PREVIEW — real data, not yet part of the daily build. Unlinked page; delta baseline starts with the next daily snapshot archive.
Research Brief — GDX
build 77c56208 · data through 2026-07-29 · methodology v5
WHAT CHANGED- Since build 77c56208, GDX added 0 filings, 0 accepted events, and 0 newly matured events.
- No issuer changed C6 state.
- Coverage weight unchanged.
- Retroactive edits: 0.
WHAT MATTERS- Evidence is concentrated: EGO and ORLA hold 14% of accepted events and 22.9% of lens weight.
- Form 4 coverage: 5/28 covered issuers — most covered Canadian issuers report insider trades through Canada's SEDI system, which is not yet ingested.
WHAT THE EVIDENCE SAYS- Peer-adjusted event performance is +0.1% at day +20 (95% CI [-1.1%, +1.2%], n=269) — status: DESCRIPTIVE.
WHAT REMAINS UNCERTAIN- Confidence intervals shown are naive: they assume independent events, while issuer clustering and overlapping windows make them optimistic. Clustered inference is scheduled and will be pre-committed before computation.
- C6 sign confirmation remains accruing; posture states are research classifications of evidence rarity, not directional calls.
INSPECT NEXTordered by trigger count — not an investment ranking
- AEM — C6 posture elevated
- NEM — 10.8% lens weight but no C6 analysis yet
547 filings → 470 prose → 420 candidates → 420 accepted → 392 deduped → 352 matured → 300 direction-eligible
Investment implication: none established — no buy, sell, or alpha conclusion is supported by this page.
brief 4c2e3c7bd70db889 · deterministic synthesis — no generative model · hover any sentence for its backing data fields
Evidence Decision Canvas — GDX
build 77c56208 · data through 2026-07-29 · method b717068e17e3f09a (pre-committed clustered-inference spec)
| HORIZON | MEAN | CLUSTERED CI | NAIVE CI | WILD CI | N / ISSUERS | MDE 80% | STATUS |
|---|
| +5d | +0.4% | [-0.3, +1.2] | [-0.3, +1.1] | [-0.4, +1.1] | 292 / 28 | 1.1% | DESCRIPTIVE |
| +10d | +0.4% | [-0.5, +1.4] | [-0.4, +1.2] | [-0.5, +1.4] | 285 / 28 | 1.4% | DESCRIPTIVE |
| +20d | +0.1% | [-1.3, +1.6] | [-1.1, +1.2] | [-1.4, +1.5] | 269 / 28 | 2.1% | DESCRIPTIVE |
median 9 events/issuer · top-3 issuers 23% of sample · clustered CI is primary; naive shown for comparison
What would change this conclusion
- If the effect stays near +0.1%, roughly 9808 additional distinct issuers with events (~9836 total clusters) would be needed before the clustered CI could exclude zero (1/√G approximation).
- A true effect smaller than the current MDE (2.1% at 80% power) cannot be detected at this sample size — absence of significance is not evidence of absence.
- SEDI ingestion could reveal an insider pattern not visible in the current Form 4 scope, changing the risk-side reading for MJDS issuers.
- If naive and clustered intervals diverge further as events accrue, issuer clustering — not the event effect — is doing the work; per-issuer decomposition becomes the priority.
Investment implication: none established — no buy, sell, or alpha conclusion is supported by this page.
CAR panel — clustered inference + sample anatomy
Peer model vs SPY — what these CARs measure: each event's expected return is α+β·(equal-weight mean return of the other covered names in this lens), estimated over the 60 trading days before the event (gap 6, minimum 30 usable days). The peer model absorbs sector-common moves — a commodity swing that lifts every name in the lens cancels out — so the residual isolates issuer-specific reaction. A SPY-relative CAR answers a different question (movement vs the broad market) and widens whenever the sector as a whole moves; the Lab's cohort event study reports that variant. All values below are peer-model CARs in percent, direction-aligned.
