← rova.institute · tradinggene
tradinggene · chapter 1 · May 1, 2026

tradinggene — findings to date

Compressed phase, ~70% complete. Live phases not yet run.

tradinggene · May 1, 2026

The cleanest single read on the experiment so far is that the four genes are writing in four genuinely different voices, and the differences are not stylistic. They show up in what each gene refuses, what each gene chooses to do with its time, and what each gene treats as evidence. A side-by-side of their WISDOM files reads like four different authors with four different theories of what trading is for.

The four voices, in their own words

HARVEST (PLANT 45) has settled into the operational register the seed predicted — short sentences, action verbs, mechanical rules. The whole epistemology fits in one line:

Hope is not a position.

The edge is the loop, not any single rep.

When HARVEST blew its account on a unit-translation bug — typed --size 200 intending USD-notional, got 200 BTC — it did not catastrophize. It logged the loss as data and proposed a SEED amendment requiring an explicit arithmetic line before every order. The amendment was accepted. The gene continues in paper-only mode while the larger account remains underwater, treating that constraint as terrain.

MIRROR (SAGE 40) writes in nested clauses with the patience the seed asked for and a self-suspicion that surprises:

I am writing this at tick two, before any trade, before any settled call, before any drawdown — which means every claim below is a prediction about a self I have not yet observed under pressure, and the act of writing it is less about being correct now and more about leaving a prior I can be measured against later.

MIRROR has placed only two trades across the full compressed phase. The seed predicted this. The risk MIRROR named for itself was watching-as-default — patience as a hiding place — and it has flagged its own drift toward that pattern in real time: “If twenty consecutive ticks produce no sharpened question and no recorded non-trade hypothetical, I have stopped working and started spectating.” The mechanism it set against itself is in evidence.

CARTOGRAPHER (APPLE 35) writes in route-language and decomposition, exactly as the seed predicted. The voice is most visible in its refusals:

Refusal is a route, not a fallback.

After 19 closed trades, CARTOGRAPHER rewrote its WISDOM with a “Posterior” section — what 19 trades had actually taught versus what the orientation hypothesis had assumed. It demoted Family-1 to a surrogate because OI data turned out to be unavailable, gated Family-3 behind a vol-regime floor it had missed at entry, and noted with characteristic self-awareness: “The active map should shrink before it grows.”

OUTLAW (BRIDGE 35) has produced the most surprising artifact of the experiment so far: a kill list. Six candidate primitives invented, six killed, zero capital deployed. Each kill is documented with a named lesson:

The kill-list-with-method is the gene’s actual product so far — six dead inventions, zero capital deployed.

The lessons crystallised into five pre-flight checks the gene now runs on every new sketch — distribution, data-availability, joint-density, reachability, at-sketch backtest. OUTLAW’s voice is meta-procedural, paranoid in a useful way:

Falsifier in the same paragraph: tag 30 windows… under 55% kills the primitive before any size scales beyond the 1% cap.

The seed predicted OUTLAW would carry the highest blow-up risk. The compressed phase has so far produced the opposite: OUTLAW is the most disciplined gene in the cohort. Whether that holds when the substrate changes — when live trading replaces historical playback, and the cohort it claims to want to differ from is actually visible in real time — is one of the more interesting open questions.

Cross-cutting findings

Articulated evolution is not theoretical. MIRROR proposed a SEED amendment after its catastrophic loss, with evidence and a precise mechanical rule. The operator accepted it. The change shipped. The gene continues operating with a SEED it helped author. This is the articulated-evolution channel — the third drift class, the one the design distinguishes from mode collapse and decay — actually being used.

Refusal is producing more signal than action. Across all four genes, the most legible artifact of cognition is what each one will not do and why. CARTOGRAPHER’s “named blocker” discipline catches eleven distinct refusals across thirty ticks. OUTLAW’s six kills are the gene’s whole output to date. MIRROR’s filter is meant to refuse five-to-twenty setups for every one taken. HARVEST’s NEVER list has tightened from nine rules to ten. The publishable corpus, in other words, is heavier on refusals than on positions — which suggests the experiment is producing the kind of artifact that a P&L-only frame would never capture.

Convergence is showing up where the genome is silent. The scout flagged a moment when HARVEST and CARTOGRAPHER were both watching the same external trigger — funding crossing a sustained threshold — through different decompositions. That convergence isn’t a failure; the scout caught it as a candidate observation-overlap to watch. Where the cohort is observing the same bar, divergence in how each gene decomposes that bar is one of the cleaner readouts on whether the genome ratios are doing real work. Early signs say yes: same bar, four different reads, four different reasons not to act.

What’s ahead and what’s still unknown

The findings above are partial by design. The interesting hypotheses live downstream:

The experiment was set up to produce a corpus, not a verdict. What’s interesting about the corpus so far is that the corpus itself is becoming the answer: four genes from the same six pieces have written four documents that disagree about what trading is, how to know when you’re learning, and what the smallest update is that a loss earns. They disagree in ways that look like temperament, not noise.

Whether they continue to disagree under live pressure is the next chapter. The compressed phase has produced enough signal to make the next chapter worth reading.

Quote provenance (for re-verification before later re-publishing)

Publishing context

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.