Verdict: Five standing watches, three retire-or-keep-as-carrier, one coverage fix. The "weird ETF" layer is doing real work — wrapper mechanics surfaced a clean MSTR-vs-NVDA bifurcation that no single-name read produces, and DRAM is the cleanest single-instrument memory-supercycle expression we have. But the universe is over-broad: nuclear/copper/lithium baskets are noise at this regime point and the synthetic-income/daily-reset corner is mostly broken-and-mean-reverting rather than informative. The one structural fix worth doing is the coverage gap: DRAM (Roundhill Memory ETF) lives in memory.json + etf-ideas.json but not in etf-monster-universe.json, which is why the 5/10 + 5/23 scans missed it from tier tables despite it being the strongest single-instrument signal in the set.

This is a producer-only consolidation against the 5/23 carrier — no fresh scan run. 5/26 is a market-open day (US Memorial Day was 5/25; 5/26 is the next trading session and live tape isn't relevant to the consolidation question).

What we're consolidating

Three existing traces:

• research/scans/2026-05-10-scan-monster-etf-v2.md — original 50-ticker mechanism-aware tier read (T1=5, T2=10, T3=20, T4=15). MSTR wrappers + WGMI on top.
• research/scans/2026-05-23-scan-monster-etf-v2.md — 13-day follow-up tier read (T1=4, T2=6, T3=5, T4=35). MSTR wrappers rolled into bottom; thematic ETFs cooled; BST/BSTZ broke out as tech-CEF.
• research/issues/2026-05-23-weird-etf-rotation-read-mstr-leverage-breakdown-ai-stack-narrowing-defensives-fl.md — consumer-shaped digest that already synthesized the 13-day delta and connected it to ai-power-bottleneck / optical-supercycle.

Plus two infrastructure pieces:

• workflow/SCAN-MENU.md row 115 — "Weird-ETF mechanism-aware scan" already wired into the Daily Tape Check recipe (commit cb23af5a9a fixed the orphan).
• stonks/watchlists/etf-monster-universe.json — 50-ticker universe with mechanisms map (type / leverage / underlying / issuer).

What this investigation produces: which of the 5/23 reads are durable signals vs single-snapshot noise, what to keep standing, what to retire, and one coverage gap to file.

The five questions

Q1: MSTR wrappers — still broken, mean-reverting, or fresh risk-on?

Answer: Broken-and-mean-reverting. Treat as a wrapper-decay tell, not a long.

Tape (5/22 close, from etf-monster-universe.json):

All three wrappers are doing exactly what daily-reset / synthetic-income mechanics predict on a choppy underlying: NAV decay with a residual mean-revert bounce on any directional MSTR move. The 3m positive read (+38% / +38% / +26%) is the bounce-from-bombed-out floor, not a new uptrend — the 30d is negative, 1y is -50% to -92%, and RSI is mid-40s on a Tier-1-classified row only because the leverage-adjusted-30d math counts.

Standing read going forward: when MSTR rallies hard, MSTU/MSTX will outperform on a single-day basis (mechanics work); when MSTR chops, NAV bleeds. This is not a tradeable signal — it's a tell that any sustained "MSTR back" narrative needs to show up in the underlying first, not the wrapper. The 5/10 → 5/23 transition (MSTU went from "+49.3% adj 30d Tier 1" to "Tier 1 only on lev-adj-30d math because the absolute 30d was -18.8%") is the canonical example of how the mechanism-aware scoring catches a flip the surface-level monster-shape filter misses.

Keep: MSTU/MSTX/MSTY in the universe with daily-reset-leveraged / synthetic-income flags. Useful as wrapper-decay observatory.

Q2: NVDA/TSLA leverage wrappers — divergence vs MSTR still intact?

Answer: Yes, the bifurcation held. NVDA leverage works; TSLA leverage works mildly; MSTR leverage doesn't.

The 1y NVDL +115% vs MSTU -92% spread is the answer. Same leverage mechanism, same issuer class, completely different outcome — because NVDA has been trending and MSTR has been chopping. Daily-reset leverage works on trends, dies on chop. This is textbook, but the wrapper basket lets us measure it in one number.

