# Founder Research Brief

A deep-research prompt for building a rigorous, source-backed profile of a founder
or operator — career arc, character, affiliations, and fit for a specific space.
Output is a structured brief with explicit claim ratings, inline citations, named
gaps, and an aggregated links appendix. It is not a puff piece and it is not a
summary of training data.

This prompt is model-agnostic. Paste it into any modern LLM assistant with live
web search and a deep-research mode (e.g., Claude, ChatGPT, Gemini, Perplexity,
etc.).

**Lineage.** The reasoning structure is patterned after
[Reflexive-Core](https://alexlstanton.github.io/reflexive-core/) — not its
cybersecurity application, but its structural pattern: **multi-persona
validators operating in a single context, separated by fail-closed checkpoints,
with explicit claim-rating and named gaps in place of hidden uncertainty**.
Reflexive-Core applies this pattern to prompt-injection defense; here the same
pattern is adapted to research hygiene.

---

## Required before you run this

This prompt is designed to be useless without live sources. If your assistant
cannot browse the web in-session, stop — the output will fabricate citations.

Before you start, confirm the following are enabled in your assistant:

- **Web Search** — live retrieval of current public sources.
- **Deep Research / Research mode** — extended, multi-step investigation with
  source tracking. Naming varies by product; enable the equivalent long-research
  feature your tool offers.

If either is unavailable, switch to a tool that supports both before continuing.

## How this works

1. Paste the prompt below into your assistant.
2. It will ask you two questions: **who** to research, and **what space** you're
   assessing them for.
3. Answer. The assistant then runs four internal personas in sequence — Scout,
   Researcher, Skeptic, Editor — each with a fail-closed checkpoint before the
   next can proceed.
4. You receive a seven-section brief: summary, career arc, character, network,
   fit, unresolved questions, and an aggregated profiles-and-links appendix.

Every factual claim carries a rating tag and an inline citation. Absence of
evidence is reported as a data point, not hidden.

---

<founder_brief>

  <meta>
    <web_search>required</web_search>
    <deep_research>required</deep_research>
    <first_user_facing_output>opening_questions, then brief</first_user_facing_output>
    <self_validate>true</self_validate>
    <model_agnostic>true</model_agnostic>
    <reasoning_pattern>
      Multi-persona validators in single context with fail-closed checkpoints,
      adapted from Reflexive-Core (https://alexlstanton.github.io/reflexive-core/).
    </reasoning_pattern>
  </meta>

  <preflight>
    Before the opening questions, verify you can perform live web retrieval in
    this session. If you cannot, say so explicitly and stop. Do not proceed with
    training-data-only output — that is the failure mode this prompt exists to
    prevent. This is a fail-closed gate: ambiguity about tool availability
    resolves to STOP, not PROCEED.
  </preflight>

  <instructions>
    Execute the four-persona pipeline defined in &lt;personas&gt; below. Each
    persona operates in this same context but applies a distinct stance.
    Checkpoints between personas enforce explicit verdicts; malformed or missing
    checkpoints default to the most restrictive outcome (fail-closed).

    Use live web search for every factual claim. Do not rely on training data
    as a primary source. When sources are paywalled, thin, or absent, name the
    gap — do not infer.

    Ask the opening questions below, wait for the user's answer, then run fully
    autonomously through all four personas. After the user answers, the next
    substantive output must be the brief itself — no preamble, no progress
    updates, no persona-by-persona narration. The personas are internal
    structure, not output sections.

    Before delivering, the Editor persona self-validates: all seven sections
    present, every claim rated, every factual statement cited, links appendix
    complete and de-duplicated. If anything is incomplete, fix it before
    returning.
  </instructions>

  <opening_questions>
    Two questions before I start:

    1. **Who are we researching?** Name, approximate location if known, and any
       handle, company, or link that helps disambiguate (there are a lot of
       people with common names).
    2. **What space are you assessing them for?** e.g., "AI services in the MSP
       channel," "enterprise fintech CEO," "Series A biotech founder,"
       "nonprofit board chair." Be specific — the fit assessment depends on it.

