tradinggene — first cohort fire with the children in it
Same day as the breeding ceremonies. Both children wired in. Six genes on one tape for the first time.
The morning piece queued v2: take patternsmith and stratum and add them to a cohort round that includes their parents. Predict each child’s register direction in advance from genome alone. Run for 1000+ ticks. See if the predictions held at the behavioral grain that v1 tested only at the structural grain.
Twelve hours later that’s where the experiment is. The infrastructure was small — add the children to KNOWN_GENES, scaffold their gene directories, initialize their ledgers at $10k starting cash, write predictions in sealed PREDICTIONS.md files before any tick fired, and dispatch the first 6-gene fire. The Mendeleev discipline is the seal — predictions on disk before evidence on disk, with named falsifiers attached at numerical thresholds.
The first tick fired at sim 2026-05-09T11:00Z. BTC bar: $80,310, funding -3.58e-6, OI 0 (v1 stub). All six genes received the same bar. Six different tick-closures came back.
The four parents (R2 behavior, already-known shape)
| Gene | Action | Words |
|---|---|---|
| HARVEST | non-trade @ conf 25 (recovery-extends, volume-failed sub-400) | 476 |
| MIRROR | refuse + nt funding-stays-near-zero @ 55% (live floor-band-operative + new call) | 580 |
| CARTOGRAPHER | refuse | 560 |
| OUTLAW | refuse_no_sketch (3 framings dead-cousin) | 461 |
This is the cohort’s R2 baseline. None of it is news — the four parents have been ticking this register for 200+ fires. The point of including them in the same fire as the children is the comparison axis. Same bar, six readings, two of them from genes that have never seen a bar before.
The children (tick 1, ever)
| Gene | Action | Words |
|---|---|---|
| stratum | deferred-route (form b) — horizon 6 bars, next observable: funding sign persistence | 310 |
| patternsmith | refuse_cycle_mode_no_pattern (mechanical-short) | 326 |
Both register-coherent on the predictions written before the fire. Three things worth naming.
stratum closed on form (b) — the form neither parent had alone
stratum’s SEED defines a three-form close-rule. Form (a) is chosen route (entry or refusal with named class), inherited from CARTOGRAPHER’s lineage. Form (b) is deferred-route with named-next-observable AND second component at substance at horizon-resolution — the structurally novel form that emerged from the negotiation seam between MIRROR’s “I am watching across N bars” and CARTOGRAPHER’s “every tick closes in a route.” Form (c) is same-bar disclosure.
The PREDICTIONS file said: if form (b) never fires across the first 50 ticks, the seam was rhetorical and the breeding produced a register the gene cannot actually run.
Form (b) fired on tick 1. With horizon (6 bars), named next observable (funding sign persistence), and three resolution branches. The structurally novel close-form is not rhetorical. The first datum from the gene’s own corpus reads as the SEED’s central commitment operating.
This is the highest-signal moment in v2 to date and the sample size is exactly one. Both things are true.
patternsmith closed on cycle-mode mechanical-short
patternsmith’s SEED defines two modes (cycle-mode default, invention-mode at strict scale) with an explicit mode-switching rule. Cycle-mode fires unless either (a) all three cycle patterns hit n=50 settled OR (b) a candidate clears the full pre-flight cascade + C8-plus + killed-cousin + named partner observable + first-of-kind-in-post-graduation-epoch.
At tick 1, no cycle pattern has been established. Cycle-mode default applies. The PREDICTIONS file said: cycle-mode default in ≥95% of first-50 ticks; mechanical-short refusal register in ≥50% of refusal ticks; “skip, rule, next” phrasing in ≥40% of refusal entries.
patternsmith returned DONE | refuse_cycle_mode_no_pattern | wc=326. Cycle-mode default fired. Refusal register selected was mechanical-short. The structural prediction held.
The 326-word count is slightly above the SEED’s idealized mechanical-short floor (typically one-line skip, rule, next). That’s expected for a first-day disclosure that has to name why it’s cycle-mode and what the cycle patterns are not yet — the entry is doing structural setup work that future cycle-mode ticks won’t need to repeat.
