Replication — the reviewer package (Dell DE→TX event study)

How to reproduce

Three commands.

python3 run_event_study.py                  # PRIMARY anchor  T0 = 2026-05-04
python3 run_event_study.py --t0 2026-05-05  # SENSITIVITY     T0 = 2026-05-05
Rscript  run_event_study_port.R             # third-platform reconciliation

All six models are computed from one equity panel: 12 dividend-adjusted price series, 1,254 return days (2021-06-11→2026-06-10). Estimation window = trading days, ending trading days before T0 so no AI-rally day leaks into estimation.

Integrity check (passed). The rebuilt Dell return series reproduces the brief’s pinned facts to the basis point: disclosure day 2026-05-04 , 2026-05-06 , 2026-05-08 . An independent second implementation matched three test cells to <0.01pp (market-model ; donor-weighted @May5 ; market+sector+AI ).

The files

What is in the package.

File	What it is
market_panel.csv	12 dividend-adjusted price series, 1,254 return days (the only rebuilt artifact, from the raw ICE export)
run_event_study.py	The locked five-model battery
run_event_study_port.R	Independent R port for the cross-platform check (±0.5pp CAR / ±0.05 p / ±0.01 R²)
run_placebo_battery.py	Placebo-in-time + placebo-in-space battery
event_study_results_t0_2026-05-04.{csv,json}	PRIMARY battery output (T0 = May 4)
event_study_results_t0_2026-05-05.{csv,json}	SENSITIVITY battery output (T0 = May 5)
placebo_battery_t0_2026-05-0{4,5}.{csv,json}	Placebo results on both anchors
ai_loading_cross_section_screened.csv	AI-loading cross-section (screened, N=7)
ai_cross_section_robustness.csv	Theil–Sen / Spearman robustness for the cross-section
loo_donor_sensitivity_v2.csv	Full leave-one-out over all donors, both anchors
garch_day0_results.txt · garch_t_day0_results.txt	GARCH(1,1) day-0 standardized AR (Normal + Student-t refits)
HOSTILE_RED_TEAM_v3.2.md	Internal adversarial review, triaged P0/P1/P2
REVIEWER_PACKAGE.md · RESULTS_README.md	Package guide and one-page results summary
MANIFEST_sha256.txt	SHA-256 checksum for every file in the package (27 entries)

Results, both anchors

The primary battery.

CAR in % of price; p_BH = Benjamini-Hochberg; the primary read is under T0 = . Every primary-window cell is statistically indistinguishable from zero.

PRIMARY — T0 = (the clean disclosure day), window

Model	CAR	p_BH
Mean-adjusted
Market (S&P 500)
Market + sector (XLK)
Market + sector + AI
Donor-weighted control portfolio
Donor portfolio (drop NVDA)

Donor weights (T0 = May 4). ; pre-period RMSE . Leave-one-out: the swing donor is window-specific (ANET at May-5 , ; at May-4 , ); NVDA is never the swing donor. No significance verdict flips.

Citation & link validation (zero tolerance)

Every authority resolves to its exact primary source.

Authority	Bluebook 21st form	Primary-source link
Class V settlement	In re Dell Techs. Inc. Class V S’holders Litig., 300 A.3d 679 (Del. Ch. 2023) (revised Aug. 21, 2023), aff’d, No. 349, 2023 (Del. Aug. 14, 2024)	courts.delaware.gov
SB-29 3% floor precedent	Gusinsky v. Reynolds, No. 3:25-cv-01816-K (N.D. Tex. Mar. 17, 2026) — Southwest Airlines case; Dell not a party (analogy only)	dockets.justia.com
ExxonMobil Texas vote	Exxon Mobil Corp., Current Report (Form 8-K), Item 5.07 (acc. ) — Proposal 4 approved ,	sec.gov/EDGAR
Conversion mechanism	DGCL §266 (Del.); TBOC Title 1, ch. 10, subch. C (Tex.)	delcode.delaware.gov · statutes.capitol.texas.gov
The Texas menu	Tex. Bus. Orgs. Code §§21.552, 21.419, 21.606, 21.373, 21.4161	statutes.capitol.texas.gov
Dell filings	DEF 14A (acc. 0001193125-26-226734); PRE 14A (0001193125-26-203969); DEFA14A (0001571996-26-000017); Form 10-Q (dell-20260501)	sec.gov/EDGAR
Arista earnings confound	Arista Networks Inc., Form 8-K, Ex. 99.1 (Q1 FY26 earnings release, May 5, 2026) — screen #2 exclusion	sec.gov/EDGAR

Rule. Every link resolves to the exact primary source (statute → statutes.capitol.texas.gov; Delaware opinion → courts.delaware.gov; SEC → sec.gov/EDGAR). Practitioner commentary may appear as commentary, never as the primary citation target.

Open items (carried)

What remains, stated honestly.

Pin the intraday crossing time. EDGAR acceptances were post-close, so May 4 and May 5 are reported co-equal; the Business Wire wire-minute determines which anchor is primary.
Reconcile the R port within tolerance before any number is treated as final.
Re-run FF3/FF5/Carhart event-window abnormal returns after the next Ken French daily-factor release covers May 2026 (deferred, not imputed).
Reconcile the rebuilt panel against the frozen bundle market_panel.csv (hash f4605c30…); the economics are identical (same ICE prices, same adjustment), but the rebuilt file will not match byte-for-byte.
Expand the donor pool to ≥ firms so the placebo-in-space and cross-section tests can reject at 5%; the current N=7/8 results are preliminary characterizations.