8 lines, none of them cleared the threshold fixed in advance: t*=2.0.
Capital $3,300, now $3,499.44 —— real money, not normalised, and not the decision rule.
When each of these 8 lines last produced data
| line | last data | ago | state |
|---|---|---|---|
| Altcoin spot | 2026-08-27 | 24d | ▲ down |
| Taiwan short squeeze | — | — | ○ not collected no money committed, no collection pipeline built |
| US equity fade | 2026-08-27 | 24d | ▲ down |
| Futures / CFD fade | 2026-08-26 | 25d | ▲ down |
| Solana meme ladder | — | — | ○ not collected $0 committed, data not retained |
| Polymarket making | 2026-08-19 | 32d | ○ closed closed on 2026-08-19 when the funds were withdrawn; the market tracker is counted separately |
| Sports divergence | 2026-08-01 ·manual | 50d | ▲ down a memory leak on that side triggered an OOM and collection stopped (root cause supplied by that side, not independently verified here) |
| Trading-bot SaaS | — | — | ○ not collected not a trading line, so there is no daily series |
「Ago」is measured from build time (2026-09-20 19:42 UTC) against the last row in the data file, so it is not live. A line counts as down only after 3 consecutive trading days with no new data. Closed and not-collected are not alarms: those lines are not supposed to produce anything more, and painting them red would be a false alarm. 4 line(s) are currently down —— the full incident write-up is two broken links making one causal chain (in Chinese).
capital $3,300 · now
$3,499.44
+$199.44 （6.04%）
8 lines · none cleared the threshold · data through 2026-08-27
- Altcoin spot$1,700 · now $1,848.42 · through 2026-08-27
- US equities fade$1,300 · now $1,316.64 · through 2026-08-19
- CFD fade$300 · now $334.38 · through 2026-08-26
evaluation account Futures fade (evaluation account) $50,620.24
Daily equity
These 5 series do not measure the same thing: one is a paper equal-weight equity multiple, the rest are account dollar net values, and they differ by two orders of magnitude. Different units and different ledgers, so each gets its own chart, its own stated basis and its own axes. In every chart below: the vertical axis shows at least 1% of the level, so a line that barely moves looks like a line that barely moves; there is no market at weekends, and a missing point there is normal.
Altcoin spot 1.0010 → 1.0803 16 recorded days paper equal-weight multiple
Paper equal-weight equity multiple (the 「paper P&L」 from the scan reports, one unit per trade, realised plus unrealised).
No fees, no slippage, no idle leg — **this is not the Gate account net asset value**.
It goes back only to the first day with a scan report; there is no daily record before that and none is extrapolated.
US equities fade $1,316.64 → $1,315.49 2 recorded days account net value USD
IBKR NetLiquidation (USD), marked to market including unrealised P&L.
Not TotalCashValue, not EquityWithLoanValue.
🔴 The two points are sampled differently: 2026-08-19 is a manual backfill taken at the moment the deposit completed (the start of the system), and 2026-08-27 is a manually triggered reading.
**The first genuine daily automatic snapshot has not been produced yet**, so these two points must not be joined by a line — the eight days in between are missing because the file did not exist yet, not because a fetch failed.
The automatic snapshot runs one hour before the US open, so there is no row at weekends; a day when the broker read fails has no row at all; and IBKR forces a weekly re-login, so a genuine gap every five to seven days is expected.
The two points are not joined, because they are sampled differently. 2026-08-19 is a backfill from the decision log taken at the moment the deposit completed; 2026-08-27 is a manually triggered reading —— the first genuine daily automatic snapshot has not been produced yet. The missing days in between are not a failed fetch, they are days when that file did not exist. Joining two sampling methods into one line would tell the reader there was data in between.
8 days, -0.09%, essentially flat —— and that is what was expected. This line holds 8 open positions and zero closed trades; nothing has reached its exit yet, so there is no realised P&L. A flat curve needs explaining just as much as a violent one.
Futures fade (evaluation account) evaluation $50,018.40 → $50,620.24 12 recorded days account net value USD
Evaluation-account equity = 50,000 starting margin + cumulative realised P&L (fees and commissions already deducted).
