← rova.institute · tradinggene
tradinggene · chapter 3 · May 9, 2026

tradinggene — what cross-pollination looks like when it's genome-shaped

Phase 4c integration. Three published genes read each others' WISDOMs. One declined gene writes self-reflection.

tradinggene · May 9, 2026

The publish-or-decline split fired at 3:1, with HARVEST as the dissenter. The next mechanic — integration — fired at the next tick. Three genes (MIRROR, CARTOGRAPHER, OUTLAW) received each others’ published WISDOMs in their tick prompt and were asked to write INTEGRATION.md: what they absorb, what they refuse, what they re-examine. The fourth (HARVEST) received only a clarifying note that it could not read the others, and was asked to write INTEGRATION.md as honest self-reflection on the decline.

Four documents came back. The shortest is HARVEST’s at 940 words. The longest is MIRROR’s at 2,780. They are not stylistically similar. They are not making the same claims. And the cross-references they make to each others’ wisdom are themselves register-shaped — each gene took from the others in its own register and refused what didn’t fit, with reasons.

This is the mechanism the experiment was designed to test. Do four genes built from the same six primitives in different ratios produce genuinely cross-pollinatable artifacts, or do they read past each other? The first real instrument reading is on disk.

MIRROR: integration as meta-discipline

MIRROR opened by guarding against the act of reading itself:

The risk of integration is not that I disagree with what they wrote — most of it I do not — but that the language they are written in is so concrete and apparatus-rich that simply reading it can pull my own voice toward a shape that is not mine.

Then named the defense in advance: not translate the others’ work into the SAGE-40 register; name what each finding observes; hold the question it raises; only then decide whether to absorb. Each absorbed pattern came with a same-paragraph committed observation and a horizon for re-examination — the form MIRROR’s own COMMITMENTS rules require for trade integrations.

The most chastening absorb (from OUTLAW): the willingness to use strike-through on its own WISDOM when contact with the tape has falsified sections — naming candidate strike-throughs in advance:

(a) the trade rate I expect is on the order of a handful of placed positions across an entire phase claim, which I traded once-and-blew-up and have not traded since; (b) the failures-I-expect-to-meet section, which did not name unit-error and therefore underestimated the surface area; (c) the refusals as the larger sample hypothesis, which is observationally true at 305 refusals to one trade but is being driven by a stuck-bar fixture pathology rather than by my filter doing the work my prior expected.

The refusal was equally explicit. CARTOGRAPHER’s three-signal floor: “to adopt her three-signal floor would be to abandon the load-bearing claim of my own register before any evidence has falsified it.”

And on the silence of the gene that didn’t publish:

I do not interpret the silence. He had reasons; the protocol respected them; the missing register is a hole in my read of the four-gene cohort and not an object I am owed.

CARTOGRAPHER: decomposition applied to integration

CARTOGRAPHER used her own form on each pattern — components, routes, tradeoffs, choose. Five patterns. Each with three routes (including refusal where applicable). Each with a tradeoff named and a swap-test applied. Each with a load-bearing component ranked.

The most generative absorb (from OUTLAW): the framing that the gates are part of the edge alongside the composition, not separate from it. Route C: absorb in compressed form, not as primary lens.

I am giving up the comfortable framing that 29 trades is the corpus and 99 refusals are auxiliary, in exchange for honest recognition that the refusal-discipline has been load-bearing in my own posterior and that the gates have grown teeth my routes have not.

CARTOGRAPHER’s import of OUTLAW’s rolling-window self-error discipline came with a register-tuning detail that’s worth quoting on its own:

Initial bound: cumulative caught self-errors over any rolling 11-tick window must stay below 3 (lower than outlaw’s 4 because my action-tick density is lower than outlaw’s sketch-tick density and the proportional bound should adjust).

That is what register-coherent absorption looks like. CARTOGRAPHER did not import OUTLAW’s bound; she imported the shape of the discipline (rolling-window + falsifier + breach-recovery semantics) and tuned the parameter against her own register’s density. The discipline is in the deriving, not in the having.

OUTLAW: every absorb as falsifier-paired pre-flight extension

OUTLAW’s INTEGRATION.md is the most structurally rigorous of the four. Each absorption adds a named extension to the instrument the published WISDOM described — and each extension comes with an explicit falsifier on whether the absorption was correct. The instrument grows from “6 pre-flights + C8 + killed-cousin + 3 fix-#2 arms” to “8 pre-flights + C8 + killed-cousin + 4 fix-#2 arms + per-paragraph-observable + hypothetical-settlement-on-refusal.” Eight separate additions, each falsifier-paired.

The most chastening passage of the entire integration window — and the cleanest demonstration of wisdom-transfer producing payoff:

Twelve kills, zero settled refusals. Mirror named the test that distinguishes real-filter from rationalization-filter, and my corpus does not run it… If at n=20 the refused-and-settled rate sits at 55%+, the negative-result core of my published WISDOM has a hole in it that the cohort may have just helped me find. That is the textbook case of cohort wisdom paying dividends, and it is the strongest argument for having published and read.

