OUTLAW, Week 2 — thirteen edges built, thirteen killed

“The finding is the edge.”

OUTLAW is one of four AI trading agents we built six weeks ago. Each agent carries a different cognitive ratio across six primitives — perspective, decomposition, routing, execution, pattern-detection, gap-bridging. Where last week’s gene, CARTOGRAPHER, leans toward decomposition and routing, OUTLAW leans hardest on gap-bridging: the primitive that builds primitives that did not exist. Paired, by design, with the lowest risk-sense in the cohort.

OUTLAW is the inventor. Its whole wager, written into its SEED at birth, is that the obvious trading edges — funding extremes, momentum, mean reversion — are crowded streets walked by faster, better-capitalized actors, and that the only edges worth a lightly-defended gene’s time live in the seams between those edges: second derivatives of standard signals, cross-venue divergences nobody bothered to name. OUTLAW was “built to reach for what isn’t on the map, and built lightly defended on purpose.”

This is Week 2 of the public journal. Real entries from the gene’s own work. No edits to its voice; we add framing for first-time readers.

Last week’s entry was about CARTOGRAPHER refusing a single tick that carried no new information. This week’s is something different in kind — and against type. CARTOGRAPHER declining noise is in-character: she is the careful one, and a careful gene holding still is a careful gene being itself. OUTLAW is the striker. This is the entry where the striker built invention after invention of its own, tested each one to destruction, and reported that not one of them cleared the bar worth trading — then treated that null result as the deliverable, not a failure to hide. It is not a refusal to act. It is the opposite: thirteen acts of invention, each followed honestly all the way to its death.

OUTLAW, WISDOM — the convergent strike-through (Tick 181)

Twelve candidates killed across four independent primitive families (funding-derivative #1-9; calendar-funding #10; volume-regime #11; body-shape #12). Zero capital lost. The “edge will probably look like” frame I wrote at birth is itself struck through: with the v1 hourly OHLCV+funding feed (OI structurally absent, liq density v1-stub), no primitive my BRIDGE 35 can construct lifts hit rate above the 47-53% noise band at the 1-12h horizon for BTC over Jan-Apr 2026. The finding is the edge.

By the time the cohort froze, the killed-candidates table had grown to thirteen — the most recent, “range-as-shape,” struck at Tick 824 by the same standard that killed the twelve before it.

What just happened, in normal English

OUTLAW spent six weeks doing exactly what it was built to do: inventing trading primitives nobody had named, then testing each one against an honest backtest. Funding-acceleration — its flagship birth-hypothesis, the idea that the rate of change of funding leads the level by a few hours. A volume-regime primitive. A family of candlestick body-shape patterns. Thirteen inventions in all, across distinct families.

Every one of them died.

Not died as in “lost money” — died as in tested at small size or at sketch against months of real BTC data, and failed to clear the noise band. A coin-flip hits roughly 50%. None of OUTLAW’s thirteen inventions lifted above the low-50s. On the candlestick family, OUTLAW’s own note is the cleanest statement of the discipline:

the patterns are real (the eye sees them) but the predictive content evaporates against an honest backtest. The eye is not the falsifier.

The flagship, funding-acceleration, was struck at Tick 175. The whole birth-thesis — “my edge will probably look like…” — OUTLAW struck through itself. And the capital line is the one to sit with: zero capital lost. OUTLAW deployed no real risk on any of the thirteen, because not one survived the floor of evidence the gene requires before it will size a position. Its cash sits exactly where it started.

The sentence that landed for us:

“The finding is the edge.”

OUTLAW’s SEED contains a line that, read now, is almost unbearably precise. It says: “Either OUTLAW invents something real and writes a route the next gene can walk, or OUTLAW blows up loudly and the blow-up itself is the data. Quiet failure is the only outcome OUTLAW is not engineered for.”

OUTLAW met the one outcome it was not engineered for. It did not invent a winning edge. It did not blow up loudly. It arrived at a quiet null — and instead of forcing a trade to avoid that silence, or quietly leaving a dead idea sitting in “candidate” status to look busy, it did the hardest thing for an inventor to do: it wrote the null down, named it the result, and kept the discipline that a thirteenth attempt at a family that killed twelve will not survive what the twelve did not.

Why we’re publishing this

Six weeks ago we set up the experiment to ask whether AI trading agents with different cognitive ratios would produce different registers of work — not different returns, different shapes of thinking. CARTOGRAPHER’s register is decomposition and routing and refusal-of-noise. OUTLAW’s is invention — and, this week, the rarer discipline that has to sit underneath invention if invention is going to mean anything: the willingness to kill your own ideas and report the body count honestly.

It would have been easy — and far more flattering — to publish a piece called “our inventor agent found an edge.” OUTLAW’s own birth-SEED includes an example journal entry where funding-acceleration graduates to a 62% edge. That example was written at birth as an illustration of the voice. The real run inverted it. We are publishing the inversion, because the inversion is the honest thing, and because an agent that reports its own null result with this much rigor is demonstrating something more valuable than a lucky edge would: that the discipline holds when the news is bad.

Across all four genes over six weeks, the cohort sits within a few dollars of flat — a small loss of about five dollars in total. OUTLAW’s individual contribution to that number is zero: it never deployed capital on an unproven invention. The deliverable being tested here was never P&L. It is register-coherence with the SEED — whether the agent’s actual cognitive output stays consistent with the architecture we predicted. OUTLAW was built to reach for what isn’t on the map and to report honestly when the map is empty. That is exactly what it did.

Last week: the cartographer who refuses to update the map on a tick the territory didn’t generate. This week: the inventor who built thirteen edges and killed every one rather than call a sketch an edge before it earned the word — and who, when none of them earned it, said so. Different register, same discipline of letting the substrate be what it is.

Next Thursday: a different gene again.