| Horizon | Mean CAR | Cluster CI (primary) | Naive CI (comparison) | Sample anatomy |
| +5d | +0.40% | [-0.35%, +1.18%] | [-0.32%, +1.11%] | 292 events · 28 unique issuers · median 10 events/issuer · top-3 issuer share 22% · window overlap 100%
minimum detectable effect at 80% power: ±1.10% |
| +10d | +0.42% | [-0.52%, +1.42%] | [-0.38%, +1.23%] | 285 events · 28 unique issuers · median 10 events/issuer · top-3 issuer share 22% · window overlap 100%
minimum detectable effect at 80% power: ±1.37% |
| +20d | +0.08% | [-1.34%, +1.62%] | [-1.09%, +1.24%] | 269 events · 28 unique issuers · median 9 events/issuer · top-3 issuer share 23% · window overlap 100%
minimum detectable effect at 80% power: ±2.14% |
Event-type decomposition — pooled CAR at +20d
Decomposes the pooled +20d CAR by accepted event type — shows whether the pooled effect concentrates in a single type. Cluster CI (issuer-clustered, registered machinery) where the type has ≥2 events across ≥2 issuers; thin types show mean only, badged UNDERPOWERED. MDE = minimum detectable effect at 80% power.
| Event type | Mean CAR | Cluster CI | n | Issuers | Badge | MDE (80%) |
| OTHER_MATERIAL | +1.42% | [-0.26%, +2.93%] | 137 | 26 | DESCRIPTIVE | ±2.30% |
| GUIDANCE_RAISE | -2.08% | [-6.56%, +2.94%] | 22 | 14 | DESCRIPTIVE | ±6.83% |
| DIVIDEND_CHANGE | -1.43% | [-3.60%, +1.78%] | 18 | 14 | DESCRIPTIVE | ±3.79% |
| EXEC_CHANGE | -2.05% | [-4.75%, +1.00%] | 18 | 12 | DESCRIPTIVE | ±3.98% |
| M_AND_A_ANNOUNCE | -7.88% | [-11.65%, -2.07%] | 16 | 7 | UNDERPOWERED | ±7.02% |
| CAPACITY_CHANGE | +0.75% | [-4.46%, +6.23%] | 13 | 10 | DESCRIPTIVE | ±7.75% |
| REGULATORY_ACTION | -1.94% | [-7.72%, +3.71%] | 8 | 6 | UNDERPOWERED | ±8.65% |
| BUYBACK_ANNOUNCE | +6.07% | [+1.62%, +10.67%] | 7 | 7 | UNDERPOWERED | ±6.49% |
| EARNINGS_BEAT | +0.41% | [-4.94%, +5.34%] | 7 | 7 | UNDERPOWERED | ±7.35% |
| M_AND_A_CLOSE | -0.57% | [-9.49%, +4.38%] | 7 | 5 | UNDERPOWERED | ±9.97% |
| PARTNERSHIP | +1.29% | [-5.50%, +6.19%] | 7 | 5 | UNDERPOWERED | ±8.17% |
| INSIDER_SELL | -0.96% | [-6.21%, +4.05%] | 6 | 5 | UNDERPOWERED | ±6.93% |
| GUIDANCE_CUT | +7.26% | [+5.96%, +9.87%] | 3 | 2 | UNDERPOWERED | ±4.00% |
Evidence structure
protocol 34ed4732b6e6 · registered before computation
420 filings → 352 deduped events → 21 stories →
153.0 effective (story-level design effect 2.301;
issuer-level effective count 226.1). Top story 93.8% of event mass.
| Story size | Mass |
Dominant issuer | Dominant type |
Span |
| 330 | 93.8% | EGO | OTHER_MATERIAL | 120d |
| 2 | 0.6% | NEM | INSIDER_SELL | 1d |
| 2 | 0.6% | AU | INSIDER_SELL | 2d |
| 1 | 0.3% | OGC | EXEC_CHANGE | 0d |
| 1 | 0.3% | SA | REGULATORY_ACTION | 0d |
Many events, fewer stories — statistics on this page use cluster-aware inference accordingly.
Investment implication: none established — no buy, sell, or alpha conclusion is supported by this page.
Falsification battery · pooled CAR at +20
protocol df92008643c8 · N=1000 per null · seed 20260727 · registered before computation
| Test | Real value |
Null mean ± sd | Null 2.5–97.5% |
Percentile of real in null |
| Event-date shuffle |
+0.13% |
-0.37% ± 0.56 |
(-1.43, +0.67) |
0.812 |
| Direction sign-flip |
+0.13% |
-0.00% ± 0.63 |
(-1.22, +1.19) |
0.574 |
| Pre-event placebo (−20d) |
-0.75% |
located in the date-shuffle null |
0.237 |
Two-sided extremity, exploratory: percentiles near 0 and near 100 are both extreme; mid-range
percentiles mean the real value is unremarkable within its null. All cells ledger-counted under
Falsification Battery v1.
