SMH Covered-Constituent Evidence Lens

Why not 100%? 13 disclosed constituents remain outside the covered sleeve — all of them US filers outside the 79-ticker scoring universe (a scoring-universe decision, not a filer-class gap). The foreign private issuers ASML, NXPI, STM, TSM are covered as of 2026-07-31 through the evidence tier (6-K current reports + 20-F annual reports, exhibit-level prose extraction — the same path shipped for the Canada Resources Evidence vertical); they are evidence coverage only and are never scored. Foreign-holdings ETFs (e.g. KORU-class country funds) remain methodologically out of scope: their constituents do not file with EDGAR at all, so there is no evidence stream to aggregate — a scope statement, not a caveat.

Research classifications, not recommendations. C6 is the evidence-impact component of the composite score; the earlier C6 finding is that insider-sale evidence may function more plausibly as a risk-state input than a near-term directional signal (see methodology).

CIs assume independent events; distinct ticker-days can still share calendar days across tickers, so the intervals are optimistic — stated, not hidden. Types with n < 10 are shown for completeness and are statistically uninformative. Form 4 (insider) events are a parsed batch covering 2026-02-18 to 2026-05-15; live Form-4 ingestion is a tracked pipeline item, so the live-era sample is 8-K derived and small.

Falsification context for this table (same population, registered battery): event-date-shuffle percentile 0.697 — unremarkable within its null, so an era-generic component cannot be excluded; direction sign-flip percentile 0.036; pre-event placebo +6.24% at null percentile 1.000. An envelope that excludes zero should be read WITH these beside it, whichever way they cut.

Matched controls (protocol bbe19aca5dd9, registered): each event paired with a same-day, event-free sleeve issuer nearest in standardized momentum+volatility (53 pairs, 5 excluded, median match distance 0.933 sd). Matched-control-adjusted E4 = -12.74% CI (-17.01, -6.25) [PRELIMINARY]. Read together with the falsification context above: the timing of the adverse alignment is not distinguishable from its era, but relative to matched factor twins on the same days it is issuer-day-specific — both facts print; neither is a recommendation.

The adverse point estimate persists under all four registered weightings; envelopes exclude zero for E1 event-weighted, E3 ETF-weighted, E4 capped-ETF-weighted and include zero for E2 issuer-weighted. Envelope = wider of the issuer-/date-cluster bootstrap CIs; the formal CGM two-way interval is reported in the table beside it.

Two-sided extremity: percentiles near 0 and near 100 are both extreme; mid-range percentiles mean the real value is unremarkable within its null. Exploratory; every cell ledger-counted under Falsification Battery v1 (protocol df92008643c8, registered before computation).

Deterministic XML parse of 607 ingested filings; 10b5-1 is the filing-level checkbox; dollar mass from non-derivative rows only. Display-only — zero LLM, scoring inputs untouched. Research classifications, not recommendations.

Point-in-time pipeline counts; the live era is never blended with the backfill era in any statistic.

Renormalization statement: posture and E3/E4 estimand weights renormalize issuer fund weights over the covered 67.15% sleeve only; no statistic on this page weights by full-fund share. Constituent weights are dated issuer disclosures, not market data.

Rounding footnote: the issuer's disclosed weights sum to 100.01%, so the undisclosed residual computes to -0.01% — shown as measured, an artifact of the issuer's two-decimal rounding, not a data error.

Evidence coverage per covered constituent — how much evidence stands under this classification — coverage, not prediction (Evidence Coverage v1, registered protocol). Names outside the 79-name scoring universe print an em-dash: the registered computation covers U79 only.

Many events, fewer stories — statistics on this page use cluster-aware inference accordingly. Linkage rule (pre-committed): same issuer within 5 trading days, or same type and lens within 3. Story-level design effect 1.49 (issuer-level effective count 92.6); with only 3 stories the story-level variance read is itself thin — both structure and variance are shown rather than choosing one. Top story carries 95.1% of event mass. Investment implication: none established — no buy, sell, or alpha conclusion is supported by this page.

Sign held in 6/7 computed cells · CI excluded zero in 1/7 · breaks in: era: live (k=20) · UNDERPOWERED cells: k=60 (backfill only), vol-high (k=20 bf). Context grid pre-declared at registration; every cell ledger-counted. 449 secondary cells ledgered registry-wide; expected false positives at alpha=0.05: 22.45. Investment implication: none established — no buy, sell, or alpha conclusion is supported by this page.

Matched-control-adjusted E4 (pairs fixed at the +20 selection) (secondary cells, same grid): sign held 5/5 (coherence 1.0) · breaks: none · empty (SPY-history bound): trend<0 (k=20), trend>=0 (k=20).

Median time-to-peak across qualifying types: 20 trading days — at the window edge for every qualifying type, with no half-life reached. An absolute cumulative path rises mechanically with horizon, so this pattern is what no-decay looks like under this estimator; it is not evidence of late-arriving impact. Types below the n≥15 floor, listed not plotted: M_AND_A_CLOSE (n=13), BUYBACK_ANNOUNCE (n=13), M_AND_A_ANNOUNCE (n=11), CAPACITY_CHANGE (n=10), REGULATORY_ACTION (n=9), EARNINGS_BEAT (n=9), PARTNERSHIP (n=5), GUIDANCE_CUT (n=1). Evidence freshness (this sleeve): median age 14d, 37.5% within 7d, 0.0% older than 30d. Investment implication: none established — no buy, sell, or alpha conclusion is supported by this page.