# QSR trailing-arm sweep — account `paper-main`

Generated 2026-08-27T19:43:56.181Z · 64 anchor+runner pair(s), 44 single-leg buy event(s)

Current live setting: `QSR_V1.trailActivationPct` = **5%**, trail width `QSR_V1.trailPct` = **3%** (fixed across this sweep — only the activation threshold varies). Anchor's own `hardSellPct` = **5.5%**, also fixed.

**Research only** — re-simulated from real daily bars, stop-before-target pessimism, no portfolio constraints. See this script's header for the full methodology and limitations. Never merge into the journal.

## Runner leg only — the actual question

This is what trailActivationPct actually acts on. Read **%/day**, not %/trade — a later activation can only pay off later, never sooner.

| Activation % | Runners | Win % | Exp %/trade | Avg days held | **Exp %/day** | Trail/TP hit | Stopped | Timed out |
|---|---|---|---|---|---|---|---|---|
| 3% | 64 | 68.8% | -0.06% | 5.5 | **-0.010%** | 42 | 19 | 3 |
| 4% | 64 | 68.8% | +0.21% | 6.0 | **+0.035%** | 41 | 20 | 3 |
| 5% ← current | 64 | 67.2% | +0.68% | 6.7 | **+0.102%** | 39 | 21 | 4 |
| 6% | 64 | 62.5% | +0.82% | 7.8 | **+0.104%** | 34 | 21 | 9 |
| 7% | 64 | 60.9% | +0.78% | 8.3 | **+0.094%** | 27 | 21 | 16 |
| 8% | 64 | 59.4% | +0.73% | 8.9 | **+0.082%** | 23 | 21 | 20 |

- Best by **%/day**: **6%** (0.104%/day, 0.82%/trade over 7.8 days)
- Current 5%: 0.102%/day (0.68%/trade over 6.7 days)

## Runner leg, symbol-day clustered (the honest significance view)

64 runner leg(s) collapse to **35 independent symbol-day(s)**. Means below are per symbol-day.

| Activation % | Symbol-days | Mean %/trade | 95% CI | Differs from zero? |
|---|---|---|---|---|
| 3% | 35 | -0.77% | -2.31% … +0.78% | **no** |
| 4% | 35 | -0.77% | -2.41% … +0.88% | **no** |
| 5% ← current | 35 | -0.46% | -2.36% … +1.45% | **no** |
| 6% | 35 | -0.47% | -2.46% … +1.52% | **no** |
| 7% | 35 | -0.55% | -2.55% … +1.45% | **no** |
| 8% | 35 | -0.74% | -2.75% … +1.26% | **no** |

## Adaptive (widening) trail vs the current flat 3% trail

dan's question, 2026-08-27: tighten the trail right at the arm point (protect more of a move that reverses immediately), only widen it once the move has proven itself further. Schedule tested: arm at 5% (unchanged), trail **2%** while peak gain is 5-9%, widen to **3%** once peak gain reaches 9%+.

| Design | Runners | Win % | Exp %/trade | Avg days | **Exp %/day** | Hit trail | Stopped | Timed out |
|---|---|---|---|---|---|---|---|---|
| Current: flat 3% trail | 64 | 67.2% | +0.68% | 6.7 | **+0.102%** | 39 | 21 | 4 |
| Adaptive: 2% → 3% at +9% | 64 | 67.2% | +1.17% | 6.0 | **+0.195%** | 42 | 21 | 1 |

Symbol-day clustered (the honest significance view):

| Design | Symbol-days | Mean %/trade | 95% CI | Differs from zero? |
|---|---|---|---|---|
| Current: flat 3% | 35 | -0.46% | -2.36% … +1.45% | **no** |
| Adaptive 2%→3% | 35 | -0.01% | -2.02% … +1.99% | **no** |

**Verdict:** adaptive beats the current flat trail by +0.093%/day (row-level). Read the CIs above before acting on this — see the symbol-day view for whether either design's mean actually clears zero.

## Whole buy event (anchor + runner combined, qty-weighted)

What the ENTIRE buy decision returned, including single-leg buy events (unaffected by this sweep, included for completeness). Days = the longer of the two legs' exit days, since capital isn't freed until both close.

| Activation % | Buy events | Win % | Exp %/trade | Avg days | **Exp %/day** |
|---|---|---|---|---|---|
| 3% | 108 | 75.9% | +2.07% | 5.8 | +0.357% |
| 4% | 108 | 75.9% | +2.18% | 5.9 | +0.369% |
| 5% ← current | 108 | 75.9% | +2.35% | 5.9 | +0.398% |
| 6% | 108 | 75.0% | +2.41% | 6.5 | +0.372% |
| 7% | 108 | 75.0% | +2.40% | 6.8 | +0.354% |
| 8% | 108 | 74.1% | +2.38% | 7.1 | +0.335% |