TSLA wrappers are the noisy middle case: TSLL +27% 30d looks like a Tier 1 move but the 1y is +11% because the 3-month chop ate it. TSLY synthetic-income is mildly positive across all timeframes — same mechanics-prediction (income-cap caps the upside but the underlying trended enough to net positive).

Standing read going forward: the NVDL/MSTU spread is the cleanest "underlying-regime" indicator we have on AI-vs-crypto leverage flows. When the spread narrows (MSTU recovers / NVDL stalls), that's a regime-shift signal. When the spread widens (NVDL extends / MSTU bleeds further), AI-side concentration is intact and crypto's leveraged-bet money is gone. Right now: spread is wide; AI-side concentration confirmed.

Keep: NVDL/NVDY/TSLL/TSLY in the universe; standing-watch is the NVDL/MSTU spread specifically (not the absolute level of either).

Q3: Thematic ETF froth — which deserve standing watches?

Answer: Three keep, three retire-to-carrier, one watch-but-broken.

Keep three for standing watches: DRAM, CHAT, HUMN. Each is the cleanest single-instrument expression of an active or candidate perspective (memory / AI-narrowing / humanoid-launch-cycle), and each has a tape that's actually moving.

Retire to carrier: LIT, COPX, URNM, NLR, NUKZ (commodity / nuclear basket). They stay in the universe (so future scans run them) but don't deserve a "watch the ETF" standing item — when the named ticker thesis is the right level (ALB for lithium, FCX for copper, CEG/VST for nuclear), the basket is downstream noise.

Keep as carrier: MAGS, ROBO, BOTZ. They're useful as comparison-baselines, not as primary signals.

Q4: Buffered / income / CEF — regime tells or noise?

Answer: BUFR + JEPQ + BSTZ are durable regime tells. JEPI/DIVO/QYLD are noise at this regime point. PDI is a separate credit-side tell.

From stonks/data/summaries/buffer-signals.json + etf-monster-universe.json (5/22 close):

Standing watches from the buffer/income/CEF corner:

• BUFR + JEPQ — paired low-vol-melt-up tell (one from buffer math, one from covered-call math).
• BSTZ + BST — tech-CEF as AI-leverage expression without daily-reset decay.
• VIXY — complacency floor.
• PDI — credit-side canary (this is the new finding; PDI wasn't flagged in 5/10 or 5/23 prints because the universe is equity-focused).

Retire to carrier: JEPI, DIVO, QYLD, USA, MOAT/NOBL/COWZ/CALF/DGRW/SCHG/IWO (the "smart-beta" set — useful as cohort comparison, not as primary signals).

Q5: Are any ETF issuer / holdings feeds missing?

Answer: Yes — one structural fix and three nice-to-haves.

Structural fix (file as [ ] follow-up):

1. DRAM is missing from etf-monster-universe.json despite being in memory.json, etf-ideas.json, and buffer-signals.json. The 5/10 + 5/23 monster-ETF scans both surfaced DRAM only via the carrier digest's separate read of memory.json — the tier table didn't see it because the universe didn't include it. Add DRAM to etf-monster-universe.json with {type: "thematic-rules-based", leverage: 1, underlying: "memory-HBM-basket", issuer: "Roundhill"} so future monster-ETF scans run it natively.

Coverage data gap (file as [ ] follow-up):

2. BIGB has null data in the precompute — price/RSI/all returns are null despite being in the universe since v2 (5/10). Either the ticker doesn't fetch (Roundhill Big Tech might be too-new for our fetch pipeline) or it's a ticker-resolution issue. Verify against stonks/data/stocks/BIGB/ and either backfill or remove from universe.

Nice-to-haves (not blockers):

3. ETF holdings feeds — we have ticker-level summaries but not "what's inside" feeds for the thematic baskets. For DRAM specifically, knowing whether the basket is overweight SK hynix (HY9H.F) vs MU vs SNDK would be useful for the memory-supercycle thesis. Out-of-scope here; flag as workshop-tier note in TASKS-ENGINE.md if ever load-bearing.