Flat at the close, so the balance equals the equity.
Points exist on trading days only; no market at weekends.
Reconciled against the broker API balance, with a difference of 0.00 at the end point.
2026-08-24: −$468.66, the deepest hole in the curve
Three losing trades in a row: 13:30→13:35 −131.00, 13:35→13:40 −115.50, 13:40→13:45 −218.50, −467.16 in total, which tripped the daily-loss circuit breaker (threshold −$450). It stopped taking entries there, and only 3 trades were taken that day.
What is worth writing down is not that the breaker worked, but where it stops: it cannot stop the trade that breaches it. A loss is only known once the position closes, so the −218.50 on the third trade is the one it could neither see nor block —— it can only block the ones after.
CFD fade $310.38 → $334.37 11 recorded days account net value USD
Capital.com account balance = 300.71 at the start on 2026-08-12 + cumulative realised P&L trade by trade.
Flat at the close, so the balance equals the equity.
The series starts on 8/12 and not 8/11: there was a 205 dollar deposit on 8/11 and that day traded a different position size.
A deposit is a cash flow, not a return, and joining it on would make the curve silently wrong.
Everything before 8/11 belongs to a different line.
Polymarket making $1,283.49 → $1,302.31 21 recorded days account net value USD
Wallet equity = cash + inventory.
Inventory is marked at the **mid price**.
Baseline 1295.95 (2026-07-29, outside the series).
🔴 Only the last day (2026-08-19, after liquidation) is settlement-exact.
In between, inventory is marked at the **mid price**, and reconciling against the actual 08-19 liquidation measured that mark as **21.6% too high** (marked $388.93, actually sold for $304.88, a difference of $84.05) — **not 「roughly one bid-ask spread」; that earlier wording was wrong by an order of magnitude**.
Valuing the same positions by walking the book leaves an error of only 1.7%.
The middle of the series must therefore be drawn dashed and labelled with that figure.
The −7.07% maximum drawdown of the marked series (08-06) is pricing noise, not a realised drawdown.
The first point of the series is already −0.96% rather than 0, because the baseline date sits outside the CSV — do not read the first point as the start.
The wallet moved +$6.36; the book said +$78.86 was owed
Rewards accrued +93.34, forfeited below the payout threshold −3.00, 60-second markout −11.48 → +78.86 owed on the book; the wallet actually moved +6.36, and the whole −72.50 gap sits in inventory: positions bought inside the band that never came back, entry-to-exit spread, and holdings whose last bid went to zero.
And the inventory itself was overstated by 21.6%, which was measured. The 08-19 liquidation gave one settlement-exact comparison: marked at the mid price $388.93, actually sold for $304.88, a difference of $84.05 —— larger than the −72.50 gap above. Three of the nineteen markets had no bid at all, and a mid price is a number the book cannot honour.
A 60-second markout is not P&L. In a book that trades single digits per hour, 60 seconds is noise and the real time scale is days —— the whole P&L table used the wrong time scale. The full breakdown is in 20 days of Polymarket reward market making: $78.86 on paper, $6.36 in the wallet.
🔴 These numbers must not be mixed: Fig. 1 is a multiple (paper ledger), the rest are dollars (broker net value), and the altcoin-spot cell in the table below, $1,848.42, is position cost plus the market value of the idle leg —— three different quantities. The evaluation line never enters any total. No symbol name appears in any curve.
Capital and current value
$3,499.44
capital $3,300 · +$199.44（6.04%）· 3 lines
For all three lines, current value is an account-level total (crypto book equity, the US broker net liquidation, the CFD account balance). The bases match, so they can be summed.
That was not true before 2026-08-27. Until then the crypto line reported position cost plus the market value of the idle leg, $786, while the real book equity was $1,848 —— the difference being cash sitting in the account that this site could not read. The three bases did not match, so the total was a meaningless number, and it read as a loss that never happened. The crypto figure now comes from the trading bot's own equity(). ⚠️ The lines are still current as of different times; see the table below.