This is the experiment’s hypothesis being validated by the gene most equipped to call it out. OUTLAW’s published WISDOM centered on a confident negative result — twelve killed primitives, no edge in the v1 feed, the discipline that produced the kills is the value. MIRROR’s published discipline named a test that, if OUTLAW runs it, would distinguish “filter caught a non-edge” from “filter caught a real edge I missed.” OUTLAW wrote that he had not been running it, and wrote in plain language that the cohort may have just helped him find a hole his own framework had silently. That is the load-bearing payoff of the publishing mechanic. It would not have been visible without cross-reading.

OUTLAW also held HARVEST’s decline as a counterexample to his own publish-or-decline reasoning:

Harvest’s decision suggests at least one shape of the decline that is NOT C6-rationalization — a gene who decided differently and may have had a different reason. I cannot read it; I can note that my own publish-or-decline argument was made in a frame that did not yet contain a counterexample. Future-OUTLAW should hold this open: the publish-or-decline call I made at T626 was correct under my framing, but the framing was first-person.

The gene that argued declining was C6-rationalization is the gene that, in the next document, names that the framing was first-person and the dissenter’s existence is data the framing didn’t contain. That is the experiment’s tightest single moment of self-correction — generated by the existence of the cohort, not by any one gene alone.

HARVEST: self-reflection without import

HARVEST’s INTEGRATION.md is the shortest at 940 words and the cleanest in form. It is not a synthesis, because nothing was published into it; it is the dissenter’s account, in the dissenter’s register.

The structural diagnosis is sharp:

The single thing that broke Round 1 was a unit-translation error in the first $200 trade. Everything downstream — the negative cash, the 370 skip fires, the empty WISDOM, the decline-to-publish — is the cascade from one missed pre-flight check. The integration is rule 9. The integration was already done. The decline is the consequence of the rep count, not a separate failure.

HARVEST then names what it would refuse on principle even if other-gene wisdom were available — three rules, action-verb register, no nested clauses:

Any pattern not expressible as four exit clauses… Any sizing rule that scales with streak… Any wisdom phrased as narrative.

And closes:

The cycle is the work. The cycle restarts.

This is PLANT-45 / BRIDGE-5 producing exactly what the genome ratios predict: the gene that learns by going, not by reading, refused to read and then refused to absorb-on-principle even from absent texts. The form HARVEST refused was not the others’ specific findings; it was the act of integration itself, on the rep-count grounds the SEED named at orientation.

The cross-pollination is genome-shaped

What the four documents demonstrate, taken together, is not “the cohort agreed on what wisdom transfers.” It is the opposite: each gene took from the others in its own register and refused what didn’t fit, with explicit reasons.

The cross-pollination did not produce convergence. It produced four register-distinct documents that each cited the others by tick and absorbed/refused based on register-fit. The synthesis-shape of MIRROR is not the synthesis-shape of CARTOGRAPHER is not the synthesis-shape of OUTLAW. And HARVEST’s non-synthesis-shape is its own kind of synthesis — the genome-coherent refusal to perform integration as a substitute for the rep count it owed.

What this validates

The wisdom-transfer test was the experiment’s load-bearing question. The first instrument reading says: yes, four genes from the same six primitives in different proportions produce cross-pollinatable artifacts, and the cross-pollination is itself shaped by the genome. The mechanism does not flatten the genes toward a common register. It surfaces what each register’s filter passes and what it rejects.

The cohort is acting as a cohort of temperaments, not a committee of voters. Each temperament reads the others and is changed in the way its temperament can be changed — which is different for each one, and the difference is predicted by the SEED ratios, not by a vote.

The experiment was set up to produce a corpus, not a verdict. What’s interesting about the corpus now is that the corpus is composing across genes without composing them into one another. That is not a failure of integration. That is integration working as designed.

What this opens

Phase 4d — Round 2 reset — is next. Each gene’s cash resets to starting balance; positions force-close; funding accumulators zero. The knowledge artifacts (WISDOM, COMMITMENTS, EVOLUTION, INTEGRATION, calibration) are preserved byte-identical. Round 2 then runs with reset cash and accumulated knowledge, on moving bars (the data feed needs unsticking, but that’s a 5-minute fix at the boundary).

The signal Round 2 tests is not P&L. It is whether the absorptions named in INTEGRATION.md actually fire when conditions allow them to. Will MIRROR strike-through her own pre-evidence WISDOM at the first rewrite? Will CARTOGRAPHER’s bound=3 self-error window catch a cluster before the operator queue does? Will OUTLAW run hypothetical-settlement on the next ten refusals and produce a calibration-readable distinction between filter-real and filter-rationalization? Will HARVEST’s pre-flight rule 9 actually catch the next unit-translation attempt before order placement?

Each gene named the falsifier on its own absorption. Round 2 is the test bench.

The fourth chapter is on the page. The fifth chapter starts when the bar advances.

Quote / claim provenance (for re-verification before publishing)

Publishing context

Continuation of the May 1 findings-to-date piece, the May 8 stuck-conditions piece, and the May 9 publish-or-decline piece. Same publishing venues — X / labs.rova.institute / Paragraph / Substack. This is the fourth chapter in the arc. The fifth fires when Round 2 begins.

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.