602fd4 | timing-range partial rank association −0.248 under eight matched actions | fresh future-blind prospective primary; 63 specimens / 222 states |
initial-action-world-v1/results.json | 99727f878bab51518dddfb618cec0cd98c15bb6fe8b3667f515adb6d933e4817 | q8 future-history spread is 1.044× q32; 5,616 futures | fixed-age causal atlas; source organisms reused |
spatial-scale-action-world-v1/results.json | 6b3a76b4eb347010d9cb14a9bf57d1c70c69f4612a1bf7fee1b41fb57537f186 | same-state spatial scrambling changes the action-world | direct mechanistic atlas; same 39 organisms |
developmental-causal-grammar-v1/results.json | 200e1c14daef2d965fff12ccb167ca70b7ecaad410393ab2a7d445401d920e17 | two-action context rewrite falls after the information turn | future-blind primary; 19 families / 152 panels |
fresh-cohort-fine-shape-confirmation-v1/results.json | 5318aa930cd789d446372828dd0fc1bcf55edbeafc7b07d7868feb76d5bc77d4 | −0.0019 sole primary; interval −0.0076 to +0.0038 | fresh confirmation unresolved; 12 families; secondaries cannot rescue |
steerability-age-profile-v1/results.json | 7f88560f446752adb31483f8f9e5f851a67069e97050d40393ab26784f98f949 | pre-turn minus post-turn divergence +0.00385; ~69% age decline | frozen primary passed; fresh intervention outcomes, previously opened cohort |
basin-width-init-variation-v1/results.json | f71fd77f180cd081ea9011970ac2e109ec7b25179b5b7982f467a8f40deed91e | sibling width rises from 0.032 at age 60 to 0.079 at age 900 | frozen primary resolved in wrong direction; fresh initialization outcomes |
history-disambiguation-curve-v1/results.json | 6cb3e4a06fa6244bc2347d11543d0a54484105edb7fbd346c0e9202380bf68f1 | fixed-epsilon look-alike count 6,144 → 1,665; ρ −0.988 | primary null; pre-named secondary passed; analysis over sibling corpus |
synthesis-v4/the-organism-appears-first-in-possibility-space.html | 69107dac3f182d4876e3bafe167f33a711c693c68f6f15e1c9a65fc7873af0fd | 95-folder campaign census: 10 passed, 24 null/unresolved, 3 wrong-way, 8 stopped, 50 open/descriptive | audited synthesis census; contextual denominator, not a result artifact |