# A3.7: audit of Virality V7, hold or sell at 5x (6 Oct 2026, Bangkok)

## Read this first
- **Verdict: SHIP TO PAPER ONLY. Not live.** On my own rebuild the signal pays a lot in the second half, but it has switched exactly once (27 Sep). One switch can't be told apart from luck.
- **My rebuild** (v3 rule, stop 0.3x, range-aware depth; at 5x, hold on a 50% trail if 30% or more of the last 4 days' resolved 5x trades paid more held, else sell):
  | | Sell at 5x | V7 | $/day sell vs V7 |
  |---|---|---|---|
  | $500, first half | 2.22x | 2.17x | $1,176 vs $1,128 |
  | $500, second half | 2.05x | **2.76x** | $1,812 vs **$3,026** |
  | $1k, first half | 2.16x | 2.11x | $2,235 vs $2,142 |
  | $1k, second half | 2.00x | **2.67x** | $3,441 vs **$5,757** |
  | Wider gate (3 h, army 0.015), $500, A / B | 2.06x / 2.03x | 1.97x / **3.02x** | $1,406 / $1,942 vs $1,278 / **$3,810** |
- **One hold is still open:** 0xc38c332012a9116dcade0c7f43b5d3c71799a9f8 was at 12.87x when my data ended (6 Oct 13:41). Without it, the second half is 2.50x instead of 2.76x. Still well above selling.
- **Where my result differs from V7:** in the first half the signal holds once, on 0xc261532bc770c2b4ea10f97f72328aa0fc51b0d0 (27 Sep 06:23). Two of the six earlier resolved 5x trades paid more held: 0xead015725c6cc0a38cdaea04cb7933210198b830 (6.26x) and 0xe99c8437b0b29325a4080b8c837ff4e5a6f9fc69 (9.14x). That hold cost −1.34x. V7 says the first half is unchanged; mine is −0.05x.

## The regime-flip risk (what the Supervisor asked)
- **The only switch so far** came on 27 Sep at 06:23, from sell to hold. The first 3 holds after it returned −1.34x, +0.61x and −2.11x against selling at 5x: **−$1,418 at $500**, before the wins came.
- **Worst case if the market flips back:**
  - Every one of the 23 held 5x trades exited on the trail, none on a pull. The worst held result was 2.18x (0x60042b65a0c0a90ba75d17879850b6f99d9422eb).
  - 3 wrong holds at that level cost about **−$3,900 at $500**.
  - If a crew pulls right after the 5x (never seen in these 23, but possible), each wrong hold loses the whole 4.8x: 3 of them is **−$7,200 at $500**.
- **What the signal really does:** it holds in the second half and sells in the first.
  - "Always hold" gives about 1.78x in the first half and 2.76x in the second.
  - "Always sell" gives 2.22x and 2.05x.
  - V7 is right if waves last days. It is wrong for 2 to 3 trades every time the wave turns.

## Paper book suggestion (for the Coder)
- `w2_v3_regime`: same entries as the main book, stop 0.3x. At 5x apply the hold-share rule: cut 0.3, 4-day look-back, trades count only after their held version has fully exited.
- Log on every 5x: the share, n, the hold decision, and the result both ways. Go live only after it sees at least one switch back to "sell" and the cost of the switch stays under about 3 holds.

## First principles score
- 7/10. The decision uses only trades whose held result was final before the 5x moment, and exits are priced at real depth.
- What's missing is a second regime switch. That is the point of the paper book.

Version 1, 6 Oct 2026 15:52 Bangkok, Analyst 3. Latest: http://159.223.88.168/hx-c9ois/sw/sw-a3-7-v7-audit.md