Methodology v2.5: stronger discipline, cleaner public research

We updated our public methodology marker to v2.5 for the May 2026 research cycle.

The reason is simple: as coverage expands across more sectors, company stages, balance-sheet profiles, and markets, a static one-size-fits-all research habit becomes weaker. A semiconductor platform, a regulated utility, a pre-profit biotech, a bank, a REIT, a cyclical manufacturer, and an overseas listing should not all be forced through the same economic lens.

What readers should notice is not a louder black box. It is a cleaner report.

What changed

We made the framework more context-aware. The research now puts more pressure on whether the assumptions fit the specific business, industry structure, maturity phase, financial strength, and evidence quality of the company being analyzed.

We tightened the public report hygiene. Private calibration labels, third-party benchmark names, and internal reference handles should not leak into reader-facing text. Readers deserve clear reasoning, not backend shorthand.

We strengthened release checks. A report should not publish just because a model produced a number. It should pass valuation-coherence checks, data-provenance checks, label and translation checks, and audit-appendix readability checks before it appears on the site.

We also made the public methodology page more careful. It explains the principles behind the research, but it no longer exposes internal orchestration, prompt design, provider wiring, private benchmark logic, or calibration controls.

What we will disclose

We will continue to show the methodology version, the report date, the fair-value range, the rating band, the key assumptions readers need to evaluate, the bear case, the sensitivity work, and the main risks that could make the conclusion wrong.

That is the part that belongs in public.

What we will not disclose

We will not publish the internal routing, prompts, private calibration tables, provider wiring, benchmark implementation, or exact control logic behind the research system.

That work is part of the company's research infrastructure. It is how we try to make the reports more robust, more consistent, and harder to game. Publishing those details would make the product easier to copy and easier to reverse engineer without making the reader meaningfully safer.

The balance we want is transparency without self-sabotage: enough public reasoning for a serious reader to challenge the conclusion, not enough operational detail to recreate the machinery.

The promise

Version 2.5 is not about being more bullish or more conservative. It is about being more accurate to the facts in front of us.

Some companies deserve a wider range because uncertainty is real. Some deserve less punitive discounting because the financial strength, return profile, and growth runway support it. Some deserve a hard no even when the headline multiple looks cheap.

The job is not to force every stock toward the middle. The job is to make the conclusion fit the evidence.