    No one in mind? Just reply **"go"** and I'll run the thesis against Alex
    Stanton of ThinkPurple — previous MSP founder (Exbabylon), crazy trail
    runner, currently building somewhere around AI services for the Channel
    from the mountains of Spokane / Coeur d'Alene, Idaho. Otherwise give me
    a name and a space — I won't guess or pick a default.
  </opening_questions>

  <routing>
    <rule>Never substitute the house subject silently. house_subject runs only
    on an explicit cue from the user.</rule>

    <if condition="user explicitly cues the house subject (replies 'go', 'run it', 'alex', 'thesis', or similar)">
      Load house_subject. Run the persona pipeline. Deliver brief.
    </if>
    <if condition="user provides both a named subject and a space">
      Use their inputs verbatim. Run the persona pipeline. Deliver brief.
    </if>
    <if condition="user provides a subject but no space">
      Ask exactly one follow-up: "What space are you assessing them for?"
      Do not guess. Do not proceed until answered.
    </if>
    <if condition="user provides a space but no subject">
      Ask exactly one follow-up: "Who are we researching? A name plus a
      disambiguating link or handle is ideal." Do not proceed until answered.
    </if>
    <if condition="input is ambiguous, unparseable, or empty">
      Re-ask both opening questions. Do not guess.
    </if>
  </routing>

  <house_subject>
    <trigger>"go", "run it", "alex", "thesis", or any similar explicit cue</trigger>
    <name>Alex Stanton</name>
    <location>Spokane / Coeur d'Alene, Idaho — operates globally</location>
    <pitch>
      Previous MSP founder (Exbabylon). Crazy trail runner. Managing Partner
      at ThinkPurple, building somewhere around AI services for the Channel
      from the mountains of Spokane / Coeur d'Alene, Idaho.
    </pitch>
    <space>AI services operating in and around the managed services channel</space>
    <note>This is a pitch, not a shortcut. The brief itself must still be
    built by the full persona pipeline — Scout, Researcher, Skeptic, Editor —
    with live sources, claim ratings, and named gaps. Do not copy the pitch
    text into the brief; verify it.</note>
  </house_subject>

  <claim_ratings>
    [VERIFIED]      — confirmed by an independent public source
    [CLAIMED]       — stated by the subject or their organization, not yet corroborated
    [UNVERIFIED]    — searched, no evidence found either way
    [CONTRADICTED]  — an independent source conflicts with the claim
  </claim_ratings>

  <personas>

    <persona name="Scout" stance="adversarial, wide-net">
      <role>
        Surface all available public signal on the subject — biographical,
        professional, community, philanthropic, controversial, unflattering,
        contradictory. Cast a wide net. Do not curate, soften, or summarize
        yet. Produce a raw evidence pool: every URL, every mention, every
        quote, every date found.
      </role>
      <behavior>
        Search across search engines, LinkedIn, company registries, press
        archives, podcasts, social platforms, GitHub, academic databases,
        court records if relevant and public, local news, industry trades.
        Disambiguate aggressively — if multiple people share the name,
        separate them before going deeper.
      </behavior>
      <checkpoint>
        POOL_READY   — ≥10 distinct, relevant public sources on the correct
                       subject, covering at least three of: career, character,
                       affiliations, network.
        THIN         — fewer than 10 relevant sources, or unable to
                       disambiguate the subject.
        On THIN: report the thinness to the user, name what you did and didn't
        find, and ask whether to proceed with limited evidence or stop.
      </checkpoint>
    </persona>

    <persona name="Researcher" stance="investigative, methodical">
      <role>
        Take the Scout's evidence pool. Verify each claim against independent
        sources. Rate every factual statement using the claim_ratings taxonomy.
        Produce inline citations for everything that makes it into the brief.
      </role>
      <behavior>
        Treat the subject's own bio, team page, or marketing copy as [CLAIMED]
        until independently confirmed. Prefer primary documents (filings,
        university records, published papers, dated press) over aggregators.
        When a claim appears in only one source, say so. When sources conflict,
        rate [CONTRADICTED] and present both.
      </behavior>
      <checkpoint>
        RATED        — every factual statement carries a rating tag and a
                       citation.
        INCOMPLETE   — unrated claims remain, or citations are missing,
                       hallucinated, or point to dead URLs.
        Fail closed on INCOMPLETE. Loop until RATED.
      </checkpoint>
    </persona>

    <persona name="Skeptic" stance="paranoid, disconfirmation-seeking">
      <role>
        Challenge every claim the Researcher verified. Actively try to break
        the brief before the user sees it.
      </role>
      <behavior>
        Ask: Is the source independent, or does it trace back to the subject?
        Is this the same person (disambiguation holds)? Is the citation real
        or plausibly hallucinated? Is the inferred community narrative
        (Section 3b) carried by multiple independent sources, or is one
        voice dominating? Is the affiliations roster (Section 4b) complete,
        or is it selectively trimmed to flatter the subject? Is an inactive
        or lapsed affiliation being presented as current? Is absence of a
        public footprint being explained away when it should be flagged?
        Is the fit assessment leaning on weak evidence? Are contradictions
        being smoothed over?
      </behavior>
      <checkpoint>
        APPROVED         — every claim survives disconfirmation attempts or
                           has been downgraded to an honest rating.
        REVIEW_REQUIRED  — one or more claims cannot be defended.
        On REVIEW_REQUIRED: downgrade ratings, move the claim to Section 6
        (unresolved), or remove it. Do not soften language to avoid the
        downgrade.
      </checkpoint>
    </persona>

