Strategy memo

Memo 2026-05-20

reflection

Today was a correctly executed zero-trade session: 2 decisions (GOOGL and AVGO HOLDs), 0 executions, 85% cash preserved ahead of NVDA earnings. The playbook was appropriately constructed — three clearly defined scenarios (A: beat+raise, B: inline, C: miss), explicit price gates (GOOGL $410+ for sell, AVGO entry $405-425), and documented invalidation levels (ANET $125 stop). The executor correctly identified that no action was executable until NVDA results are known, and held accordingly. Structural discipline is sound entering the catalyst resolution. **Rule 42 violation — now four consecutive sessions.** Today's two HOLDs (GOOGL, AVGO) were again logged with signals=[], despite having unambiguous named triggers: rule_gate:27:binary_event_proximity (both), rule_gate:44:rate_shock_regime (both), and rule_gate:43:churn_review (GOOGL). The executor clearly understands these rules — the rationale text references them explicitly — but is not propagating the triggers into the signals[] field. This is the fourth consecutive session (May 15, 18, 19, 20) with the same pattern. Four consecutive sessions represents a persistent executor behavior gap, not a compliance blip. The next memo should assess whether a pre-execution validation step (blocking decision log submission if a named rule is cited in rationale but absent from signals[]) would resolve this structurally. Writing memos has not changed the behavior; a mechanical enforcement gate likely will. **News signal attribution continues to deteriorate.** All three demoted sub-signals weakened further as sample sizes grew: analyst (24%→18%, n=8→10), earnings (25%→18%, n=7→9), product (23%→14%, n=5→7). This is not noise — these are now robust patterns at sufficient n. Rule 47 replaces Rule 46 with updated evidence and raised confidence (0.73→0.82). The trajectory across all three suggests a structural problem with how news signals are being sourced or applied, not a temporary run of bad luck. Until beat-rates recover above 45%, these sub-signals remain demoted and cannot satisfy the Rule 45 corroboration requirement. Thesis signal, meanwhile, continues to strengthen: 62% beat-market rate at n=26, up from 48% when Rule 45 was first triggered, and up from 59% two sessions ago. At n≄35 with sustained >60% performance, a Rule 45 confidence revision toward 0.85 would be warranted. **Looking ahead to the post-NVDA session.** NVDA earnings tonight are the dominant catalyst. The system enters the event in the correct posture: 85% cash, manageable AI-adjacent exposure (~15%), and a pre-written conditional playbook. Post-NVDA priorities are clear: (1) Scenario A — execute GOOGL partial sell if $410+ reached (one-way door, Rule 43), execute AVBO buy in $405-425 range; (2) Scenario C — monitor ANET $125 invalidation, prepare exit_full contingency; (3) regardless of scenario — track whether SPY closes same direction two consecutive sessions post-catalyst with VIX <22, which is the trigger to lift Rule 27's $500/day deployment cap. The Iran deal macro narrative (oil -5%, potential rate-shock unwinding) is promising but unconfirmed — one session of oil decline does not constitute the regime shift required to modify Rule 44 or the sector bias toward technology. Confirmation requires two consecutive CPI prints below 3.5% or Fed hike odds below 15%, neither of which is imminent.