Seven & i Holdings, July 2025.

A contested take-private. An arb book exposed to a family-governance outcome that was in the seed material if you knew where to look. The desks that lost money had the information. They just did not rehearse it.

Arbitrage funds lost

$90M

in a single day · $200 to 400M broader event-driven book damage

T=0

2025-06-27

Three trading days before the 30 June IC meeting

01 / The decision

A deal that everyone was reading as progressing.

By 30 June 2025, the Seven & i Holdings / Alimentation Couche-Tard contested take-private had been in NDA and standstill phase for 60 days. Couche-Tard's divestiture workstream was active in the US convenience-store footprint. METI had formally reclassified Seven & i under the Foreign Exchange and Foreign Trade Act core-industry designation, which every merger-arbitrage desk on the street read as a signal that METI was cooperating with the bid process rather than blocking it. ValueAct and Artisan were applying coordinated pressure through public letters. Merger-arbitrage desks globally were holding roughly $1 billion of arbitrage exposure to the deal. The IC meeting on 30 June was three trading days before Seven & i's 10 July Q1 FY26 earnings event.

That is the archetype. A special-situations Managing Director on an event-driven or activist book at a multi-strategy fund, weighing a position in a contested M&A or activist situation where the stakeholder dynamics are on the public record but the outcome is not yet set. The question is not "what will happen?" but "what will this specific set of stakeholders do, given what they have said publicly and their known incentives?"

On 16 July 2025, Couche-Tard withdrew. Seven & i fell 7%+ on the day. Arbitrage funds long Seven & i lost $90 million in a single day. Broader event-driven book damage was conservatively $200 to 400 million globally. Couche-Tard itself walked away from an estimated $150 to 200 million in deal advisory, banker retainer, and due-diligence costs. The stakeholder dynamics that drove the withdrawal — the Ito family standoff, the METI reclassification read, the pressure ceiling on ValueAct and Artisan, the Japanese retail shareholder reaction pattern — were all on the public record before the 30 June IC meeting.

T=0 / The cutoff

Evening of 27 June 2025.

The rehearsal was run on the evening of 27 June 2025, using only documents publicly available through the 27 June close — three trading days before the IC meeting. In working memory at T=0: the 60-day NDA timeline, the divestiture workstream filings, the METI core reclassification, the ValueAct and Artisan public letters, the Ito family governance structure, and the Japanese retail shareholder reaction pattern from prior Seven & i corporate actions.

Held out of the rehearsal and never uploaded to Glasshouse: the 16 July withdrawal, the 7%+ single-day drop, every piece of post-withdrawal reporting, and the broker closing marks from 16 July. Gemma 4's training cutoff is January 2025; the 16 July 2025 withdrawal is six months post-cutoff.

02 / The rehearsal

What Glasshouse produced.

The rehearsal surfaced the Ito family's stonewall posture, the METI latitude, the ValueAct and Artisan pressure ceiling, and the Japanese retail reaction pattern as the four load-bearing stakeholder signals driving the deal trajectory inside the simulation. 1 The divestiture workstream, which every arb desk on the street was reading as a deal-progressing signal, was correctly placed inside the simulation as a Couche-Tard-side cost-sunk commitment that did not change the probability of Ito-family assent.

The dominant narrative emerged in round twelve of fifteen: the deal will fail on family-governance grounds before the divestiture workstream can close. 2 The narrative was not in the seed material as an explicit prediction; it was assembled inside the simulation from the publicly-known family governance structure, the Japanese retail shareholder reaction pattern from prior corporate actions, and the specific wording of the Ito family's public responses to the offer.

The simulation's minority-view personas — the ones who placed the deal on a successful-close trajectory — were three of fifteen. The three were anchored to the divestiture workstream as a progressing signal, exactly the read the street was carrying into the 30 June IC meetings. The minority view decayed monotonically across rounds twelve through fifteen as the family-governance signal tightened. 3

03 / What actually happened

Rehearsal alongside record.

Glasshouse rehearsal · T=0 2025-06-27

Stakeholder signals & dominant narrative

01 Ito family stonewall posture — dominant governance signal
02 METI latitude — cooperation read, not block read
03 ValueAct and Artisan pressure — ceiling, not floor
04 Japanese retail reaction pattern — not supportive of take-private
R12 Dominant narrative: deal will fail on family-governance grounds before the divestiture workstream can close

The dominant narrative emerged in round twelve of fifteen. The IC meeting was three trading days away.

Real-world record · 16 to 17 July 2025

Documented outcome

01 Couche-Tard withdrew on 16 July 2025
02 Withdrawal reasoning named family-governance and Ito-side terms as the breaking point
03 Seven & i stock fell 7%+ on the day
04 Merger-arbitrage funds long Seven & i lost $90 million in a single day
05 Broader event-driven book damage conservatively $200 to 400 million globally

Primary reporting from Bloomberg, Reuters, Financial Times, and the Globe and Mail coverage dated 16 to 17 July 2025.

The stakeholder dynamics were public. The outcome was not. The desks that lost money had the information. They just did not rehearse it.

04 / Methodology

Four rules. No exceptions.

Rule / 01

T=0 is explicit.

T=0 for this case is the evening of 27 June 2025, three trading days before the 30 June IC meeting. The seed contains only documents publicly available through the 27 June close.

Rule / 02

T=0 is post-cutoff.

Gemma 4's training cutoff is January 2025. The 16 July 2025 withdrawal is six months post-cutoff. The model has no memorized outcome to retrieve from training data.

Rule / 03

Financial anchors are real.

The $90 million single-day arbitrage loss, the $200 to 400 million broader event-driven book damage, the 7%+ single-day drop, and the $1 billion pre-withdrawal arbitrage exposure come from primary reporting in Bloomberg, Reuters, Financial Times, and the Globe and Mail.

Rule / 04

Never tuned to pass this case.

Glasshouse is tuned on aggregate eval results across many scenarios. This case passes the rubric as a byproduct of aggregate tuning, not because Glasshouse was reverse-engineered around the Seven & i withdrawal.

05 / Related rehearsals