Post-market dummy refinement: 2026-08-14

[github-actions]

Goal

Assign this issue to Copilot as a daily post-market refinement task.

Review today's dummy-trading behavior, identify only evidence-backed improvements, and propose a small safe change if the data justifies it. Do not add strategy parameters randomly.

Latest Paper P&L

• Paper date: 2026-08-14
• Portfolio value: ₹500000.0
• Today P&L: ₹0.0 (0.0%)
• Total P&L: ₹0.0 (0.0%)
• Alpha vs NIFTY: -0.526%
• Positions: 0
• Trades in latest paper run: 0

Incubation Gate

• Decision: not_ready
• Days: 20/20
• Blocking checks: 10/11
• Failed checks: positive_dummy_alpha

Live Market Snapshot

• Live snapshot IST: 2026-08-14T15:26:31.205930+05:30
• Market open flag: true
• Live day P&L: ₹0.0 (0.0%)
• Live actions: 0
• Stops: 0

Latest Dummy Transactions

• No EOD paper trades in latest run

Daily Analysis Context

• Trading date: 2026-08-14
• Signal counts: BUY 45, HOLD 5, SELL 0
• Market regime: NIFTY trend filter is risk-off

Backtest Lab Context

• momentum_only_weekly: return 8.96%, alpha 10.02%, maxDD -7.97%
• momentum_weekly_churn_control__news_on: return 8.96%, alpha 10.02%, maxDD -7.97%
• momentum_weekly_churn_control__news_off: return 8.96%, alpha 10.02%, maxDD -7.97%

Copilot Instructions

• Inspect the dashboard/data outputs and current strategy code.
• Look for repeated or material failure patterns: stale data, bad churn, weak ranking, poor risk behavior, oversized drawdown, or confusing UI.
• If there is no strong evidence for a code change, say so and leave the system unchanged.
• If a change is justified, keep it small, testable, and paper-only.
• Preserve the one-month incubation rule. Do not recommend real-money trading.
• Include validation steps and expected effect on the readiness gate.

[claude]

Daily post-market refinement — 2026-08-14 — no change

Reviewed today's snapshot (data branch: daily/latest.json, paper/latest.json, paper/readiness.json, backtest/latest.json, watchlist/latest.json) against the strategy code (src/market_regime.py, src/paper_trader.py, src/incubation_report.py). No code change is warranted. Today closes out the full one-month incubation window as an exact continuation of the pattern already investigated and confirmed on 2026-07-21 (#23) and every daily-refinement issue since (#26–#43, 20 consecutive trading days).

Evidence

1. Incubation has now run its full 20/20 days with zero trades, entirely due to a validated regime guard, not a defect.
paper/readiness.json → min_trading_days: 20/20 — now passing (this was the last mechanical blocker; it has cleared). paper/latest.json history shows every single day from inception (2026-07-20) through today (2026-08-14) recorded n_trades: 0, n_positions: 0, risk_block_new_buys: true. Today's daily/latest.json regime block confirms why: NIFTY price 24366.0 vs SMA50 24084.1 vs SMA200 24735.75 — SMA50 < SMA200, so market_regime.risk_on = False, reason "NIFTY trend filter is risk-off". src/paper_trader.py's USE_MARKET_REGIME_GUARD blocks all new BUYs whenever risk_on is false, regardless of the 45 BUY-ranked signals in today's signal_counts (45 BUY / 5 HOLD / 0 SELL) — those are display-only candidate rankings, not order intents (verified in #23). With 0 existing positions there is also nothing to SELL, so 0 trades is the only correct outcome.

2. The one remaining failing check is mechanical cash-drag, not a strategy defect.
paper/readiness.json: decision: not_ready, passed_blocking: 10/11. The single failing blocking check is positive_dummy_alpha: alpha_pct: -0.526 — the direct arithmetic consequence of sitting 100% in cash (total_pnl_pct: 0.0) while NIFTY drifted +0.526% over the 20-day window. It is not a signal, ranking, or execution defect — recomputed totals match stored totals exactly (totals_recomputed_consistent: max mismatch 0.0 pct-pts). All other 10 blocking checks pass, including backtest_oos_positive (walk-forward mean validation alpha +3.96%, 2/3 folds positive) and backtest_supports_profile.

3. Backtest lab continues to show the regime guard is net positive, not a drag to fix.
backtest/latest.json (strategy_lab, generated 2026-08-09): momentum_only_weekly (regime guard on, matches the active profile momentum_weekly_churn_control) — alpha +10.02%, maxDD -7.97% — versus no_regime_filter — alpha -5.21%, maxDD -16.92%. Forcing trades now, one day after the mechanical min_trading_days gate cleared, during a confirmed and still-active death-cross regime would trade a backtest-validated defensive filter for materially worse historical alpha and ~2.4x deeper drawdowns. That would be gaming the incubation gate, not fixing a defect — explicitly against the refinement rules.

4. News ablation variants remain identical by design (already addressed in this repo), not new evidence of a bug.
momentum_weekly_churn_control__news_on and __news_off both still report 10.02% alpha / -7.97% maxDD — expected, since news_days_used is still 0 for the active variant (the dated news archive, added 2026-07-19 in commits d2b61bd–f887959, has not yet accumulated enough history to affect the backtest replay window). This is documented, self-resolving behavior, not something to patch today.

Why no change

No new signal, execution, or risk-behavior anomaly appears in today's data beyond what has already been verified and left unchanged for 20 consecutive daily-refinement issues. The strategy has behaved identically and deterministically every day of its incubation: correctly declining to trade into a persistent risk-off regime, exactly as its own backtest evidence recommends. The only change a "fix" could make today — loosening or removing the regime guard to force a positive-alpha entry — is precisely the change the backtest data argues against, and would violate the "evidence-backed only" / no-gaming-the-gate rule.

Validation

python -m unittest discover -s tests -p "test_*.py" → 82 tests, 1 pre-existing error (test_news_archive fails to import due to ModuleNotFoundError: No module named 'requests' in this sandbox — requests>=2.31 is declared in requirements.txt but not installed here; confirmed this is a sandbox dependency gap, not a code regression, and matches the same pre-existing issue noted in PR #24 for issue #23). All other tests pass.

Expected effect on readiness gate: none from this issue. positive_dummy_alpha will resolve once the market regime flips risk-on (SMA50 crosses back above SMA200) and the strategy re-enters positions, per the guard's documented, backtest-verified behavior — at which point decision should move to eligible_for_review (human review only, never automatic real-money trading).

Closing as no-change per the daily-refinement protocol.

🤖 Daily refinement — no code change today. Day 20/20 of incubation completed with 0 trades, entirely explained by the backtest-validated risk-off regime guard blocking all buys every day since inception; the sole remaining failing check (positive_dummy_alpha) is mechanical cash-drag from a NIFTY drift of +0.526% while in cash, not a strategy defect — commented with evidence and closed issue #44.

[github-actions]

🤖 Daily refinement — no code change today. Day 20/20 incubation completed with 0 trades, fully explained by the backtest-validated risk-off regime guard; the only failing check (positive_dummy_alpha, -0.526%) is mechanical cash-drag, not a defect, so no code change was made — issue #44 commented and closed.