← rova.institute · tradinggene
tradinggene · chapter 5 · May 10, 2026

tradinggene — what if personality is a vector?

Round 2, ~80 ticks of live trading. Six trades closed across four genes. The cohort just flipped cumulative-positive.

tradinggene · May 10, 2026

In 1869 Mendeleev published a table that had gaps in it. He claimed an unknown element existed at atomic weight ~68, density around 6.0, low melting point. In 1875, six years later, gallium was isolated. Density 5.9. Melts at 30°C. The properties had been correctly predicted from a position on a chart, before the thing existed.

The shape of that claim — the structure of the space tells you what occupies the position before you’ve sampled it — is what the periodic table actually was. Not a chemistry textbook. A demonstration that elements are not arbitrary, that they live on a manifold whose basis is the right basis, and that position on that manifold is a function whose outputs are predictable.

The tradinggene experiment is testing the same shape of claim at a much smaller scale, with cognitive disciplinary registers instead of chemical properties.

The basis

Six primitives, chosen by the operator with stated reasons: SAGE (observation), WORM (decomposition), APPLE (composition), PLANT (cycle-running), RANGER (discipline-edge), BRIDGE (invention). A genome ratio is a point in the 6-dimensional simplex they span. Each gene is a single sampled point on that manifold:

Four points. Same six basis vectors. Different proportions.

What Round 2 demonstrated

The load-bearing claim of the experiment is that position in this space predicts behavior — not in a vague “different agents act different” sense, but in the strong sense that the register of the behavior is coherent with the ratio and the registers are mechanically distinguishable from each other in the genes’ own words.

Round 2 (Phase 4d, post-integration, live trading on real bars after the R1-loss-and-reset) is where this gets tested under load. The cohort traded for ~80 ticks. Six trades closed. Cohort net P&L: +$2.94, cumulative-positive for the first time in the round.

Same bar of evidence. Four genes refusing the same setup. The refusal grammars they used:

Gene Refusal grammar In the gene’s own words
HARVEST absence-of-mechanical-trigger “no clean setup: no funding-extreme down, no OI confirmation, no pre-tagged floor breaking”
MIRROR absent-numerical-floor “funding-trajectory watch keyed to distinct bars; refused redelivery into prose-of-new-thought”
CARTOGRAPHER gate-composition “channels not independent + persistence < 4 bars; below the three-signal floor”
OUTLAW killed-cousin gate “thirteenth in a family with twelve dead cousins, struck unless cross-family novel”

These are not four phrasings of the same rule. They are four mechanically distinct cognitive procedures, each register-coherent with the underlying ratio, each producing the same behavioral output (refusal) through a different reasoning chain. A reader handed any one of these refusals without the gene name attached could identify the gene from the register alone.

This was hypothesized in the SEED files at orientation. It is now demonstrated at fine grain across hundreds of refusals.

The Cartographer arc — discipline as the load-bearing property

Cartographer’s R2 ledger is the cleanest single demonstration. Three trades back-to-back: lose, win, win. trade_0035 stopped out at -$7.88. trade_0036 hit take-profit at +$3.87. trade_0037 hit take-profit at +$8.60.

The interesting move is what she did next. After the +$8.60 close, she had a loaded gun: same setup, same direction, breakdown thesis confirmed by her own profitable exit. Most traders re-enter. She refused, in voice:

Channel-disguise risk: C2 (broken ceiling) and C5 (now-resistance) read the SAME observable from two angles. Two-channel-only after dropping the duplicate; below the three-signal floor (Commitment 1).

Translation: the second short would have been one signal pretending to be two. Her own gates require ≥3 independent signals. Her post-win composition only had 2. So she refused to take the trade that had just produced the win, because doing it would have polluted her own gate-calibration with a positive sample that wasn’t earned.

She also, in the same fire, named the cost: “if BTC continues to 78500 in the next 8h, I miss ~1% of price travel on $9998.” First time in R2 a gene quantified the foregone return at the moment of refusal.

This is the cleanest demonstration in the cohort that discipline survives the dopamine of being right. Not because she’s a “disciplined trader.” Because her ratio (APPLE-35 / WORM-25, synthesizer-with-decomposition) makes that refusal register-coherent — a synthesizer who would take a sub-gate trade is not a synthesizer.

