Built in public
A trading discipline layer, built in public.
Project Anchor is a discipline-first trading operating system. It enforces risk limits, keeps an audited ledger, raises fault alarms and can stop trading. It is not a trading bot and not a signal service.
We are a small project testing our own rules in the open. So far, nothing we have tested has passed. We publish the failures too.
Not investment advice. No returns are promised or implied.
What it does
Anchor sits between a strategy and the account it runs on. Its job is to say no.
Daily loss line
Tracks the account's daily loss limit and keeps every trade inside it, with a safety buffer.
Drawdown floor
Tracks the account's hard drawdown floor. A trade that could breach it is not taken.
Skip when there is no room
Risk per trade = min($150, room to the daily line − $30, room to the floor − $30). If that is under $50, the trade is skipped.
Audited ledger
Keeps an append-only record of what happened and why, so results can be checked afterwards.
Fault alarms
Raises an alarm when market data, a service or a check stops behaving as expected.
Can stop trading
A kill switch can halt trading. Anchor does not place orders for other people.
Public test recordsLast updated:
Every number below comes from a file we keep, and carries the date it was recorded. Nothing here is a claim that any strategy works. If the “last updated” time is more than two days old, the daily updater has stalled and the figures are old.
Every test result, pass or fail, is also posted to our Telegram channel. Telegram channel
Scoreboard: 10 tested, 0 passedas of
9 are trading-strategy families. The other 1, EVAL_ACCT_MGMT_V1, is an account-management rule, not a trading strategy. All were tested against pass bars written before the run. OOS means out-of-sample. Click a name for the one-page report.
| Tested | Candidate | Out-of-sample result | Verdict |
|---|---|---|---|
| Oct 23, 2026 | FUNDING_CARRY_V1 | +4.0% a year, against an 8% bar | FAIL |
| Oct 23, 2026 | BASIS_MR_V1 | −31.9% a year; failed the fee gate | FAIL |
| Oct 23, 2026 | QUARTERLY_BASIS_CC_V1 | −0.17% a year, against an 8% bar | FAIL |
| Oct 24, 2026 | TSMOM_D_V1 | +0.046R per trade; does not beat the random-entry 95th percentile | FAIL |
| Oct 24, 2026 | FUNDING_CONTRA_V1 | Failed the fee gate; out-of-sample shown for reference only | FAIL |
| Oct 24, 2026 | HOUR_SEASON_V1 | −0.058R per trade; failed the fee gate | FAIL |
| Oct 24, 2026 | EVAL_ACCT_MGMT_V1 | No variant met the bar; we kept the fixed $100 risk | FAIL |
| Oct 26, 2026 | MULTICOIN_TREND_V1 | +4.7% a year, Sharpe 0.31, against a random-sign 95th percentile of 0.86 | FAIL |
| Oct 27, 2026 | REGIME_BULL_SURVIVE_V1 | Selected variant: +8.0% a year vs buy-and-hold +54.0%; drawdown −40.7% vs −53.0%; bull-market capture 29% against a 50% bar | FAIL |
| Oct 27, 2026 | REGIME_BULL_SURVIVE_V2 | Running, not a verdict. Checkpoints Apr 27, 2027 and Oct 27, 2027 | RUNNING |
| Oct 27, 2026 | REGIME_BULL_LEV15_V3 | Running, not a verdict. Checkpoints Apr 27, 2027 and Oct 27, 2027 | RUNNING |
| Oct 28, 2026 | REGIME_BULL_VOLTARGET_V1 | +78.7% a year vs buy-and-hold +250.3%; drawdown −37.1% against a −30% line; bull-market capture 26.7% against a 50% bar | FAIL |
Forward paper tracking (running, not a result)Last updated:
Two paper-money shadows of the bull-market survival idea, started on Oct 27, 2026 with 10,000 of paper money and run on real BTC prices. No orders are placed and no account is involved. They are not judged until the checkpoints on Apr 27, 2027 (interim) and Oct 27, 2027 (12 months). “Insufficient evidence” is the expected verdict unless both a real bull and a real bear fall inside the window. Rules are frozen; the rule hashes are below.
| Line | Paper equity | Since start | Worst drawdown so far |
|---|---|---|---|
| V2 (b): 200-day filter, volatility-targeted, 1× (judged) | 9,665.02 | -3.35% | -4.89% |
| V2 (a): plain 200-day filter, 1× (information only) | 9,665.02 | -3.35% | -4.89% |
| V3: 1.5× leverage with a 20% drawdown brake (judged) | 9,477.56 | -5.22% | -7.35% |
| BTC buy-and-hold (benchmark) | 9,665.02 | -3.35% | -4.89% |
Rows are UTC daily closes. Latest one in the ledger: the candle for Oct 31, 2026, closed . Days recorded: 6. V2 regime: ON. V3 exposure: 1.50×, brake OFF, liquidations: 0.
