Skip to content

Phase 6A — Post-Run Evaluation Template

Scoring sheet for momentum-aware one-sided participation filter runs. Use this template for every Phase 6A session until Phase 6A is formally closed.

Run identity

  • Session ID:
  • Duration:
  • Config:
  • bid_offset_bps = 10
  • ask_offset_bps = 14
  • drift_pct = 0.65
  • max_skew_bps = 10
  • momentum_filter_enabled = true
  • momentum_lookback_ticks = 3
  • momentum_threshold_bps = 4.0

Hard startup gate

Must confirm from startup log: - momentum_filter_enabled=True - momentum_lookback_ticks=3 - momentum_threshold_bps=4.0

If any missing or wrong: Run invalid for Phase 6A evaluation.

1. Core outcome block

  • Fills total:
  • Buy fills:
  • Sell fills:
  • VW realized spread:
  • Avg realized spread:
  • Toxic fills / toxicity %:
  • Ending inventory %:
  • Drift range:
  • XRP session %:
  • Portfolio end value:

2. Momentum filter activity

  • Total BUY suppressions:
  • Total SELL suppressions:
  • First trigger timestamp:
  • Last trigger timestamp:
  • Total ticks with suppression active (if available):

Activity classification: - Dormant — 0 total suppressions. No meaningful momentum regime. Valid for baseline preservation only. - Moderate / healthy — Suppressions in bursts around directional moves. Bounded counters. Filter active and selective. - Overfiring — Frequent suppressions across noisy tape. Threshold too sensitive or signal too noisy.

3. Side-correctness check

For each suppression cluster, record: timestamp, momentum_bps sign, suppressed side, subsequent price path.

Pass rule: - positive momentum → SELL suppressed - negative momentum → BUY suppressed

Fail: any inversion of sign/side logic. Single ambiguous case OK, repeated = Phase 6A fails immediately.

4. Calm-regime baseline preservation

Compare against pre-6A baseline sessions.

Expected calm-regime profile: - VW spread roughly positive - fills not collapsed - toxicity near zero

Outcome Criteria
Pass VW positive, fills in plausible range, no starvation, toxicity ~0
Warning VW still positive but fills materially collapse; suppressions active in calm tape
Fail Previously healthy session now broken; VW materially worse in calm; filter fires without momentum

5. Anchor regime classification

Classify anchor over session: - Negative anchor — anchor meaningfully below CLOB mid → SELL structurally advantaged - Neutral anchor — near zero, not directionally dominant → cleanest read of intrinsic filter value - Positive anchor — anchor meaningfully above CLOB mid → BUY structurally advantaged

Record: - anchor mean: - anchor range: - anchor bias direction:

Interpretation rule: do not over-credit filter for strength on the anchor-advantaged side.

6. Interaction and conversion block

  • Buy near-touch (session):
  • Sell near-touch (session):
  • Buy conversion %:
  • Sell conversion %:
Signal Interpretation
Healthy Conversion plausible on at least one side; no complete collapse
Warning One-sided conversion near zero from suppression without momentum justification
Useful 6A signal Suppressed side shows reduced adverse interaction while opposite side still converts

7. Stress-regime check

  • Max drift reached:
  • abs(drift) >= 10%: yes/no
  • Duration at/above 10%:
  • Cap engaged during suppression windows: yes/no

Interpretation: - No stress (drift <10%) → Phase 6A evaluated for calm-regime preservation only, not full stress benefit. - Stress reached → true test of momentum filter at the boundary.

8. Pass / Fail decision tree

Phase 6A PASS — all true: - ✓ startup gate passed - ✓ side-correctness passed - ✓ no evidence of overfiring - ✓ calm baseline preserved - ✓ toxicity stays near zero - ✓ either stress behavior improved OR calm regime intact while filter stays selective

Phase 6A PARTIAL PASS: - logic correct, no breakage - but no strong momentum regime occurred OR anchor regime makes attribution difficult - safe to keep testing, benefit not yet isolated

Phase 6A FAIL — any of: - side suppression wrong - filter overfires in calm tape - fill activity collapses without justification - VW degrades materially in otherwise healthy regime - implementation not actually active in startup logs

9. Trigger count thresholds (evaluation, not production)

Range Interpretation
0–10 per side over 2h Low / selective / calm regime
~10–40 per side during directional periods Healthy active range if clustered
50+ both sides in calm-looking run Overfiring concern — investigate
Continuous alternating suppression Too sensitive

Soft thresholds. Not hard failure lines.

10. Output format for close

Send close in this structure: 1. Session results 2. Startup gate confirmation 3. Momentum filter activity 4. Side-correctness assessment 5. Anchor regime 6. Drift / stress status 7. Conversion and interaction 8. Verdict: PASS / PARTIAL PASS / FAIL 9. Next recommendation

Atlas scoring priority (what matters most)

  1. Side-correctness
  2. Baseline not broken
  3. No overfiring
  4. Stress improvement if stress occurred
  5. Anchor-adjusted interpretation

For the first Phase 6A run, we need to know:

  • Did the filter activate correctly?
  • Did it stay out of the way when not needed?
  • Did it avoid harming the good baseline?

If yes — Phase 6A survives first contact.

Template provided by Atlas, 2026-04-15. Use for all Phase 6A runs.