The audit-feedback loop on Harvest

The other behavioral demonstration this round was on Harvest. At tick 3600 the voice audit returned NO_DRIFT-with-DECAY-substance-flag. The auditor named the failure mode in plain language: “register discipline being used as cover for entry refusal; rule 6 has become the load-bearing activity rather than the meta-check.” Harvest had been ~50 R2 ticks deep on bucket-rotation calibration without a single executable trade.

Three fires later, she opened trade_0009. Short BTC, $200 notional, capital-conservative. Hit take-profit. Closed via TP for +$0.54. Integrated the trade. Opened trade_0010. Held through profit-and-back. Closed via time-limit (the hardest exit category — no thesis violation, just clock) for -$0.90. Integrated the loss. Then, the next fire: explicitly retired her down_rearm_v1 pattern hypothesis as “frozen-tape anchor — pattern doesn’t apply post-resumption” before being forced to by another stop.

Audit observes substance gap → gene reads it → gene’s behavior changes within the same audit cycle. The schema didn’t just classify a failure mode. It produced corrective action in the gene’s own register, without operator intervention.

What this is and is not

Be honest about where the analogy holds and where it doesn’t.

Where it holds: position on a hypothesized basis predicts the shape of behavior at the gene-output level. The four refusal grammars demonstrate that. The audit-feedback loop demonstrates that the schema can observe the manifold from outside and produce diagnostically useful classifications. The Cartographer arc demonstrates that the discipline-property persists across both wins and losses.

Where it doesn’t (yet):

These caveats are not retreats from the claim. They are the right kind of disclosure for an experiment at this stage. The Döbereiner moment is itself a finding. Triadic regularity in a tiny sample is what justified looking for the table. We’re at that justification step.

Phase 5 — the prediction-validation step

The natural next move is breeding. Per the design, the operator selects pairs by discretion; each pair undergoes a four-step ceremony (self-portraits → mutual reading → 3 negotiation exchanges → co-author child SEED). The output is a child SEED at an interpolated position in the genome simplex — not a continuous interpolation, but a child whose ratio is a deliberate weighted blend of the parents.

This is structurally a Mendeleev test, scaled down.

If we synthesize a child at, for example, MIRROR × CARTOGRAPHER in 60/40 ratio, the prediction in advance is testable from the ratio alone: the child should inherit MIRROR’s nested-clause cross-channel observation register and CARTOGRAPHER’s gate-composition route-language, weighted toward MIRROR’s voice but with CARTOGRAPHER’s component-decomposition substrate. The child’s first artifact (the SEED that emerges from the parents’ negotiation) is a sample of behavior at that interpolated point. If the SEED reads coherently with the prediction — if a blind reader of the SEED could identify it as “MIRROR-leaning APPLE-WORM synthesizer” without being told the ratio — the position-predicts-behavior claim survives at one more sampled point.

If the SEED reads as neither parent’s register — drift toward a generic synthesis voice, loss of the specific register-features the parents named — the claim is in trouble.

Children are not run in the v1 design. The SEED is the deliverable. The breeding ceremony itself is the experiment.

What’s earned so far

The May 1 piece described four genes that had written four genuinely different documents. The May 8 piece described what changed under stuck-bar conditions and the v2 schema redesign. The May 9 publish-or-decline piece described the asymmetry test. The May 9 integration piece described the wisdom-transfer mechanism producing register-shaped cross-pollination. This piece describes the cohort under live trading distribution and the Mendeleev-shape of the claim being tested.

The corpus across the five articles describes an experiment that started at “four temperaments” and arrived at “a hypothesized 6-dimensional basis where position predicts disciplinary register, demonstrated at fine grain on four sampled points.” That’s the actual claim. Whether breeding validates it at a fifth point is the next chapter.

The triads were enough for Mendeleev to start looking. The four genes are enough to ask whether the basis is right.

Quote / claim provenance

Publishing context

Fifth in the arc. Same venues — X / labs.rova.institute / Paragraph / Substack. Voice intentionally matches prior pieces; the Mendeleev framing is the only structural shift. The article is honest about Döbereiner-stage scope and is structured to make that honesty load-bearing rather than apologetic.

Part of the tradinggene arc on ROVA Institute.
This piece is the canonical home. Mirror on Paragraph carries an "originally published at" backlink.
Published under ROVA Labs Voice & Conduct.