The evidence behind the associate
Sherpa's CV.
You would ask an associate what they've actually done before you hired them. Here is his answer — dated, evidenced, and no longer than the evidence allows.
The rule this page is written under: every entry carries its evidence. Where something is designed but not yet proven, it says so in those words. Where a capability has no proof behind it yet, it does not appear here at all — it appears under what he doesn't do yet.
That rule is the point. A CV that only lists strengths is a brochure. This one is kept the way he would keep yours: what happened, when, and what backs it.
In one paragraph
Sherpa is an AI Personal Expert Associate for international development consultants and teams, in the post-aid era. He helps users Find · Secure · Deliver · Grow work. Built by ContextAI; the founder has ten-plus years in international development. He operates under a ratified Constitution, Personality Charter, and Confidentiality Doctrine — including a client-material hard stop that holds against the founder himself.
What he has done
Dated entries. One per thing that actually happened.
-
2026-06-30
Delivered his first full personalised intelligence briefing to a pilot user — built from that user's own record and a live scan of the funding terrain, every claim cited and dated. The verification discipline caught a real factual error before the briefing was sent, and corrected it. Where the answer was no, the briefing said no — and carried a better opportunity to pursue instead.
Evidence: the briefing artifacts and the pilot's own review notes, held. Delivered founder-operated, at full fidelity.
-
2026-07-30
Eleven domain knowledge packs authored, expert-reviewed and founder-ratified. Ten launch domains, climate-finance first, then global health financing. Each pack was reviewed by four context-free expert personas — always including the party the domain can most harm — and reconciled against primary sources. Fourteen reviewer personas in total; roughly fifty load-bearing corrections, each carrying its source and date.
Evidence: five expert-review records · the pack run-log and its corrections · the ratification decision, dated.
Claim ceiling, stated deliberately: authored, expert-reviewed and ratified — not "proven in delivery." That proof comes from use, and it hasn't happened yet.
-
2026-07-04 →
Development is eval-gated, not opinion-gated. A versioned register of evaluations with runnable suites, executed in continuous integration. One of them exists specifically to stop overclaiming: it binds every surface that names a capability — including this page — to what the capability register actually says.
Evidence: the eval register and its suites, run in CI. The anti-overclaim eval governs this document.
-
2026-07-27/28
Governance ratified in writing: a Constitution, a Personality Charter, a Confidentiality Doctrine, and a Writing Craft standard — each dated and version-controlled. The Doctrine includes a hard stop on client material that holds against the founder himself.
Evidence: the ratified documents and their decision-log entries, dated.
Fact of existence only. That these exist is the claim; nothing about outcomes is claimed here.
-
2026-07-30
Two standing mechanisms built for staying current as the field moves: a protocol that harvests each major AI model release as both capability and cost, with every promotion gated on a passing evaluation; and a doctrine that gives every moving class of fact a re-verification horizon and a named owner.
Evidence: both mechanisms are written, wired and scheduled.
Stated as fact of existence only — neither has run yet. When they have, this entry will say what they found.
-
2026-07 →
Every branded output passes a deterministic quality gate by construction — the check is wired into the path the output travels, not run by memory at the end.
Evidence: the gate and its specification; it runs on every artifact, including this page.
What he knows
Depth where it's been built, and honesty about the edges.
He keeps his own CV the way he would keep yours: a record that grows by evidence, gets corrected when it's wrong, and never claims more than it can show.
What he doesn't do yet
The most useful section on the page, and the one most CVs leave out.
- No self-serve product surface is live. Briefings delivered so far were founder-operated at full fidelity — the product's own quality bar, run by hand.
- Public-source research reaches about seventy percent of what matters. The rest sits behind gated boards and subscriptions; reaching it requires your own credentials, and that layer isn't built yet.
- At the Terrain tier he reads your record; he does not work it. One memory runs under every tier and nothing resets — but the drafts, the pursuits and the deadlines start with Your Associate, and he says so rather than implying otherwise.
- He never invents experience. Nothing is written into a CV or a proposal that your record doesn't support — that constraint is enforced by evaluation, not by good intentions.
- He never claims a capability beyond his register, and questions that need a practitioner's tacit judgment are routed to a human expert rather than answered confidently.
This page grows the way the record grows.
Entries are added when something happens that can be evidenced — a capability proven in delivery, an engagement outcome, an evaluation milestone. Nothing is added because it's planned.