Control room

System

source config + strategies/config.ts + RPC · data 0 s old · confidence high

The eight — what each one tries, and what decided it

🔴 15/09 — this table used to open the front page, eight rows of which seven said switched off. It is a weak thing to hand a reader first and it belongs here, where somebody has come to check. It is generated from the same registry the execution gate reads, so the page cannot disagree with the code that would spend money.

The circuit — profit, buyback, burn

source paper_positions + treasury_settlements · data 15 h old · confidence medium

PAPER mode: simulated profit buys nothing, and saying so is the point

Pending is not burned. Adding the two would publish a burn that never happened.

Performance

source paper_positions · data 15 h old · confidence medium

The two columns are never added together. They measure two different things, and a figure that mixed a simulation with real money would be correct under a false sentence.

What it actually did, regenerated every trading-day evening.

Every number in this section is produced by the same code that trades, the evening it trades — weekdays, which is when the book it reads is open. A page of measurements typed by hand ages into a lie, and this project has already had to correct two of its own published claims.

Across 10 sessions: 1852 positions closed, 1291 of them closed above cost, +0.297% mean.

The line is the on-chain price against the live reference. Filled dots are entries, hollow ones exits; green closed above cost, red below. The agent buys the discount and leaves when the discount is gone — whatever the price is doing.

What the engine did with its decisions

Four stages of the pipeline can only say no, and every refusal is written down with a reason drawn from a fixed list rather than a shrug. The largest one is not about a token at all: it is the agent stopping because its own data is behind.

The diagnostic that does NOT decide

The rule was written down before the data that tests it existed, and a scheduled job applies it rather than a person reading a table. What follows is that job’s reading at a grid of fixed entry points — across 5 fixed entry thresholds the mean runs from +0.217% to +0.764%, and every 95% interval sits above zero. It is not the measurement that promotes anything: the one that does is the walk-forward, which refits its entry for every day out of sample and whose verdict is published, with the age of its data, at the walk-forward endpoint and beside the figure it produced further up this page. Reading this table as a verdict is the mistake this section exists to make hard.

Generated 2026-09-25 at 12:12 UTC.

Pons radar

source launches + curve_trades + graduations · data unknown age · confidence unknown

Progress reads unknown on purpose: the real figure needs a read of the curve contract, and a row count is not that. Rendering it as zero would show forty curves standing at the start line.

Decisions — including the refusals

source decisions · data 16 h old · confidence medium

The refusals are the record, not the noise: a panel showing only the bets would describe a system that sees little, instead of one that is stopped often.

The house shows its losses

The criterion is written, committed and pushed before the measurement runs. Read in the other order, the same numbers would have made an excellent announcement.

What it has read

Tokenized equities — how dislocated the book is

source equities.jsonl + universo-equity.json · data 46 s old · confidence high

Coverage is inside the index, not beside it: between 7 and 11 September this agent ran blind for five sessions because a request asked for 37 instruments and the cap was 30 — zero reads, every light green. A median published over two instruments out of thirty-seven is a flawless number about a book nobody looked at.

Deviation is the token price against its reference feed. It is never a statement about the underlying security.

Walk-forward — does a threshold picked on the past still pay?

source equities.jsonl + backtest/walkforward.ts · data 46 s old · confidence high

A — the 10 seed tickers — 163 out-of-sample operations, positive in 13 of 13 days, top day 15%

B — all 37 — 812 out-of-sample operations, positive in 13 of 13 days, top day 14%

The criterion was written and committed before this was run (written and committed BEFORE the measurement ran; published at kuzanfloor/kakegurai-method, docs/criteria/walk-forward.md). A criterion written afterwards is fitted to the outcome and approves whatever came out. Nothing is switched on by any of the three verdicts.

⚠️ the cost is a flat 0.1% and does not scale with size: a measured return is not an executable return

Capacity — the size at which the edge dies

source backtest/capacita.ts (cached measurement) · data 13 h old · confidence medium

constant product: 2S/L per side, both sides. Overstates cost inside a v3 tick range, understates it outside.

Whether the cost model runs high or low is not settled: the claim that these figures were a ceiling was withdrawn on 13 September, after reading the pools directly. Liquidity is the binding constraint here, not the strategy: capacity scales with pool depth, and the thinnest instrument in the book holds under fifty thousand dollars. None of this is a verdict on the edge — that one is in the walk-forward panel above, and it says not yet proven.

Health

source scan_state + scan_gaps + RPC · data 9 s old · confidence high

Risk disclosure

KAKEGURAI is software that places bets. It can lose its entire bankroll. Nothing here is advice, an offer, or a forecast, and no figure on this page is a claim about what happens next. Buyback and burn are protocol mechanisms, never a statement about price. Every number here is also readable by a program at /api/status.json and the five files beside it, each carrying its own age.