Paper equity, start = 10,000: ■ V2 (b) · ■ V3 · ■ BTC buy-and-hold
Rule hashes (SHA-256 of the frozen PARAMS files): V2 8092f576c2be… · V3 faea68bfd065…
Paper fills at the daily close with a flat 15 bp per side; real fills, outages and exchange rules are not tested.
Raw data: records.json · V2 ledger (CSV) · V3 ledger (CSV)
Sample lie-detector result: TSMOM_D_V1, daily BTC momentumtested Oct 24, 2026
A real test of one of our own candidate strategies, net of fees. It made money in-sample and out-of-sample, and it still failed.
| Trades | Mean R per trade, net of costs | |
|---|---|---|
| In-sample | 140 | +0.108 |
| Out-of-sample | 88 | +0.046 |
| Random-entry control, mean of 200 seeds | +0.002 | |
| Random-entry control, 95th percentile | +0.139 |
Verdict: FAIL. The out-of-sample +0.046R sits inside the range that coin-flip entries with the same exits reached, so we cannot credit the signal. Full test report
Rules SHA-256, locked Oct 24, 2026, 12:46 AM ET (04:46 UTC) before any backtest ran:
4657ac56e6f5c8eacec015fed9ff07d9f85ba84d06527695ed8d897c9a9c7b39
Disclosures: this was our own strategy and we were also the tester. The hash was not published before this run, so for this test the lock time rests on trust. The same batch had a process slip: out-of-sample numbers were computed for two families that had already failed the fee gate, and were labelled reference only. Since then we post the hash publicly before running. The full details are in the sample report below.
Evaluation status: Kraken Prop $10k account, MF4 BTC strategybalance as of
- Start
- $10,000
- Target
- $11,000
- Drawdown floor
- $9,400 (static)
- Last recorded balance
- $9,954.28, read by hand from the Kraken evaluation page on ; no open position
- Status
- Not passed yet
Our first live test is a Kraken $10k prop-firm evaluation running the MF4 BTC strategy. Kraken Prop has no API, so orders are placed by hand and the balance is recorded by hand. MF4 is our own strategy.
What our own tests say about it: MF4 backtested from 2024-10 to Oct 24, 2026 averages −0.07R per trade net of costs. A Monte Carlo of 20,000 paths from $9,954.28 gives roughly a 25–28% chance of ever reaching the target at any sensible risk size, against 27% for a strategy with no edge (Oct 24, 2026). In plain words: the guardrails are the point of this test, not an expectation that the strategy wins.
Free strategy test
If you have a trading strategy written down, we will test it for free and tell you honestly whether it passes. This is the “strategy lie detector”.
- Send your written rulesPlain words or pseudocode: entries, exits, stops, sizing, instruments, timeframe and your cost assumptions. Message us on X.
- We lock the pass bar firstBefore running anything we write down the rules and pass criteria and record their SHA-256 hash. You can publish the hash yourself. Nothing changes after the lock.
- We run the testsA fee gate, an in-sample / out-of-sample split, and a random-entry baseline: your exits with coin-flip entries, to see whether your signal beats luck.
- You get a plain PASS or FAILA full report, structured like the sample below, whatever the outcome. We publish it only with your consent.
Free. We only need your written rules. No account access, no API keys, no funds. We never trade for anyone and never manage money. A pass would be a statement about a historical simulation, not a promise of returns. Capacity is limited, so we may not be able to take every request.
Downloads (PDF)
What we won't do
- Trade for anyone, or manage anyone's money.
- Run copy-trading or send trade signals.
- Ask for your exchange logins, API keys or funds.
- Promise or imply returns.
- Publish your rules or results without your consent.
- Take money for anything on this site. It sells nothing.
- Message strangers out of the blue.
- Call a result an edge when it is within what luck can explain.
Contact
To reach us, message us on X. The Telegram channel only posts test results; it does not take messages.
We have no contact form and no email address on this site.