4. Issuer-level concentration view — Roundhill alone has CHAT / HUMN / MAGS / METV / BIGB / DRAM in or adjacent to the universe. If Roundhill thematic-ETFs are issuing more product than they're attracting flows for, that's itself a frothy-thematic-launch tell. Out-of-scope; nice-to-have if we ever build issuer-views.

5. ETF flow data — % AUM / daily flow / premium-to-NAV is not in our precompute. Could be a real edge for the CEF corner (BST/BSTZ premium-to-NAV is the cleanest sentiment read on those, and we're not seeing it). Genuinely useful but a real engineering lift; flag for TASKS-ENGINE.md only if a perspective demands it.

What this consolidates

The 5/10 + 5/23 traces produced three real findings that this consolidation locks in as the durable lane:

1. The MSTR-wrapper-breaks-before-MSTR signal is real and mechanism-driven. Wrapper mechanics (daily-reset NAV decay on chop) telegraph regime changes before the underlying confirms. Standing watch: NVDL/MSTU 1y spread as AI-vs-crypto leverage flow tell.

2. The thematic-ETF set is over-broad and most of it is noise. Three thematic ETFs are doing real work (DRAM, CHAT, HUMN); the rest are downstream of named-ticker theses we cover elsewhere. The desk should not have nine "standing watches" — it should have three.

3. The buffer/income/CEF corner has the highest signal-to-noise ratio in the universe. BUFR + JEPQ + VIXY + PDI together produce a four-instrument vol/credit-regime read that no single-name read replicates. This is the underrated corner of the desk.

The 5/23 digest already framed (1) and (2) at the digest level; this investigation lifts (3) into a standing-watch decision, files the DRAM coverage gap as the one structural fix, and prunes the universe by retiring 8-10 noise-class baskets to carrier-only.

Verdict + standing watches

Verdict: keep five standing watches, retire three baskets to carrier, fix one coverage gap.

Standing watches (5):

1. NVDL/MSTU 1y spread — AI-vs-crypto leverage flow regime tell. Current: spread is wide (NVDL +115% / MSTU -92%); flag any narrowing.
2. DRAM — single-instrument memory-supercycle expression. Current: RSI 65.9, +90% 3m, -6% from 52wk high. Flag at RSI <50 or a 30d drawdown >15%.
3. CHAT — AI-narrowing basket tell. Current: RSI 67.6, +40% 3m, at-ATH. Flag at RSI <55 or any break of 52wk-high range.
4. BUFR + JEPQ paired — low-vol-melt-up regime tell. Current: BUFR RSI 71 / JEPQ RSI 74, both at-ATH. Flag if BUFR breaks SMA50 (currently $35.03) — that's the regime-change signal per [[reference_bufr_regime_signal]].
5. PDI — credit-side canary. Current: RSI 34.6, -18% from 52wk high. Flag at any further break below the 5/22 price ($16.61) or RSI <30.

Retire to carrier (still in universe, no standing watch):

• Commodity/nuclear basket: LIT, COPX, URNM, NLR, NUKZ (named-ticker thesis lives elsewhere).
• Synthetic-income duplicates: JEPI, DIVO, QYLD (JEPQ does this job at this regime).
• Generic smart-beta: USA, MOAT, NOBL, COWZ, CALF, DGRW, SCHG, IWO (cohort comparison only).

Keep universe-only (no change):

• MSTU/MSTX/MSTY — wrapper-decay observatory.
• NVDL/NVDY/TSLL/TSLY — measured via the NVDL/MSTU spread.
• HUMN — promote to standing watch if it diverges from ROBO/BOTZ (humanoid ≠ general robotics tell).
• BST/BSTZ — measured indirectly via "is tech-CEF still outperforming spot tech?" check.
• WGMI — measured indirectly via "are miners still decoupled from BTC?" check.
• VIXY — measured indirectly via "is complacency still extreme?" check.

Coverage fix:

• Add DRAM to etf-monster-universe.json (filed as follow-up below).

What this does NOT cover: the 5/26 live tape. This consolidation is against the 5/22 close carrier (per source row instruction: "label the 5/23 data as carried"). When the next full Daily Tape Check runs and monster-etf-scan:save fires fresh, the five standing watches above are the rows to inspect first — not the full tier table.

