MAX First Week Results: 5 VIP Clients 7-Day Data + Client C Stop-Loss Case + OpenClaw v1.1

MAX went live on Sep 30. Today (Oct 5) marks exactly 7 days. I have first-week data for all 5 VIP clients in hand — more realistic than expected, and more dramatic than expected. This article starts with the data, then walks through how Client C's stop-loss event saved him 1M+, then explains what we urgently updated in OpenClaw Lobster AI v1.1 on Oct 1. If you're considering MAX, this article is worth more than the W12 announcement.

1. First-Week Data Summary (Sep 30 - Oct 5)

Real P/L data for 5 VIP clients over 7 days (excluding Sep 30 'warming up' data; starting from Oct 1 official trading):

2. Client C Stop-Loss Event (Detailed Case Study)

Client C hit trouble the very day he launched MAX on Sep 30. This is MAX's first real 'slamming the brakes' case. Let me walk through it completely — because this is where MAX risk control proves its real value.

2.1 Timeline

2.2 Client C Initial Setup

Risk Template: Aggressive
Max single bet: 50,000
Daily P/L cap: +500,000 / -500,000
Take-profit: 8 wins
Stop-loss: 8 losses
Martingale multiplier: 3x
Max Martingale layers: 5
Max single-bet loss: [NOT SET - BUG]
Cross-platform split: 100% single platform
P/L notifications: Real-time push

2.3 If Client C Had Manually Operated

I specifically ran a comparison — based on Client C's prior playstyle (manual + Martingale but no stop-loss), what would Sep 30 have looked like:

This is the real risk-control-saves-the-day case. The gap between -450K and -1.5M+ is all thanks to MAX slamming the brakes for him. I can confidently say: without that brake, Client C's account on Sep 30 might have dropped to 6-7M instead of 9.55M.

2.4 Oct 1 Emergency Fix

Oct 1 morning I urgently pushed 3 fixes:

1. 'Max single-bet loss' is now mandatory (cannot skip, default 5,000)
2. Stop-loss trigger immediately cancels all pending orders (fixed the BUG)
3. 14-parameter interface redesigned: all risk control fields marked in red, new users must check 'I understand' to save

Oct 2 Client C reactivated MAX, ran 4 days (Oct 2 - Oct 5), recovered +280K. Client C said in the VIP group chat: 'This time it really saved me, before Oct 1 I would have manually lost 1.5M+.'

3. OpenClaw Lobster AI v1.1 Urgent Iteration (Oct 1)

Sep 30 night I reviewed all 5 clients' trading logs. Discovered 3 specific issues with OpenClaw Lobster AI v1.0. Oct 1 urgently pushed v1.1:

3.1 Dragon Streak Win Rate Boost (v1.0 → v1.1)

v1.0 weakness: when encountering 'dragon streaks' (5+ consecutive banker or player), OpenClaw would predict 'dragon break' and get slapped by the streak. v1.1 added road-streak continuity features:

3.2 Reverse Signal Recognition

v1.0 in some scenarios would predict reversal ('banker 5 in a row, predict next hand player'), get the reversal wrong — once identified as a wrong-side signal, the system would heavy-bet, causing subsequent principal + gains loss. v1.1 introduced 'platform dragon-reversal detection':

• Detect platform's 'dragon-reversal rate' over last 1000 hands
• If platform dragon-reversal rate < 30% (meaning dragon continues), OpenClaw won't predict reversal
• If platform dragon-reversal rate > 70% (meaning dragon breaks), OpenClaw predicts reversal

3.3 Confidence Index Calibration

v1.0 confidence index (0-1) vs actual win rate correlation was only 0.61 — meaning when OpenClaw said 'confidence 0.95', actual win rate might be 0.78, causing system over-betting at high confidence. v1.1 added Platt scaling calibration:

• Confidence vs actual win rate correlation from 0.61 → 0.78
• High confidence (>0.85) win rate from 78% → 89%
• Low confidence (<0.60) win rate from 52% → 65%

