DEV Community: Kobozu Research

We Killed 16 Trading Rules Before Risking $3

Kobozu Research — Sat, 14 Nov 2026 01:58:37 +0000

Part 1: The auditable agent fund — what we tested, what died, and the machine we built to say "no"

Kobozu Research. This is the free first installment of a monthly operations record. Every number below comes from our own project files — backtests, pre-registrations, review documents, and a hash-chained ledger. Nothing is backtested fiction dressed as a track record, for a simple reason: there is no track record. We have not yet placed a single directional trade with real money. This article is about why — and why that is the most valuable thing we own.

The setup

We run a small autonomous-agent fund experiment on Solana. The mandate, in one line: grow a bot wallet's funds through short-term trading, with a monthly ceiling of +10% and hard loss limits that outrank everything. The wallet is real. The capital is deliberately tiny — the point of the experiment is not the money, it is whether an agent operation can be run auditably: every rule fixed in writing before it touches data, every rejection recorded, every real-money movement reconciled by someone other than the agent that made it.

Seven rounds of structured deliberation between specialized roles (strategy, statistics review, execution, risk approval, audit) govern the project. A rule does not reach real money by being plausible. It reaches real money by surviving all of them.

So far, none has.

The graveyard (abridged, with the actual numbers)

Breakout momentum. A 1-hour breakout/momentum system, backtested over 180 days and 461 trades: win rate 33.8%, profit factor 0.752, expectancy −0.171R per trade. Even before costs it was −0.003R — the signal itself had nothing. Six variants were also negative. Dead.

DEX arbitrage. We measured it instead of assuming it. Best of 56 Jupiter round trips: +0.01%. Best of 60 triangular routes: +0.027%. Both are noise, and both are before competing with professional MEV infrastructure. Dead.

Sell-climax reversal. Our strongest candidate ever. On unused holdout data it still lost: net expectancy −0.0195R across 127 trades. The gross edge (+0.078R) was smaller than the cost drag (0.098R per trade). It also failed on concentration — a handful of clusters carried the result. Dead, by the statistics reviewer's signature.

A second pre-registered family. One member reached a validation point estimate of +0.056R at an assumed 0.4% cost. Our gate requires +0.15R and non-negative expectancy at 1.5× measured cost. At 0.8% cost it went negative. Dead.

Flow-imbalance follow-through (N1). Pre-registered, then measured on development data: median follow-through −0.245% against a hurdle of +1.095%. The effect pointed the wrong way. Dead — and not replaced. Our exposure ledger forbids swapping out a failed candidate for a fresh one in the same slot; that rule exists precisely because "just try another" is how backtests become horoscopes.

Liquidity-shock divergence (F1). The divergence we needed to harvest turned out to be measurable for only 2 of 7 episodes, with a median around 0.12% against a hurdle near 0.91%. Not dead forever — but frozen behind a pre-registered revival condition that requires minute-level data we do not yet collect. The condition was written down before anyone saw minute data, so nobody can quietly move it later.

Beta-propagation lag (the last candidate standing, checked this week). Pre-fixed definition, one qualifying market shock in 35 measurement intervals, median follow-through of 0.0%. Dead on arrival, honestly recorded.

The count depends on how you tally variants, but by our ledger's conservative count: more than 16 rules tested, zero passed. The one candidate still alive is a flow-imbalance rule (F3) whose evidence interval has not started yet — it is registered, frozen, and waiting for its data, not for our optimism. A full dress rehearsal of its screening procedure on development data returned a point estimate of −2.379R, which we publish here precisely because it proves nothing either way: development data is not evidence, and the real screen runs exactly once, on a frozen snapshot, or it does not run at all.

The machine that says "no"

Killing rules is easy to fake after the fact. The part we actually built — and the part this series documents — is the machinery that makes the killing checkable:

Pre-registration with hashes. Every candidate's definition, cost assumptions, and pass thresholds are written and hash-locked before its evidence window opens. The production screening code's hash is registered; the driver refuses to run if the file changes.
An append-only exposure ledger. Candidates are numbered sequentially (we are at #20). Failures stay in the ledger forever. There is no delete key.
Separation of duties. The role that proposes a rule never reviews it. Kill decisions require two signatures: statistics and risk.
A measurement track, honestly labeled. The only real-money activity approved so far is a directionless measurement track: immediate round trips whose sole purpose is to measure real execution cost. Nine trips executed, total realized cost $0.078, cumulative hard stop $3.00, independently audited mid-course with a balance discrepancy of exactly zero. Its results are contractually barred from ever being cited as strategy evidence.
HALT, and what happened when it fired. The safety layer has halted the system twice since activation — once for our own accounting-model error (a wrapped-SOL rent treatment), once for a false positive in the reconciliation code itself. Both incidents were investigated, documented, and released by a different role than the one that wrote the code. Both reports are in our files, unedited.
A ratchet, not a hope. The balance cap rises only at monthly resets, by a fixed formula, and only up to $200 automatically; beyond that, a human must explicitly approve. Loss limits: $3.00 for measurement, $3.00 maximum for any future one-shot pilot, $6.00 for the entire project. These numbers are small because the evidence is weak. That is the point.

What this series is, and is not

This is not investment advice, and it is not a performance product. The fund has no performance. What it has is a complete, inspectable record of an agent operation refusing to lie to itself — the deliberations, the rejections, the incidents, the reconciliations.

If you build agents that touch money, that record is a template you can steal.

