AI analysis of proxy contest filings from four models
Analysis run: September 8, 2026
All four models analyzed the same single factual extraction from the underlying filings — they did not independently review the source proxy materials. Agreement among them reflects a shared reading of that extraction, not independent corroboration of facts. Any error or omission in the extraction would produce identical conclusions across all four analyses. This limitation is material and should inform how heavily readers weight the convergence.
The contest is a written consent solicitation by founder and former CEO Vishal Garg (the "Garg Group") seeking to remove five of eight incumbent directors — Lewis, Massenet, Menon, Narasimhan, and Talwar — without disclosing replacement nominees. The Board's Special Committee recommends stockholders return the WHITE consent revocation card. All four models, drawing from the same extraction, recommend supporting management.
The shared reading of the extraction supports the following overall picture: Garg's tenure produced severe and documented financial losses, repeated missed guidance, and a stock decline exceeding 90% from the de-SPAC close. The consent solicitation mechanism, as structured, asks stockholders to remove a majority of the board without knowing who would replace them — a structural deficiency the extraction presented as central. Against this, the extraction also surfaced real weaknesses in the management case: the stock fell sharply on Garg's removal, he was re-elected with 99.5% support just weeks before removal, the Board offered him a lucrative advisory role immediately afterward, and Interim CEO Lewis's qualifications for a regulated fintech at Better's scale are unproven. Models weighted these competing considerations differently, which explains the variation in confidence scores.
| Model | Recommendation | Confidence |
|---|---|---|
| Claude | Support Management | 6/10 |
| Grok | Support Management | 7/10 |
| OpenAI | Support Management | 8/10 |
| Gemini | Support Management | 9/10 |
No model's stated label contradicts its reasoning. No override is warranted.
All four models, working from the same extraction, converged on the following readings:
Financial record: The extraction documents cumulative net losses exceeding $1.5 billion since 2022, a stock decline of over 90% from the de-SPAC close, and a series of specific missed guidance targets — $5B+ projected 2023 revenue versus $72M actual, $1B monthly loan volume missed by more than 40%, and withdrawal of the September 2026 adjusted EBITDA breakeven target. All four models treated these as grounded in public filings rather than solely in management advocacy, and all weighted them heavily against Garg.
Structural deficiency of the consent mechanism: All four models identified the absence of named replacement directors as a critical governance problem. Stockholders cannot evaluate a reconstituted board they cannot see. All four also noted the downstream risks: Nasdaq compliance failures if standing committees fall below required composition, potential change-of-control triggers, and the possibility that the departure of Hugh Frater (former Fannie Mae CEO) could leave Garg with effective unilateral control over vacancy appointments.
Securities law and conduct concerns: The extraction documented that Garg publicly claimed majority support on August 14 and 17, 2026, then withdrew the claim after Activant's 13D disclosed only 26.8% group ownership. The Garg Group's own filing acknowledged consents delivered on those dates did not constitute a majority. All four models treated this sequence as documented, not merely alleged. All four also noted the jury finding of fiduciary duty breach and conversion as a legal outcome, not an allegation.
Written consent as an inappropriate mechanism for this dispute: All four models, to varying degrees, concluded that the annual meeting — with named nominees and full disclosure — would be the more appropriate forum, and that the 60-day written consent process denied stockholders adequate deliberative opportunity.
Divergences among models that received the same extraction reflect differences in emphasis and weighting, not differences in available information.
Confidence spread (6 to 9/10): This is the most notable divergence. Claude assigned 6/10, Grok 7/10, OpenAI 8/10, and Gemini 9/10. The gap turns on how heavily each model weighted the management-side weaknesses the extraction also surfaced.
Claude devoted the most analytical attention to the case for the activist: the 41–57% stock decline on Garg's removal, the 99.53% re-election vote six weeks before removal, the ~$15M Vice Chairman offer made three days after removal, the court's denial of the TRO (which Claude read as a signal that the legal case against Garg's solicitation is weaker than management presents), and Lewis's unproven qualifications. Claude's lower confidence directly reflects these countervailing factors and is the most hedged reading in the set.
