Skip to content

H-09: VWAP × vol ablation on 30-day real 1-min (diagnostic, not pass/fail) #10

Description

@Kevin-Power

Layer C (real 1-min, 30-day window already in data/tx-1min.json). Q8: VWAP and the 0.4–2.4× OR-median vol band are still glued together inside ALPHA-37. H-06 split the gap filter; this is the matching split for the two C-layer filters.

Not a pass/fail. n=30, all in the 2026 bull, no 2025 veto. Same framing as H-01-30d probe grid. Official Q8 pass/kill still needs real 1-min from 2024-08-26.

Skeleton (locked): probeMin=37, targetR=0, stopOrFrac=0.55 (not TORB's 0.7 — stop fraction stays out of this ablation), no regime / settlement / foreign / weekday filter. Cost model = 2026 MTX.

2×2, gaps OFF (isolate C-layer):

  1. none (TORB-37 skeleton at 0.55)
  2. vwap only
  3. vol only
  4. vwap+vol

Companion, not in the 2×2: ALPHA-37 as shipped (vwap+vol+gaps).

Real path vs synthetic path, each with its own OR-width history (same replay as scripts/h01-probegrid.ts). Seed 0 for the table; 20-seed synthetic distribution is optional companion and cannot upgrade this to a pass.

Do not: change DEFAULT_PARAMS, retune 0.8 / 0.55 / 37, add presets, treat 30-day PF as Q8 pass or kill.

Read-only diagnostic: report ΔPF of each cell vs none on the real path, plus skip-reason counts. Sign of ΔPF only tells us whether the filter is still worth waiting for the long real window.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions