独立证据审查。 本文回答了两个不应混为一谈的不同问题。第一：公开记录是否独立验证了所宣传的 PHOENIX 管理交易服务 hamster-bot的表现？第二：与我们的研究相关的可复现 ZigZag, MRS および ZZBoba 策略想法是否经得起诚实测试？
这不是欺诈、操纵或犯罪行为的认定。私人商业 PHOENIX ポートフォリオは非公開であり、再現できた戦略ルールは代理指標にすぎません。代理テストの失敗は、その非公開ポートフォリオを反証も再現もしません。
PHOENIXハムスターボットのレビュー：提供されるものは何か？
公式 PHOENIXページ は、顧客自身の取引所またはブローカー口座での暗号資産の定量運用を説明しています。運営者のIPアドレスに紐付けられた取引可能なキーを要求し、出金許可は不要としています。ページは、月間平均リターン3%超、総リターン+131.0%、推奨入金額15,000ドル、四半期サービス料、15%～30%のパフォーマンス手数料、紹介顧客の手数料の5%～20%のパートナー報酬を宣伝しています。
顧客の口座に資金を保持することで、カストディリスクの一種は軽減されます。しかし、取引権限、レバレッジ、清算、執行、プラットフォーム、戦略、利益相反のリスクは排除されません。出金できないキーでも、損失を生むポジションを開くことはできます。そのため、当社はこのエクスポージャーを、接続前に独立した検証が必要なほど高いものと分類しており、読み取り専用の分析製品と同等とは見なしていません。
Does TradeLink verify PHOENIX returns?
The short answer is: it provides useful third-party telemetry, not a complete independent audit. At our evidence cutoff, the official page reported +2.48% for January through May 2026. The linked Binance TradeLink portfolio feed was approximately +2.56% over the same cutoff. That limited agreement matters: the reported period is not contradicted by that linked account.
It is still insufficient to reproduce the advertised composite. The official page says tracking moved to a TIGER broker account on 2 February 2026, but its monitoring table still labels the TIGER link “coming soon.” The linked portfolios allow selected start dates, include backfilled history from before their public creation, and the Binance view joins two keys at a later date. Published account weights, cash flows and a continuous equity reconciliation from the earlier accounts into the current account were not available.
TradeLink itself explains that its portfolios are calculated from raw exchange data including trades, deposits and withdrawals, and that exchange feeds can be incomplete or change without notice. Its documentation on compound portfolios and exchange limitations is useful and candid. It also explains why a portfolio link proves less than an audited, account-weighted track record.
| Public evidence can support | Public evidence does not establish |
|---|---|
| A linked account feed existed and reported exchange-derived activity. | That every commercial client received the displayed result. |
| The Binance feed was close to the advertised 2026 result through May. | The continuous +131% composite across FTX, Binance, Bybit and TIGER. |
| The operator publicly discloses fees, referral terms and requested trading access. | Strategy identity, account ownership, complete account selection or future returns. |
Phoenix ZigZag: the apparent edge that disappeared
Our Phoenix ZigZag work began with a result that looked exceptional. Across the research grid, 119 cells initially met a live-candidate gate. An execution audit then found intrabar look-ahead: the simulator could use the current bar's high to advance a trailing peak before checking the same bar's low. That ordering gives a backtest information a live order never had.
After the repair, 119 candidate cells fell to 1, and median performance fell from roughly +284% to +29.2%. A later 342-coin, four-hour pass still looked promising in isolation: 230 coins were positive, 283 beat hold, and one live candidate remained. But the median maximum drawdown was about 34%, and the family still had to survive costs and repeated-hypothesis control.
Costs reversed the headline. The same high-turnover portfolio, covering 4,930 trades, changed from +720.1% under a low 5-basis-point assumption to −599.3% under the 0.2%-per-side spot fee used for the relevant Gate scenario. Finally, a per-strategy walk-forward comparison with Holm–Bonferroni correction across 57 strategy hypotheses left 0 survivors. Even phoenix_zz_slow, whose raw probability looked strong, missed the corrected threshold and weakened again at higher fees.
