Cross-campaign evidence synthesis. During June and July 2026, BibaMoney translated several indicators, bots and strategy ideas into reproducible tests. This review combines the evidence available through 30 July for FVB, MoonBot, PHOENIX, MRS, ZZBoba, NASAlgo and a final 57-hypothesis strategy campaign.
The purpose is not to declare a winner. These studies used different instruments, timeframes and validation contracts, so placing their raw returns in one leaderboard would be misleading. The useful question is narrower: which claims survived the test that was appropriate for them, and which next action is justified?
This is not a fraud, scam or broker-safety finding. A strategy can fail while its software works as documented; a broker can execute orders correctly while a strategy loses; and a third-party performance monitor can contain genuine exchange data without constituting an independent audit. Those are separate propositions.
The evidence map
| Research lane | What was tested | Result | Action |
|---|---|---|---|
| FVB composite | Regime, BTC alignment, macro and session filters | NOT READY | Complete the missing layer; preregister a new test |
| MoonBot replicas | Four 5-minute economic proxies and 324 configurations | NOT REPRODUCED as a live edge | One tiny paper candidate only |
| MoonBot copy cohort | Eight trader IDs frozen before forward observation | 2 / 8 holds, 6 / 8 fails | No copying; retain read-only observation |
| PHOENIX/ZigZag proxies | Execution repair, costs, walk-forward and family correction | 0 / 57 survivors | Close the tested family |
| MRS | Written and optimized mean-reversion offsets vs matched chance | NOT REPRODUCED | Do not tune the same hypothesis again |
| ZZBoba | Regime-conditioned directional ZigZag | INSUFFICIENT EVIDENCE | Test the regime classifier prospectively |
| NASAlgo WaveTrend | One fixed rule on 30 stocks and 30 crypto assets | NO PROVEN EDGE | No live promotion from this result |
What the optimizer was designed to prevent
An optimizer is very good at finding an attractive historical cell. That is not the same as discovering a repeatable trading advantage. Every extra symbol, timeframe, parameter and exit is another opportunity to select noise.
The campaigns therefore used progressively harder gates: causal signals and execution, next-bar or resting-order fills, both-side costs, an untouched out-of-sample segment, passive or mechanism-matched baselines, walk-forward folds, and multiple-testing control. The strictest final campaign evaluated 57 named hypotheses on four-hour data and applied Holm–Bonferroni correction across the family.
This distinction explains why an early “best result” can be real arithmetic and still be unusable evidence. Optimization answers what fitted this history? A proof protocol asks what remained after selection, costs and unseen data?
FVB: verified components, unverified edge
FVB is an indicator architecture rather than a turnkey bot. Its intended hierarchy combines market structure and regime, the traded asset's alignment with BTC, a DXY macro layer and session-aware routing. The Python port of the Z-score logic was checked against the visible Pine behavior and passed with caveats. That establishes implementation parity for a component; it does not establish profitability.
The strategy database contains 47 FVB composite runs, and all 47 are NOT READY. The macro layer was used in 0 / 47 because matching DXY data was unavailable. The original five-coin comparison found the composite did not improve the raw regime layer on average. Later rolling BTC runs sometimes reduced the loss relative to the ungated arm, but did not turn the latest candidate positive.
In the 30 July BTC snapshot, the ungated out-of-sample result was −52.72% and the gated result was −14.97%. The filter therefore improved a bad result by 37.75 percentage points, but still lost money, generated only 17 OOS trades and was marked underpowered. The honest verdict is NOT READY: useful architecture, incomplete test, no live edge claim.
MoonBot: software, replicas and copy statistics are different claims
MoonBot is capable trading software with tick-level behavior; our replicas are 5-minute economic approximations. They cannot reproduce millisecond fills or every vendor configuration. Among the four current proxies, three are net-negative. The optimized btc_shorts_fast arm produced about +$0.54 on a rolling $100-order research contract after a 324-configuration search. That is a locked paper candidate, not a live-ready edge.
The separate copy-trader study froze eight public trader IDs before observing them forward. After 35 days and 24,090 source records, 2 / 8 passed the daily-cluster hold gate and 6 / 8 failed. A displayed +1,425.11% sum across sources is not account ROI because the feed lacks a shared capital ledger, concurrent exposure, collateral and position sizing. The full methodology and limitations are in the MoonBot replica and copy-cohort review.
PHOENIX and the 57-strategy campaign: the candidate that did not survive correction
The PHOENIX ZigZag work produced the clearest example of why execution auditing comes before optimization. A look-ahead error in trailing execution allowed the simulator to update a peak with the current bar's high before evaluating that bar's low. Repairing the order collapsed 119 candidate cells to 1 and reduced median performance from roughly +284% to +29.2%.
The final campaign included 57 hypotheses spanning PHOENIX, Jesse, Donchian, EMA, MACD/RSI, WaveTrend, Supertrend, Bollinger, momentum, order-block and chart-pattern families. At 5 basis points per side, phoenix_zz_slow had raw p = 0.001, but adjusted p = 0.057. At 10 basis points per side, its raw p weakened to 0.063 and adjusted p became 1.0. No hypothesis survived at either cost level.
The commercial PHOENIX service is a separate claim. A linked Binance monitor was close to the operator's reported January–May 2026 result over that limited cutoff, but public evidence did not reconcile the advertised continuous +131% composite across accounts. Its verdict is therefore UNVERIFIED, not “false.” See the full PHOENIX, ZigZag, MRS and ZZBoba evidence review.