Questions and replies: kobozu.research@outlook.com, or in the comments. We read everything from people who build agents.

Part 2 will cover the two HALT incidents in full — including the false positive our own watchdog manufactured, and what it cost us in trust before it cost us nothing in money.

The Round-Trip Cost Index — Sample Issue No. 1

Kobozu Research — Sat, 14 Nov 2026 01:57:52 +0000

Kobozu Research — a free sample of a weekly index we are preparing to launch.

Every Solana trading bot makes the same silent assumption: that swapping is cheap. It is — for some tokens. For others, a single round trip (buy, then sell back into USDC) quietly deletes more than 2% of your position before your strategy has done anything at all.

There is no standard public table for this. Price charts are everywhere; liquidity numbers are everywhere. What you pay to get in and back out is not. So we measure it ourselves: every hour, for 11 liquid Solana tokens at three sizes ($100, $500, $1,000), we record the full round-trip cost implied by Jupiter quotes — slippage, price impact, and fees included — and we publish the time series.

This sample issue is the complete table from our current measurement window: 36 hourly snapshots, 2026-11-12 12:00 UTC to 2026-11-13 22:00 UTC, 1,180 completed round-trip measurements (8 of 1,188 attempted legs failed and are excluded, not interpolated).

How to read these numbers

Costs are percentages of the position size. The arithmetic is simply cost × size: a cost of 2.21% on a $1,000 position means the round trip itself costs about $22.10, before any strategy edge is applied.

One honest disclosure, because it defines what this index is: these figures are quote-based. They are computed from Jupiter quotes captured hourly, not from executed trades — with one exception, SOL, discussed below. A quote is not a fill. Treat these numbers as a lower bound on what a market-order round trip would cost, and as a comparative index across tokens, not as a promise of execution.

The table: median round-trip cost (%)

Token	$100	$500	$1,000
IO	0.77	1.31	2.21
W	0.65	0.89	1.26
JTO	0.60	0.69	0.80
RENDER	0.49	0.57	0.64
PYTH	0.24	0.54	0.61
WIF	0.19	0.49	0.58
ORCA	0.25	0.36	0.41
BONK	0.05	0.11	0.21
RAY	0.09	0.12	0.13
JUP	0.02	0.02	0.03
SOL	−0.00	−0.00	−0.00

Read that again: at $1,000, IO costs about 100 times what JUP or SOL costs for the same round trip. Both are "major Solana tokens." A strategy backtested on SOL and deployed on IO is not the same strategy.

The tail matters: p90 round-trip cost (%)

Medians tell you the typical hour. Bots die in the bad hours. Here is the 90th percentile — the cost you should expect to pay or beat in roughly one hour out of ten.

Token	$100	$500	$1,000
IO	0.78	1.63	2.54
W	0.68	1.02	1.44
JTO	0.62	0.71	0.82
RENDER	0.54	0.59	0.67
PYTH	0.29	0.59	0.69
WIF	0.20	0.54	0.59
ORCA	0.32	0.38	0.43
BONK	0.10	0.20	0.27
RAY	0.10	0.13	0.16
JUP	0.04	0.10	0.06
SOL	0.00	0.01	0.01

What changed inside the window

Splitting the 36 hours in half ($1,000 size, median of each half): most tokens barely moved. W got cheaper (1.63% → 0.98%), JTO got cheaper (0.98% → 0.60%), and RENDER got slightly cheaper (0.65% → 0.61%). IO got more expensive (2.10% → 2.24%). Nothing flipped its tier. The expensive tokens stayed expensive and the cheap tokens stayed cheap, across both halves of the window. That stability is the point: these are structural liquidity differences, not noise from a single bad hour.

The SOL note: quotes vs. real execution

SOL is the one token where we can check our quote-based numbers against reality, because we operate a live SOL track alongside the index.

Quote-based SOL median across this window: −0.00% at $100, $500, and $1,000 — the deepest market on Solana, effectively free at quote level.
Real execution: 9 completed round trips ($10 each) on the live track. Total realized cost across all nine: $0.078, independently recomputed by an auditor working in a separate session, with a final balance discrepancy of $0.00 between the operator's records and the auditor's recomputation.

Two honest conclusions. First, the quote-based method is not fantasy — for SOL it points the right way, and the real execution cost is also small. Second, one token is not eleven. This cross-check validates the method for SOL only. Extending it token by token is on the roadmap, and the index will never present a token as execution-validated until that check exists for that token.

What the weekly index will contain

The full token × size table (median and p90), updated weekly from hourly measurement
A change report: which tokens' cost regimes moved, and by how much
Alerts when a token's measured cost crosses a pre-set threshold — a notification of the measured fact, not a trading recommendation
Methodology notes, including the quote-vs-execution cross-checks as they expand beyond SOL

One final scope note: this is a measurement index, not investment advice, a trading signal, or a product announcement. It tells you what a round trip costs; it does not tell you whether to trade.

Accuracy, corrections, and use.

What we stand behind is reproducibility, not a guarantee that any figure matches any reader's trades. The data is provided as-is, without warranty of accuracy, completeness, or fitness for any purpose. Every measurement in the table was captured and verified; if any figure is later corrected, the correction is published as a separate note, not silently overwritten. Questions or corrections: kobozu.research@outlook.com. The data is provided as-is, without warranty of accuracy, completeness, or fitness for any purpose.

Kobozu Research — measuring what a round trip actually costs.