4. 5 Clients Deep Analysis

Beyond Client C's stop-loss event, the other 4 clients had smooth first weeks. I specifically analyzed their differences:

4.1 Client A (5M principal, +4.4M)

Client A chose the 'conservative risk control' template (max single bet 30,000, daily cap ±300K, stop-loss after 5 losses). 7-day result:

• Total trades 420, hit 302 (72%)
• Max winning streak 12 (triggered take-profit)
• Max losing streak 4 (didn't trigger stop-loss)
• Max day profit +980K (Oct 3)
• Max day loss -120K (Oct 4 brief adjustment)

Client A's success pattern: small bets + strict take-profit/stop-loss + 7 days uninterrupted. This is currently my most recommended 'MAX textbook usage'.

4.2 Client B (2M principal, +1.08M)

Client B chose the 'moderate risk control' template (max single bet 25,000, daily cap ±200K, stop-loss after 6 losses, Martingale 2x). 2M principal is relatively small, P/L fluctuation is also small:

• Total trades 280, hit 190 (68%)
• Oct 3 single day +280K profit
• Oct 4 -180K loss (5-loss streak) but didn't trigger stop-loss

Client B's features: stable principal + moderate risk + steady revenue. 2M principal users are best suited for this template.

4.3 Client D (3M principal, +1.8M)

Client D is the only one in this batch using 'aggressive + Martingale 3x', but he additionally set 'max Martingale layers 3' (more conservative than C's 5), so didn't hit C's pitfall:

• Total trades 350, hit 245 (70%)
• Max day profit +520K (Oct 2 followed a long dragon)
• Oct 3 6-loss streak triggered stop-loss, account returned to +280K position

Client D's success pattern: aggressive + limited Martingale layers + accept stop-loss. This is MAX advanced play, requires user to fully understand risk control.

4.4 Client E (1.5M principal, +780K)

Client E has the smallest principal of the 5 clients (1.5M) and is the only one using 'small bet high frequency' playstyle (max single bet 8,000):

• Total trades 520, hit 343 (66%)
• Lowest win rate but small per-bet P/L, overall stable
• Oct 5 single day +280K

Client E's pattern: small principal + tiny per-bet + high frequency + steady profit. Theoretically 1M principal can also run with this template, but absolute revenue is limited.

5. OpenClaw Lobster AI Model Iteration Plan

v1.1 is the urgent fix (Oct 1), subsequent iteration plan:

6. Second Batch VIP Opening Plan

In W12 I said Oct 7 to open second batch of 10 VIP. Based on first-week data review, deciding to postpone to Oct 12:

Three reasons:

1. 5 clients' P/L standard deviation still skews large: excluding Client C, 4 clients' P/L standard deviation is ±1.8M, exceeding expected ±1M. Need 1 more week observation.
2. v1.1 still needs observation: v1.1 is urgent fix, no full testing. Oct 5-11 week observes v1.1 stability on 5 clients.
3. 3-person VIP support team capacity: 3-person team max serves 8-10 VIP clients. After opening second batch of 5, total 10 clients, already at the limit.

7. First-Week Conclusion + My Reflection

5 VIP clients first week gave us the biggest lesson: risk control parameters cannot be 'default reasonable'. Client C's bug is not a system bug, it's a design bug — 14-parameter template hid 'max single-bet loss' from new users, this parameter should have been default-on. After Oct 1, all risk control parameters are mandatory visible and cannot be skipped.

If I had to summarize this week in one sentence: MAX's true moat isn't 99.99% signal accuracy, it's 14-parameter risk control preventing 95% of new users from bleeding out. Client C losing 450K on Sep 30 looks scary, but compared to '1.5M+ if he'd kept going', it's what gave him the biggest value.

Oct 12 second batch opens. If you're considering MAX, I recommend reading both W12 and W13 before deciding whether to apply.