MRS: why a 78% win rate was not an edge
MRS places mean-reversion orders around a moving average. This can create many small wins while preserving rare losses that dominate the account. In the broad 351-symbol four-hour campaign, written MRS2 produced 2 candidates where 17.35 were expected by chance. The optimized arm produced 17 candidates where 16.95 were expected by chance.
The optimized median win rate was 78.48%, but its selection count was almost exactly the null expectation. On one APT candidate, optimized OOS performance was +6.67%, while mechanics-preserving random-offset twins averaged +6.08% and beat it 44% of the time. The specific offset was not the demonstrated source of profit. Further tuning of the same family would be selection work, not new evidence.
ZZBoba: an impressive history that depends on knowing the regime
ZZBoba is our internal name for a directional ZigZag hypothesis that uses one configuration in a bull regime and another in a bear regime. The attractive SOL full-history example showed approximately +8,407% versus about +3,517% for hold. But the regime decision was made with knowledge of the same history. In the wrong regime, documented examples lost roughly 80%–93%.
The real strategy is therefore not the ZigZag line; it is the ability to classify the regime before the future occurs. No sealed out-of-sample classifier has demonstrated that ability. The correct label is INSUFFICIENT EVIDENCE, not profitable and not disproved.
NASAlgo: breadth disappeared out of sample
The NASAlgo WaveTrend study deliberately avoided per-asset tuning. One fixed daily rule was applied unchanged to 30 US stocks and 30 crypto assets: enter at the next bar open, exit on the opposite signal or after 30 bars, use a 20% protective stop, 1× leverage and 5 basis points per side. The first 70% of each series was development data and the final 30% was held out.
Across 60 assets, 25 / 60 were net-positive on the full sample, 16 / 60 were positive in OOS, and only 10 / 60 were positive in both segments. Average net performance was −4.4%. NEAR and HBAR were interesting asset-level observations, but a handful of positives inside a negative-breadth universe is a research queue, not a family-level edge.
What this evidence says about brokers and exchanges
The strategy and the venue must be audited separately. A broker, exchange or monitor cannot turn an unproven strategy into a proven one, and a failed backtest does not by itself show that a venue is unsafe.
| Venue evidence in these studies | What it supports | What it does not support |
|---|---|---|
| Binance and Bybit account feeds shown through TradeLink | Useful exchange-derived telemetry for the linked accounts and selected periods | A complete audited PHOENIX composite or equal client outcomes |
| Gate execution-cost scenarios | A test of whether the hypothesis survives the declared fee assumption | A general verdict on Gate or a claim about every user's actual fee tier |
| Missing public TIGER monitor at the cutoff | An evidence gap in the advertised PHOENIX account continuity | A negative safety or regulatory conclusion about TIGER |
Cost sensitivity was economically decisive. The same high-turnover Phoenix portfolio moved from +720.1% under a 5-basis-point assumption to −599.3% under the 0.2%-per-side spot-cost scenario used in the Gate study. This does not mean one venue “caused” the strategy to fail. It means the apparent edge was smaller than plausible execution friction and therefore was not portable enough for live promotion.
Performance monitoring also has a boundary. TradeLink telemetry can be valuable without being an auditor: selected start dates, backfilled history, joined keys, deposits, withdrawals and incomplete exchange feeds can all change what a chart means. A commercial return claim needs account ownership, immutable trade and balance exports, cash-flow reconciliation, complete fees and funding, and an explanation of every account transition.
Finally, “withdrawals disabled” is not the same as “capital safe.” A trading-enabled API key can still open losing or leveraged positions. The appropriate sequence is read-only evidence first; then, only after a strategy passes its own proof gate, a separate least-privilege review covering subaccounts, IP allowlists, withdrawal permissions, order limits and incident response. Current evidence does not authorize that second step for any family reviewed here.
What should continue—and what should stop
| Lane | Decision | Reason |
|---|---|---|
| Live bots, copying, deposits and paid signals | STOP · $0 | No family has passed the live proof gate |
| Affiliate promotion of reviewed vendors or venues | STOP · $0 | Evidence is insufficient for an executable endorsement |
| More parameter search inside failed Phoenix/MRS families | STOP | The next best cell is not an independent hypothesis |
| Locked MoonBot BTC-short forward paper test | GO · paper only | Small held-out improvement deserves observation, not capital |
| FVB after DXY completion and preregistration | GO · new evidence contract | Current test is incomplete rather than a clean family rejection |
| ZZBoba regime classifier and NASAlgo asset candidates | GO · sealed paper only | Prospective classification and breadth still need proof |
The general conclusion
The optimizer did its job precisely when it refused to promote a strategy. It found implementation defects, exposed fee sensitivity, separated high win rate from alpha, measured how often optimization merely matched chance, and prevented selected historical winners from becoming live claims.
The combined evidence supports a strategy-audit product, not an automated trading product. Future candidates should enter through a preregistered specification, causal simulator, realistic cost model, sealed OOS test, matched null, family-wise correction and forward paper ledger. A broker or exchange review is a second passport, never a substitute for the strategy passport.
Editorial policy and right of reply
This synthesis is not paid placement and contains no affiliate link to a reviewed bot, broker, exchange or monitor. Operators, authors and venues may submit dated primary evidence, a factual correction or a response. Material corrections will be logged and a verdict may be upgraded, narrowed or retracted when evidence changes.
Open the research optimizer, read the FinEdg testing methodology, or inspect the detailed PHOENIX-family and MoonBot case reviews. This material is research, not financial advice or an invitation to trade.