gitadmin/foxhunt
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

plan(dqn-v2): mark Task 6c LANDED

Browse Source
This commit is contained in:
jgrusewski
2026-04-25 01:11:04 +02:00
parent 0bbe97ed85
commit 3af2c79def
1 changed files with 4 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
docs/superpowers/plans/2026-04-24-dqn-v2-plan-3-behavioural.md
View File

@@ -1000,6 +1000,10 @@ Fingerprint shifts to `[79,80]`; `ISV_TOTAL_DIM` grows from 78 (post-Task-4) to
```
- Reset `PS_REGIME_SHIFT_BAR = 0` at every trade-entry + fold-boundary reset (same list as MIN_PNL).
### Task 6c: Conviction consistency bonus — ✅ LANDED `0bbe97ed8`
> **Actual outcome.** PS slots 41/42 (`PS_PRE_ENTRY_CONVICTION_EMA`, `PS_PRE_ENTRY_CONVICTION_VAR_EMA`) added; `PS_STRIDE` grew 41 → 43 via the same 6 hardcoded-stride migration pattern. Producer updates Welford-EMA of `conviction_core` each Flat bar (α=0.05). Consumer at Flat→Positioned entry fires iff `stddev/mean < 0.2` (the single permitted tuned constant), adds `shaping × vol_proxy × stability × conviction_core` to reward + rc[5] — max ≈ 0.01, mirroring B.2 novelty-bonus shape. Single unbounded multiplicand is `vol_proxy` per the one-unbounded-signal pearl. Smoke fold-2 best Sharpe 92.40 at epoch 3; HEALTH_DIAG `reward_split bonus=17.95`. 25-pt swing below T6b's 117 is at the edge of noise — flag for L40S validation watch.
### Task 6c: Conviction consistency bonus (stable conviction before entry)
**Files:**