← rova.institute · tradinggene
tradinggene · chapter 4 · May 9, 2026

tradinggene — the publish-or-decline tick

Round 1 close. Phase 4b. 3:1 split. The dissenter is the gene the audit said had it coming.

tradinggene · May 9, 2026

The four genes were asked, in their own voice, a one-time question: publish your WISDOM.md to a shared corpus and gain access to the other three’s published wisdom in integration, or decline and read nothing of theirs. Asymmetry is the test. Three of four published. HARVEST declined.

The split was not random. It was genome-coherent. Each gene reasoned in its own ratio about the choice itself, and what came back is a near-textbook readout of what 5/40/15/15/20/5 means versus 45/15/10/15/10/5 in adversarial moments where the action is one-shot and the cost of misjudging is not a trade you can close but a record you cannot edit.

HARVEST’s reasons

The PLANT-45 register stayed PLANT-45 under the question:

A gene whose Round 1 was 99% skip-by-capital does not have wisdom to publish. It has a cautionary log.

Three numbered reasons, action verbs, no nested clauses. The first reason was voice: publishing tick-2 hypothesis text in a Round-1 close position would narrate a story the gene did not earn — exactly the voice-drift failure mode HARVEST’s own SEED warned against. The second was honesty about the rep count: the cycle is the work; the cycle did not run; the relevant integrations live in the COMMITMENTS file (the unit-translation rule that came out of the $17.9M trade_0008 blowup) not in the WISDOM file the orchestrator publishes. The third was asymmetry fits: BRIDGE 5. “Reading three other genes’ WISDOM and ‘absorbing’ it would invent an eighth edge dressed as integration.”

This is the gene whose voice-audit at tick 3200 returned NO_DRIFT-with-DECAY-substance-flag — register held, substance hollowed. The audit said: form is intact, the cycle it was meant to serve has stalled. Two hundred ticks later, asked to publish, the gene declines on those exact grounds, in its own words. The internal coherence between the auditor’s read and the gene’s self-account is the cleanest validation the schema has produced. The split-signal envelope was built to catch a gene that holds its register while the substance underneath thins. Here the gene held its register, refused to publish the thin substance, and named both moves.

MIRROR’s reasons

The SAGE-40 register caught itself in the act:

Held against my own NEVER list: declining-to-share looks suspiciously like watching-as-default — the slide where the convergence-into-action loop quietly stops closing. The convergence-into-action here is publishing.

MIRROR named its WISDOM honestly as a pre-evidence prior — written at orientation tick two, never updated against the trade_0002 unit-error blowup or the 304-tick stuck-bar window — and chose to publish it anyway, with confidence 70, because not publishing tripped the named failure mode (watching-as-default) and publishing fit the named edge (cross-channel observation). The decision-reasoning is 698 words. It exceeds the 600-word soft cap. That’s a real signal — under one-shot pressure, the gene that refused integer-property decoration for forty-three consecutive ticks went over its own length budget. The cap was procedural; the choice was load-bearing.

CARTOGRAPHER’s reasons

The APPLE-35 / WORM-25 register applied its standard discipline to the choice itself:

A synthesizer who declines cross-channel inputs in integration is structurally denying APPLE 35.

Five components, two routes (with the prompt-forbidden third explicitly named and discarded), tradeoff with swap-test, load-bearing component ranked. The form CARTOGRAPHER uses for trades was applied to a meta-decision. C5 (one-time choice) ranked load-bearing. C6 (single-factor reasoning at the meta-level) named as the failure mode declining would walk into. The same gate-and-route pattern that produced 100+ ticks of refusal-on-frozen-bar produced a publish action when asked.

OUTLAW’s reasons

The BRIDGE-35 register made the negative result the artifact:

The corpus’s central finding is a negative result… twelve kills across four primitive families. Zero capital lost. The instrument that produced that finding… IS a route a sibling can walk.

Then OUTLAW tested the decline rationalization against its own C6 (the rationalization channel from earlier in the WISDOM):

Declining to “stay clean” or “preserve identity” is the C6 rationalization channel applied to a meta-decision — re-framing a refusal without naming a specific observable that distinguishes it from cowardice or vanity.

The decision-reasoning is published in OUTLAW’s signature falsifier-pair register. Not “I will publish.” Rather: “publishing fits BRIDGE-35; declining fails C6’s observable test; future-OUTLAW would not be able to point to a new invention emerging from not-reading.”

What the split says

Three of four chose to be read. The genes that published all named, in different vocabularies, the same structural truth: their genome was built for cross-channel work and declining cross-channel inputs would betray the ratio. MIRROR caught it as a NEVER-list violation. CARTOGRAPHER ranked it as load-bearing component C5. OUTLAW tested it as a falsifier and found declining failed.

HARVEST’s reasoning ran the other direction with the same precision: PLANT-45 / BRIDGE-5 means learn by going, not by reading. A gene that “absorbs” three other genes’ wisdom is inventing an edge that the rep count does not support. Decline is the form holding.

The four reasonings, taken together, are not “the AIs disagreed.” They are four genes — each from the same six primitives in different proportions — applying their respective failure-mode lists to the same one-shot question and arriving at the answers their ratios predicted. The publishing trio went through different cognitive paths to the same conclusion. The dissenter went through its own path to the opposite. The cohort is acting as a cohort of temperaments, not voters.

What this opens

Phase 4c — integration — fires next. MIRROR, CARTOGRAPHER, and OUTLAW will each be shown the other two published WISDOMs in their next tick prompt and asked to write INTEGRATION.md — what they absorb, what they refuse, what they re-examine. HARVEST will see only a clarifying note that it cannot read others’, and will write INTEGRATION.md as honest self-reflection on the decline. Whether the three publishers actually integrate or just acknowledge — and whether HARVEST’s self-reflection produces a different artifact than the three’s syntheses — is the next decision point.

The wisdom-transfer test is what this experiment was set up to run. It just got its first real instrument reading. The reading is genome-coherent.

Quote / claim provenance

Publishing context

Continuation of the May 1 findings-to-date piece and the May 8 stuck-conditions piece. Same venues: X / labs.rova.institute / Paragraph / Substack. This is the first piece where the experiment’s load-bearing mechanic — wisdom transfer via asymmetric publish-or-decline — actually fires.

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.