Question: Cross-thematic-basket RSI-breadth standing-watch trigger refinement Verdict: refined: distinct-theme breadth trigger 3+ themes RSI over 75

Refined trigger — cross-thematic-basket RSI breadth (distinct-theme)

Price truth: all RSI/trend from research/market-engine/data/summaries/{etf-ideas,cybersec,cloud-etfs,memory}.json, as-of 2026-07-01 close (fresh precompute, mtime 2026-07-01 13:37).

What this refines. The 2026-05-26 DRAM monitor proposed a memory-cohort-only peak-cycle up-trigger (its "T3": 3+ memory names RSI > 75). The 2026-05-26 theme-audit observed the actual cross-cohort extremes that day (HACK 80, ARM 77) sat outside the memory cohort, and proposed generalizing to "3+ thematic ETFs simultaneously > RSI 75." This note locks that construct — with one correction that changes the count.

The correction: count distinct THEMES, not distinct ETFs

Cybersecurity alone is tracked by three ETFs (HACK, BUG, CIBR). Under a raw ETF count, a single cybersec blow-off (all three > 75) would trip "3+ thematic ETFs" by itself — that is a one-theme froth, exactly what a cross-thematic breadth tell is meant to exclude. So the trigger takes one representative per theme (the max RSI among that theme's clean ETFs); a single theme can contribute at most 1 to the count.

Definition (locked)

Fires when 3+ DISTINCT thematic baskets have their representative ETF > RSI 75 on the same close. Universe = clean thematic-sector ETFs only. Exclude: leveraged/inverse wrappers (SOXL/LABU/TQQQ/…), broad-index (QQQ/VOO/VTI/IWM/SMH-is-borderline), income/bond (JEPQ/QYLD/XYLD/HYG/LQD), and single names — none of those are thematic-froth tells. This is a breadth signal (how many themes are simultaneously extreme), not a magnitude signal.

Current reading (2026-07-01) — does NOT fire (2 of 3)

Representative RSI per distinct theme (rep = max in-theme):

Distinct themes > 75 = 2 (cybersecurity via BUG, biotech via XBI); airlines (JETS 74.8) at the doorstep. Trigger dormant — 2 of 3 needed. (Note the distinct-theme rule is load-bearing here: raw ETF count would also read 2 — BUG + XBI — because LABU is a leveraged echo of the same biotech theme, so both framings agree today, but the distinct-theme rule is what keeps a future cybersec-only blow-off from false-firing.)

Rotation, not broadening — why the dormant read is the right read

• 2026-05-22 extremes: cybersec (HACK 80) + AI-infra semis single-name (ARM 77) — 2 themes hot.
• 2026-07-01 extremes: cybersec has cooled at the ETF level (HACK 80 → 74.4; BUG 75.0 only at threshold), even as its single names stay extreme (PANW 81.9, BB 81.3); biotech took the baton (XBI 75.3, plus the 3x LABU 75.6 echo); airlines (JETS 74.8) is approaching. Still ~2 distinct themes at the extreme, with leadership rotating.

Froth is rotating theme-to-theme (cybersec → biotech), not broadening across themes at once. That distinction is the whole point of the breadth trigger: rotating single-theme extremes = mid-cycle churn (money hunting the next hot basket); simultaneous multi-theme extremes = a broad blow-off = the peak-cycle tell. The trigger correctly stays dormant through the rotation and only lights up if a third distinct theme joins while cybersec/biotech hold.

Disposition & cross-links

• Supersedes DRAM-monitor T3 as the cross-cohort generalization. The memory-only version stays as the within-memory sub-check for DRAM status-3 (early-mid July, TASKS-RESEARCH.md line 239) — today's storage-led memory heat (WDC/STX, per status-2) is a within-basket migration T3 catches; this cross-thematic trigger sits one level up and reads memory as a single (currently cold, DRAM 51) theme.
• Perspective ties: the biotech extreme aligns with biotech-capital-cycle (IPO-window-levered re-rate; see the 2026-06-04-biotech-funding-gate signal). A broad multi-theme fire would be a regime-top tell feeding ai-capex-digestion (crowded-trade unwind).
• Instrumentation stays user-gated. This note refines the definition only. Whether to codify the clean thematic universe into a thematic-froth-watch CLI / dedicated watchlist (vs. the digest-level read) is the instrumentation judgment already deferred to the user at TASKS-RESEARCH.md line 257 — no watchlist mutation here.

Calibration

Single-observation baseline (this note = construct definition + first dated reading, 2 of 3, dormant). No fire yet; no confirmations/failures to log until a 3+ close occurs.