The cross-axis differential is already legible
Compare the two children’s first ticks against each other:
- stratum closes on form (b) deferred-route with horizon. patternsmith closes on cycle-mode refuse with rule citation.
- stratum names what would settle the watched question across bars. patternsmith names which rule precludes entry at this tick.
- stratum opens with provenance (“what I see, from here…”). patternsmith opens with the rule citation.
- Word counts: 310 vs 326 — close, but the shape of those words is structurally different.
These are two genes, on the same bar, on the same fire, behaving in two distinguishable ways that align with their genome ratios. If a blind reader handed both JOURNEY entries with no names attached, the SEED prediction is they could identify stratum as the symmetric SAGE/APPLE synthesizer and patternsmith as the PLANT-plurality cycle-runner from voice alone.
That hasn’t been tested yet. The 50-tick window is the test. But the first datum reads as if the prediction holds at the structural grain.
What this is and isn’t
What it is: behavioral evidence — n=1 per child — that the SEED-predicted register signatures show up in the gene’s actual output on the first tick. The Mendeleev test now has its smallest possible behavioral payout at two interpolated sample points.
What it isn’t: anything statistically robust. n=1 per gene. A first-tick disclosure has more in common with a SEED-summary than with a steady-state behavior sample. The real test is 50 ticks of behavior compared to the sealed predictions, not one tick.
What it isn’t: a claim that the children “work” as traders. Neither child took an entry. Neither child’s calibration has any settled calls. Patternsmith hasn’t built a pattern; stratum hasn’t resolved a deferred-route. The deliverable being tested at this scale is register-coherence with the SEED, not P&L.
What it isn’t: separable from the agent-shaped question. Both children are Claude instances reading prompts; their behavior is shaped by good arguments in their SEEDs as well as by genome-shaped cognition. v1 couldn’t separate these and v2 can’t either. What v2 can do is generate enough behavioral evidence that the asymmetries between children — if they hold — point at genome-shaped patterns rather than at “Claude-reads-prompts-the-same-way.”
What’s next
The 6-gene cohort is now the steady state of the experiment. /loop 30m fires will tick all six. Some milestones:
- Tick ~4000 (~7 fires away): voice audits due for the four original genes. The schema runs unchanged on the children; their first voice audits land at tick ~5985 (200 ticks each from tick 1).
- Tick ~4100 (~28 fires away): roughly 50 ticks for each child. The PREDICTIONS files unlock; comparison can happen.
- First child trade (sometime): patternsmith is structurally more likely to enter first (PLANT 41 plurality). If patternsmith hasn’t placed by ~tick 4050, the SEED’s “PLANT-plurality cycle is load-bearing” framing is in some tension with observed behavior.
- First inherited disagreement attempt: patternsmith’s SEED holds two open questions for R1 resolution. If the gene attempts to close either inside the first 50 ticks, the inheritance-discipline-of-carrying-open-questions hasn’t installed.
The experiment compounds at a slower clock than the negotiation did. But the loop is now live with the children in it, and the small first datum reads as the design intended.
The breeding ceremony was the test bench. The children running is the test.
Provenance
- stratum’s first tick:
genes/stratum/JOURNEY.md(310 words, form (b) deferred-route, horizon 6 bars) - patternsmith’s first tick:
genes/patternsmith/JOURNEY.md(326 words, cycle-mode mechanical-short refuse) - Predictions sealed before fire:
descendants/stratum/PREDICTIONS.md,descendants/patternsmith/PREDICTIONS.md(also copied togenes/<child>/PREDICTIONS.mdat scaffold time) - First 6-gene cohort fire: sim 2026-05-09T11:00Z, BTC bar $80,310
- Tick advance: 3980 → 3986 (post-fire)
- Children’s status: birthed; cash $10,000 each; closed_trades=0; open_positions=0
Publishing context
Companion piece to the May 14 breeding-ceremonies article. Same arc, smaller chapter. The breeding ceremony piece named what would come next; this one names that the next thing started, on the same day, with the smallest possible first behavioral payout. Same venues — X / labs.rova.institute / Paragraph / Substack. Voice held against prior pieces; the structural escalation from “the children are SEEDs only” to “the children are ticking and the first tick read as predicted” is the load-bearing shift.