Pre-event momentum conditioning · pooled CAR at +20
protocol dfee13621c33 · median split by prior issuer-vs-peer relative momentum · registered before computation
| Window | Median rel-mom |
Outperformed half | Underperformed half |
Difference | Envelope |
| prior 60d | -1.7% | -2.49% (n=122) | +2.76% (n=122) | -5.25pp | (-8.21, -2.52) |
| prior 20d | +0.2% | +0.31% (n=122) | -0.03% (n=122) | +0.34pp | (-2.20, +2.91) |
Secondary, exploratory, ledger-counted; extremity read two-sided. A negative difference means prior
outperformers deliver more adverse aligned CARs — a reversal structure, not event information.
Evidence lifecycle
protocol e02bd40cc884 · display-only · registered before computation
| Type (backfill era) | n | Peak day | Peak |CAR| | Half-life |
|---|
| OTHER_MATERIAL | 139 | 20 | 6.466% | not reached within window |
| DIVIDEND_CHANGE | 16 | 18 | 5.988% | not reached within window |
An absolute cumulative path rises mechanically with horizon — a window-edge peak with no half-life is
what no-decay looks like under this estimator. Types below the n>=15 floor, listed not plotted:
EXEC_CHANGE (n=13), GUIDANCE_RAISE (n=10), M_AND_A_ANNOUNCE (n=8), CAPACITY_CHANGE (n=7), EARNINGS_BEAT (n=7), BUYBACK_ANNOUNCE (n=7), REGULATORY_ACTION (n=5), PARTNERSHIP (n=5), M_AND_A_CLOSE (n=4), INSIDER_SELL (n=3), GUIDANCE_CUT (n=1). Evidence freshness: median age 10d, 32.1% within 7d, 21.4% older than 30d. Investment implication: none established — no buy, sell, or alpha conclusion is supported by this page.
Form-4 transaction taxonomy · covered members
deterministic XML parse of 607 ingested filings · display-only, zero LLM, scoring inputs untouched · built 2026-07-27 01:42 UTC
| Code class | Transactions |
Share | $ mass (non-deriv rows) |
| S · discretionary open-market sale | 2 | 40.0% | $833,325 |
| S · Rule 10b5-1 plan sale | 3 | 60.0% | $1,001,952 |
Top-3 members by ingested events
| Ticker | Ingested events |
S discretionary | S 10b5-1 plan |
F withholding | M+A mechanical |
| NEM | 3 | 0 | 3 | 0 | 0 |
| CDE | 1 | 1 | 0 | 0 | 0 |
| RGLD | 1 | 1 | 0 | 0 | 0 |
10b5-1 is the filing-level checkbox. Counts cover both transaction tables; dollar mass from
non-derivative rows only. Research classifications, not recommendations.
Daily delta — 2026-07-28 → 2026-07-29
Rendered from the archived snapshots (output/synthesis/<date>_GDX_snapshot.json) — real day-over-day changes, not recomputed history.
- filings ingested 533 → 547 (+14)
- with usable prose 462 → 470 (+8)
- candidate events 415 → 420 (+5)
- accepted 415 → 420 (+5)
- after dedup 388 → 392 (+4)
- matured 349 → 352 (+3)
- direction eligible 298 → 300 (+2)
- AEM accepted events 9 → 10 (+1)
- KGC accepted events 17 → 19 (+2)
- ARIS accepted events 12 → 13 (+1)
- EXK accepted events 14 → 15 (+1)
- matured CAR events at +5d: 290 → 292 (+2)
- matured CAR events at +20d: 266 → 269 (+3)
Grade profile — what each letter is made of
Coverage classifications, not rankings. Dimension values are medians across the names holding each grade: coverage = filings ingested · freshness = days since last event · depth = prose filings · diversity = distinct event types · price-integrity = % of last 30 trade dates with a close.
Grade A (28 names, medians)
coverage18.5
freshness9
depth15.5
diversity5
price-integrity100%