The vault the principal owns

In April 2026 I ran a commercial-diligence pass on a live sovereign-capital engagement. The pass took forty-eight hours and ran on one principal, the public record, a vault I have been curating since late 2023, and a long-context model. By the morning call the commercial position had moved: the pricing had omitted a fact already in the public record.

That pass is the evidence for the practitioner claim. At this layer the work is already being done this way. The decision in front of a principal is whose substrate will carry the next three years.

I use secondbrain only in that narrow sense: a vault I own, agents I run against it, and a billing relationship in my name. If I leave an institution, it comes with me. The last two years of curation have accrued to me. The next engagement uses it without depleting it.

What the April pass used

The stack has four commercially available layers. Retrieval, in my case Firecrawl, turns public-record URLs into Markdown I can read at speed. Reasoning, Claude Opus through Claude Code, runs long-context passes against the assembled evidence. Memory is the Obsidian vault: client briefs, person-docs, pricing notes, audit templates, voice guidelines, transcripts filed to the client directory. Above every run, a hierarchy of CLAUDE.md files carries house style and the frame for that engagement. The April pass itself ran on public-record URLs.

The combination, tuned to one principal’s taste, is the part that stays off the shelf. The vault is the IP. Specificity is what the relationship is made of, and the vault is where specificity lives. A principal who walks in with the vault has already paid the context cost.

What this does to the billing line

The MBB pyramid capitalises a partner’s judgment by multiplying the research hours underneath it. That held while partner-level synthesis was scarce and first-pass research was abundant. Through 2025 the abundance reversed. Partner-level synthesis, rerun against a curated vault, is now cheap enough that the binding constraint is taste, and taste is one principal at a time. A decade of principal-seat decisions is how that taste is trained.

The pricing implication is architectural. The buyer is paying for the name in the room and the work on the page. The mid-pyramid cost is absorbed into the inference budget. How many engagements a principal can carry is then a function of inference budget, sleep, and the quality of the substrate.

Pyramid ratios and document counts stay off this page. The April pass is the evidence I can stand behind.

Where the claim holds

The substrate multiplies judgment that is already there. An empty vault produces generic output at any model quality. The method is available to any principal with the discipline to run it.

The pattern is for principal-to-principal work, where specificity carries the conversation. Delivery at scale (implementation teams, physical ops, regulated handoffs) still needs head-count. This piece is about the advisory layer.

What I will mark this against

I will mark the claim against three observables.

By 30 September 2026: whether the next four principal-to-principal engagements I run can be repositioned by a single-principal research pass inside three days. If they stay put, this was one favourable weekend.

By the end of 2026: whether peer studios running similar stacks, with thinner vaults, ship work of comparable specificity. I will run three comparisons on the same rubric I use internally.

Through 2027: whether MBB, the Big Four, and the tier-one banks ship per-partner capability that matches single-principal output. Published briefings and client-facing announcements are the record.

I will mark the piece at year-end 2026 and again in Q3 2027.

The question

Every new engagement I take turns on whose substrate we will run on once it lands. You are choosing a stack for the next three years of advisory work. That stack is the capex of the next cycle. Waiting for the market to settle is a choice with a date on it.

The vault you have been building, on your schedule: is it already yours, or does it sit on a server you lose the day you leave? If you are running that calculation against a live stack decision, and you reach a different answer, I want to hear it.

Gopal Patel is principal at Navaro Advisory; prior roles include CTO at Auriens and operational audits across services, institutional hospitality, and AI-native procurement.