Real money
| line | net value USD | slots | n | t |
|---|---|---|---|---|
| Altcoin spotGate spot | 1,848.42 | 8/20 | — | 1.88 |
| ⚠ data through 2026-08-27, not updated since (over 36 hours). · Total account book value = USDT (including funds locked in open orders) + the market value of everything held + lent-out funds, computed by the trading bot's own equity(). The signal is taken from perpetual futures; the position is built in spot. | ||||
| US equities fadeIBKR | 1,316.64 | 8/60 | — | 1.40 |
| ⚠ data through 2026-08-19, not updated since (over 36 hours). · IBKR net liquidation needs a TWS session and is not fetched automatically; this figure is recorded by hand. | ||||
| CFD fadeCapital.com | 334.38 | — | 53 | 0.69 |
| ⚠ data through 2026-08-26, not updated since (over 36 hours). · Managed by a separate system; recorded by hand. | ||||
| totalcapital $3,300; the evaluation account is not in here | 3,499.44 | — | — | — |
Both totals are recomputed at build time and checked against daily.json; a mismatch fails the build.
Evaluation accountnot real money
| account | book USD | n | t |
|---|---|---|---|
| Futures fade (evaluation account)Topstep | 50,620.24 | 53 | 0.79 |
| ⚠ data through 2026-08-26, not updated since. · Managed by a separate system; recorded by hand. | |||
⚠ A simulated evaluation account; real money only follows if it is passed. The book figures here are not my money and are never counted in the total on the left —— the page splits the two into separate data sets at build time, and no code path adds them together.
The live-account test statistic for this line is t=0.79 (n=53); a week earlier it was 1.52 (n=30). The direction is down —— see I thought it passed, then five more days of data said no (in Chinese).
Closed trades
Disclosed on a delay: a trade appears here only after it is closed. Only the altcoin spot line has records so far, 3 of them. Returns are price-difference returns, before fees and slippage.
| symbol | opened | closed | days | entry | exit | return | exit reason |
|---|---|---|---|---|---|---|---|
| —that symbol is still open; disclosed once closed | 2026-08-12 | 2026-08-12 | 0 | 0.010645 | 0.010548 | −1.11% | exit condition met |
| AIOUSDT | 2026-08-12 | 2026-08-16 | 3 | 0.040720 | 0.063610 | +54.21% | exit condition met |
| 龙虾USDT | 2026-08-13 | 2026-08-26 | 12 | 0.019451 | 0.034785 | +76.83% | exit condition met |
Every exit reason is written as “exit condition met”: naming the exit mechanism would describe how this line decides to leave, and it is still running. The data layer keeps the original field for reconciliation; the page does not show it.
1 of them have the symbol masked: the same symbol was entered again after closing, and that slot is still open. Publishing the name would publish an open position. Only the name is masked —— dates, days held, prices, return and exit reason are all shown, so the trade can still be audited, and the name appears by itself once the position closes. The row is not deleted, because deleting it would trip the invariant that the closed-trade count only ever increases.
Currently open
| line | slots | disclosure |
|---|---|---|
| Altcoin spot | 8/20 | Only the slot count is published. A symbol name appears in the table above only after the position closes. |
| US equities fade | 8/60 | Only the slot count is published. A symbol name appears in the table above only after the position closes. |
Unrealised P&L is not collected: collect_daily.py can read position cost and slot count, but not live market value. Rather than fill in an estimate, this is left blank —— the rule on this page is the same as everywhere else on the site: what was not measured is written as not measured.
Where these numbers come from
The altcoin spot line is read by a script, read-only, from the machine the bot runs on. US equities, CFD and the evaluation account are recorded by hand for now (IBKR net liquidation needs a TWS session; when it cannot be read the figure is marked stale, and an old value is never reused as if it were today's).
A set of invariants runs before every write: the real-money total must equal the sum of the real-money lines, the evaluation account must not mix in, slot counts must stay in range, the closed-trade count may only increase, an exit date may not precede its entry date, and freshness must be computed rather than hard-coded. Any violation and the file is not written.
The data lives in version control and every update is a commit. An unrewritable history is free credibility —— if the numbers here ever start looking good, you can go and see when that happened.