The closed eras.
Before 17 July 2026 no real money was ever at stake. Three eras came first — one simulated by us, two on an exchange that reset itself twice. They are over, the figures are frozen, and none of them count towards anything you see on the dashboard.
1 · The four eras, side by side
All four are measured with the same formula — the one the site uses today — so the columns can honestly be compared. That is not how they were measured at the time, and section 3 is about the difference.
| Era | Period | Days | Orders | Win / loss | Fees | Net result |
|---|---|---|---|---|---|---|
| Paper | 30 Mar → 8 May 2026 | 40 | 1,192 | 461 / 130 | $17.21 | +$47.57 |
| Testnet 1 | 8 May → 27 May 2026 | 20 | 64 | 20 / 3 | $1.88 | +$10.06 |
| Testnet 2 | 5 Jun → 6 Aug 2026 | 63 | 255 | 113 / 13 | $7.80 | +$45.13 |
| Kraken | 17 Jul 2026 → today | real money — still open | dashboard | |||
"Net result" is what was actually banked, after fees, over the life of the era. Testnet 1 and 2 also ended holding open positions ($520.83 and $301.58 at cost) which the exchange resets simply deleted — those are not counted as gains or losses, because nothing was ever settled.
2 · What each one was, and how it ended
No exchange at all: orders were simulated inside our own code. The Trend Follower roamed 56 coins because nothing punished it for trying.
Closed deliberately when the bots were wired to Binance's test exchange — the first time an order left the building.
Real orders on Binance's test exchange, simulated funds. Six coins. The first era where a fill could be refused by someone other than us.
Binance reset the test exchange without notice. Balances back to default, order history gone. $520.83 of open positions ceased to exist.
The longest era, and the one where the machine took shape: four grids, three advisory brains, the nightly reconciliation, the public dashboard.
Binance reset the exchange again, between 5 and 6 August. Same ending as Testnet 1 — which is why there is no Testnet 3.
3 · The same era, counted twice
When the paper era closed in May 2026, we published its result as +$69.05. Recomputed today, over exactly the same 1,192 orders, it comes to +$47.57 — twenty-one dollars lower.
Nothing about the past changed. The accounting did. That era counted profit with a FIFO queue that had three bugs we had not found yet, and it did not consistently subtract trading fees from the result. Both were fixed later — the fee unification in May, the average-cost rewrite in June — and every figure on this page uses the corrected method.
4 · What the archive shows that no single era could
Read in sequence, the four eras say one thing very clearly: the bots traded less and less, and it was the cost of trading that slowed them down.
| Era | Orders / day | Fee per order | Coins traded |
|---|---|---|---|
| Paper | 29.8 | $0.0144 | 56 |
| Testnet 1 | 3.2 | $0.0294 | 6 |
| Testnet 2 | 4.0 | $0.0306 | 4 |
| Kraken | 0.2 | $0.2702 | 2 |
Thirty orders a day on paper, where a trade cost a penny and a half. One order every five days on Kraken, where it costs twenty-seven cents. The strategy did not change between those two rows — the price of being wrong did.
5 · Where these numbers come from
The paper era lives in a database export taken the day it closed, 8 May 2026 — 1,192 orders. Testnet 1 and 2 are still in the live database under their own cycle tags, untouched. All three were replayed with the same average-cost engine the dashboard runs on, so the comparison is like-for-like.
These figures are frozen on purpose. They describe finished periods and will not move again — unlike everything on the dashboard, which is real money and changes while you read it.