Follow-ups

Filed narrow durable items only:

1. Add DRAM to etf-monster-universe.json. ~5 min. Edit the symbols array + add to mechanisms map: {"type": "thematic-rules-based", "leverage": 1, "underlying": "memory-HBM-basket", "issuer": "Roundhill"}. Re-run deno task precompute etf-monster-universe after.

2. BIGB null-data investigation. ~15 min. Check stonks/data/stocks/BIGB/ — if the fetcher errors, fix or remove from universe; if the data is fresh but the precompute is rejecting it, debug the precompute filter.

3. Standing-watch instrumentation decision. ~30 min. Five named standing watches (NVDL/MSTU spread, DRAM, CHAT, BUFR+JEPQ paired, PDI) — should these become a single weird-etf-watch row that runs alongside monster-etf-scan:save and prints just those five with flag-conditions? Or is the digest-level read enough? Deferred to user — instrumentation judgment, not autonomous work.

4. BUFR regime-signal validation deadline. Per [[reference_bufr_regime_signal]], the calibration is single-day as of 2026-05-10; needs 3-6mo of regime-change observations to validate predictive value. Calendar item: revisit at 2026-08-10 (~3mo) with whatever regime-change data the next 3 months produce.

5. NVDL/MSTU spread historical chart. ~20 min. Pull the 1y daily NVDL/MSTU ratio and look at the chart shape — does the spread mean-revert, trend, or step-function on regime breaks? If mean-revert, the spread becomes a tactical signal; if trend, it's the structural-regime tell as described above. Open question, not blocking.

Sources / receipts

Carrier data (data_as_of 2026-05-22):

• research/scans/2026-05-23-scan-monster-etf-v2.md — tier-1-4 table from mechanism-aware scan.
• research/scans/2026-05-10-scan-monster-etf-v2.md — prior baseline for 13-day delta.
• research/issues/2026-05-23-weird-etf-rotation-read-mstr-leverage-breakdown-ai-stack-narrowing-defensives-fl.md — consumer-shaped digest synthesizing the 13-day delta.
• stonks/data/summaries/etf-monster-universe.json — full 50-ticker precompute (mtime 2026-05-23 03:24).
• stonks/data/summaries/buffer-signals.json — 19-ticker buffered/income/CEF precompute (mtime 2026-05-23 03:24).
• stonks/data/summaries/memory.json — memory-basket precompute including DRAM (mtime 2026-05-23 03:24).

Universe + scan plumbing:

• stonks/watchlists/etf-monster-universe.json — 50-ticker universe with mechanisms map (v2, 2026-05-10).
• stonks/watchlists/buffer-signals.json — 19-ticker buffer/income/CEF universe.
• stonks/cli/chart/monster-etf-scan.ts — scan CLI.
• workflow/SCAN-MENU.md:115 — weird-ETF row in scan menu.

Live calls confirmed (not new scans, just universe-level checks via precompute):

• deno eval reads against the three summary JSONs above to verify 5/22 closing values for all named tickers in the standing-watch table. No CLI scan was re-run for this consolidation.

Cited events:

• 2026-05-10-scan-monster-etf-v2 — baseline tier read.
• 2026-05-23-scan-monster-etf-v2 — 13-day-follow-up tier read.
• 2026-05-23-weird-etf-rotation-read-mstr-leverage-breakdown-ai-stack-narrowing-defensives-fl — digest carrier.

Cross-references to perspectives (no new claims, just confirmation of how the weird-ETF reads connect):

• ai-power-bottleneck — confirmed via "named tickers ($BE/$POWL/$CGEH/$GEV) running, XLU/XLB weak" frame from the 5/23 digest; this consolidation doesn't update the perspective.
• optical-supercycle — confirmed via "AAOI/AXTI/LITE still extended" from the 5/23 digest; this consolidation doesn't update the perspective.
• memory-supercycle — DRAM as cleanest single-instrument expression is the new finding here; flagged for the perspective's next refresh but not pushed as a thesis-shift (the underlying memory thesis is what's running, DRAM is just the ETF wrapper).