The lesson is broader than this family: signal code can be causal while execution code still leaks the future. A backtest is only as honest as its fill ordering, fee tier, slippage model and count of ideas tried before the winner was selected.
MRS strategy review: a 78% win rate without proven edge
MRS is a mean-reversion system that moves limit orders around a moving average. Price reaching an offset opens a position; return toward the average closes it. It can naturally create many small wins and a high win rate while retaining rare, large or long-lived losses.
That is exactly what the tests found. In a 30-coin, 30-minute breadth pass, 29 of 30 out-of-sample results were positive and the median win rate was 70.7%. Yet median alpha versus passive directional exposure was −0.305%, and 0 of 30 passed all gates. Eighteen were below hold, ten could not beat mechanics-preserving random offsets, and two had too few trades.
The broader four-hour campaign is more decisive:
| MRS arm | Observed candidates | Expected by chance | 決定 |
|---|---|---|---|
| Written MRS2, 351 symbols | 2 / 351 | 17.35 | Below chance expectation |
| Optimized MRS, 351 symbols | 17 / 351 | 16.95 | Indistinguishable from chance |
The optimized arm's median win rate was 78.48%, but candidate count was almost exactly the null expectation. The concrete offset was not the demonstrated source of profit. Continuing to tune offsets inside this failed family would spend compute on selection noise, so the family is closed unless a materially new, preregistered hypothesis appears.
ZZBoba strategy review: the regime call is the strategy
ZZBoba is our internal name for a regime-conditioned directional ZigZag hypothesis; it is not represented here as a commercial PHOENIX product. In a bull regime it stays long without a stop until an opposite signal; in a bear regime it switches to a short configuration.
The attractive full-history example is SOL on one-day candles: buy-and-hold was approximately +3,517%, while the selected long ZigZag form was approximately +8,407%. But the analysis used the regime choice over the same history. In the wrong regime, documented examples lost roughly 80%–93%. No sealed out-of-sample classifier has yet shown that the system can choose bull versus bear prospectively.
Therefore the verdict is not “profitable” or “failed.” It is INSUFFICIENT EVIDENCE. The hypothesis may return only as a preregistered regime-classification test with a sealed holdout and a forward paper record.
Was the idea of trading or buying these strategies justified?
No, not as a shortcut to profitable automation. The evidence does not justify connecting a live account to the reviewed operator, buying these strategies, copying them, or funding another parameter search. It does justify preserving the research as a negative-control library and turning the work into a transparent strategy-audit product.
| レーン | 状態 | 予算 |
|---|---|---|
| Live execution, copying and validation capital | STOP | $0 |
| Vendor deposits, strategy or signal purchases | STOP | $0 |
| Affiliate promotion of reviewed entities | STOP | $0 |
| Read-only evidence capture and deterministic reproduction | GO | 既存のインフラのみ |
| Human-reviewed reports, corrections and strategy-audit validation | GO | $0増分インフラ |
What evidence would change the PHOENIX decision?
- A working public TIGER monitor and immutable raw exports for every linked account.
- A cash-flow- and account-weighted reconciliation of the +131% composite, including account transitions.
- Clear mark-to-market treatment of open positions, complete fees and funding, account-selection rules and ownership evidence.
- A documented operator response and resolved service/legal-entity scope.
- For any strategy claim: causal execution, realistic costs, passive and matched-null baselines, multiple-testing control, a sealed holdout and forward paper evidence.
それまでの間、アクションは継続する 現在の証拠から接続または資金提供を行わないでください。これは可逆的な証拠判断であり、非難ではありません。
編集方針、利益相反、および反論権
このレビューは有料掲載ではなく、レビュー対象の事業体へのアフィリエイトリンクも使用していません。運営者、著者、データ提供者は 日付入りの一次証拠、事実訂正、または回答を提出できます。重要な訂正は記録され、証拠が変更された場合、判定はアップグレード、限定、または撤回されることがあります。
基礎となるテスト基準については、 FinEdg手法 および関連するフィールドノート、 セブンのトレードアイデア、正直にテスト済みをお読みください。この資料は調査であり、金融アドバイスや取引の勧誘ではありません。