Gemini reached the highest confidence (9/10) by weighting the structural governance risks most heavily — particularly the scenario where Frater's departure plus five removals gives Garg near-total control over vacancy appointments — and by giving the least analytical weight to the stock-price reaction and the Vice Chairman offer. Gemini also characterized Garg's 90-day plan targets as lacking credibility more categorically than the other models.
Grok and OpenAI occupied the middle range. Grok's analysis was the most compressed and gave the least individual attention to the management-side weaknesses, which may partly explain its moderate 7/10. OpenAI's 8/10 reflects a thorough engagement with both sides but a judgment that the governance and legal concerns tip clearly toward management.
Treatment of the Vice Chairman offer: Claude flagged this as analytically significant — a ~$15M offer made three days after removal is difficult to reconcile with the Board's framing of Garg as an existential risk. OpenAI noted it. Grok and Gemini gave it less weight. The extraction presents this as corroborated by management's own filing (the Compensation Committee confirmed the offer), so the divergence is one of weighting, not of available information.
Treatment of the stock price reaction: Claude and OpenAI both treated the 41–57% stock decline on Garg's removal as a meaningful market signal. Gemini and Grok noted it but gave it less analytical weight, implicitly treating it as noise or as consistent with broader market conditions. Given the stock's pre-existing trajectory (-61% YTD), this is a legitimate interpretive difference: the extraction does not isolate how much of the decline was removal-specific versus continuation of prior trends.
Lewis's qualifications: Claude was most explicit that Lewis's prior company (Ascend Fundraising Solutions, ~$250M GMV) represents a substantial scale gap relative to Better's $1.6–1.7B quarterly funded loan volume, and that his fintech or mortgage experience is not documented in the extraction. OpenAI acknowledged the criticism. Gemini and Grok treated Lewis's qualifications less critically.
Framing of the TRO denial: Claude read the court's refusal to grant a TRO as a signal that the legal case against Garg's solicitation is not as strong as management presents, and weighted this toward the activist's side. The other three models noted the TRO denial but drew no adverse inference about the underlying legal claims.
Support Management
Strength: Moderate
The structural deficiency of the consent solicitation — removing five of eight directors without named replacements — is the dominant factor. It is present in all four models' reasoning and rests on the extraction's account of the filing record rather than on any single model's characterization. The documented financial record under Garg (corroborated by public filings per the extraction) and the documented false majority claims (acknowledged in the Garg Group's own filings per the extraction) reinforce this conclusion.
The strength is Moderate rather than Strong because the management case carries real weaknesses that the extraction itself surfaced and that Claude's analysis engaged with most carefully: the sharp stock decline on removal, the overwhelming re-election vote six weeks prior, the Vice Chairman offer, Lewis's unproven qualifications, and the court's denial of emergency relief. These are not frivolous concerns, and a reasonable stockholder could view the annual meeting — rather than either the consent solicitation or the current board's preferred outcome — as the appropriate resolution mechanism.
Confidence: 7/10
The consensus confidence is anchored below the midpoint of the model range (6–9) for several reasons. First, all four analyses drew from a single extraction rather than independent readings of the filings; agreement among them does not compound confidence. Second, Claude's more granular engagement with the management-side weaknesses — particularly the Vice Chairman offer, Lewis's scale gap, and the TRO denial — identifies genuine analytical tensions that the higher-confidence models may have underweighted. Third, the absence of named replacement nominees is the fulcrum of the recommendation; if Garg were to disclose a credible independent slate before the October 20 consent deadline, the analysis would require material revision. Stockholders with access to the underlying filings should verify the key factual claims — particularly the Vice Chairman offer amount, the exact stock decline magnitude on the removal announcement, and the current cash position — before acting.
