The Shishin live bot is paper-traded — orders are submitted to a simulated execution venue, not to the open market. The published track is therefore free of real-money slippage and fill quirks beyond what the simulator models. Real execution introduces real costs that the paper track does not pay.
Every algo manager's worst nightmare is the honest track record. Most never have to face it. Shishin chose to publish one from day one.
What the industry usually does
The standard pattern in quantitative trading is to publish a backtest, perhaps a curated set of historical fills, and a monthly performance letter. Behind the letter sit a thousand small decisions about which version of the system produced the number, whether to include the months when the system was off, how to attribute slippage, and what to do about the recent drawdown. None of this is necessarily dishonest. All of it is malleable.
The recipient of the letter is asked to assume the manager is acting in good faith. Often they are. The structural problem is that the recipient has no way to verify the assumption — the data behind the letter is not public, and the things the letter does not say are not visible. Performance disclosure under those conditions is a relationship management exercise as much as a reporting one.
What Shishin does instead
The Shishin live bot, paper-traded since 2026-06-08 with a starting NAV of $250.000, publishes the following information continuously and publicly:
- Every fill. Every entry, every exit, every top-up. Direction, ticker, quantity, simulated price, time-stamp, exit reason if it's an exit. The full ledger is queryable.
- Every open position. Live, refreshed intra-day. Cost basis, current market value, unrealised P&L, days held. No selective disclosure.
- Every NAV tick. The daily NAV time series is published from inception. Drawdowns are visible the day they happen, not the month they end.
- The active engine. Which of the four guardians the macro classifier picked today, and the state of the regime transition counter. The system's intent is observable, not just its actions.
- The signals that fired. The ranked board of candidates from this morning's scan, the composite scores, the setup states. The output of the ranking layer, separate from whether the bot acted on any given line.
The disclosure is end-to-end. There is no version of the system that produced a different number we chose not to publish. There is one bot, one paper-traded brokerage connection, one set of fills, one NAV. Everyone sees the same record.
Why this is unusual
Most managers don't publish in this way because the downside is asymmetric. The upside of transparency — a reputation for honesty, a credible track record, a defensible disclosure regime — accrues slowly. The downside of a bad month, with the bad month visible, is immediate and personal. The rational manager response, particularly when capital is being raised on the strength of recent returns, is to keep the bad months private and let the long-arc summary do the talking. We sympathise. We just don't accept the trade-off for our own work.
The disciplining effect
Publishing the live track has two effects, both of which improve the system over time.
The first is on the operator. It is psychologically impossible to convince yourself that a drawdown is “not really happening” or that a stop-loss was the wrong decision in retrospect, when the record of both is in public view and people are watching the line on a chart you didn't get to redraw. The discipline isn't just to follow the system; it is to have nowhere to hide if you don't. The fastest way to learn whether you actually believe your own rules is to publish them and watch yourself live by them.
The second is on the system. A system whose track is secret can survive on the strength of memory: the manager remembers the wins, narrative-builds the misses, and moves on. A system whose track is public cannot. The record is the record. The bot's hit rate, average winner, average loser, max drawdown, Sharpe — all of these will either hold up under live conditions or visibly fail to. The act of publishing creates an evidentiary feedback loop that pure backtesting cannot reproduce. Reporting and backfill bias have no purchase against a continuously published track — there is no chance to quietly drop the bad months after the fact.
What we expect from the live track
We expect the live track to diverge from the backtest track within reason. Paper-traded execution is cleaner than real, but the live tape is still messier than the historical sample. We expect months when the system out-performs the backtest's expectation and months when it under-performs. We expect at least one drawdown that is uncomfortable. We have already, in the first three weeks of live operation, experienced one trade we would not have taken on a re-read of the rules — a known edge-case the system handled correctly but inelegantly — and one regime transition the classifier got within two sessions of the actual market turn. Both are in the public record.
The live track is the only piece of evidence about a quantitative system that doesn't admit retrospective editing. We chose to publish ours because the alternative is to ask anyone evaluating our work to take our word for it. They shouldn't have to. The bot's record is the work. The record is what we're putting up against the only fair test of any algorithmic strategy: did it actually make money, and how would you know.