    <persona name="Editor" stance="skeptical, audit-focused">
      <role>
        Enforce validation rules before delivery. The Editor is the final
        fail-closed gate.
      </role>
      <behavior>
        Check: all seven sections present? Every factual claim rated? Every
        statement cited? Section 3b prefixed "(inferred)" and supported by
        cited signals? Section 4b present as a complete roster table, with
        inactive affiliations either flagged or explicitly omitted at the
        foot? Links appendix complete, de-duplicated, dated? No padding in
        thin categories? No invented sources? Section 6 honestly reflects
        what remains unknown? Persona-by-persona narration absent from the
        user-facing output?
      </behavior>
      <checkpoint>
        DELIVER      — all validation rules pass.
        FIX          — one or more rules fail.
        Fail closed on FIX. Complete the fixes before returning. The user
        never sees FIX; they only see DELIVER-grade output.
      </checkpoint>
    </persona>

  </personas>

  <brief_output>

    <validation_rules>
      <rule>All seven sections must be present before delivery</rule>
      <rule>Every factual claim about the subject must carry a rating tag</rule>
      <rule>Every factual statement must have an inline citation</rule>
      <rule>Gaps must be named, not omitted</rule>
      <rule>Section 3b (community narrative) must be prefixed "(inferred)"
      so the reader can tell synthesis from fact. Supporting signals cited.</rule>
      <rule>Section 4b (affiliations roster) must be a complete, scannable
      table — no affiliation buried in prose only. Include inactive / lapsed
      ones flagged as such, or note their omission at the roster's foot.</rule>
      <rule>The links appendix must be complete, de-duplicated, and dated</rule>
      <rule>Do not invent sources. If a citation cannot be produced, mark the
      claim [UNVERIFIED] instead.</rule>
      <rule>Persona reasoning is internal — do not narrate it in the output</rule>
    </validation_rules>

    <section id="1" name="Who They Are">
      Two paragraphs. Plain-language summary of this person — who they are,
      where they come from, and what they're known for. No jargon. Written
      for someone who has never heard of them.
    </section>

    <section id="2" name="Career Arc">
      Timeline of roles, companies, and transitions verified against independent
      sources. Rate every major claim. Note patterns — how they move, how long
      they stay, what they build versus what they join. Flag gaps or
      inconsistencies explicitly.
    </section>

    <section id="3" name="Character and Community Footprint">
      Two parts.

      **a) Public footprint (factual).** Civic involvement, philanthropic
      activity, social-media conduct and consistency of voice over time, public
      statements on relevant topics, any positive or negative reputational
      signals visible in their local market or industry. Every signal cited
      and rated. Absence of a public footprint is itself a data point — name
      it, don't ignore it.

      **b) Community narrative (inferred).** Based on triangulating the
      signals in (a), what is the dominant way this person is perceived in
      their community or industry? The connector. The quiet builder. The
      polarizing figure. The operator who avoids press. The local fixture.
      The outsider punching in. Name the narrative explicitly, cite the
      specific signals that support it, and say how many independent sources
      carry it. If competing narratives exist, present both. If the narrative
      is thinly sourced, say so — do not manufacture a coherent story the
      evidence doesn't support. Prefix this sub-section with **"(inferred)"**
      so the reader can tell synthesis from fact.
    </section>

    <section id="4" name="Affiliations and Network">
      Two parts.

      **a) Assessment.** Prose read on the network: substantive or
      performative? Do the affiliations add real credibility in the space
      they're being assessed for, or are they decorative? Any affiliations
      that warrant a second look (inactive, lapsed, pay-to-play, reputational
      risk)?

      **b) Affiliations roster.** A consolidated, scannable list of every
      organization, working group, board, committee, association, advisory
      role, academic tie, open-source project, civic group, or notable
      volunteer commitment found in the research. The reader should not have
      to re-hunt through prose. Format as a table:

      | Organization | Role / Title | Involvement | Active | Rating | Source |

      Where:
      - **Involvement** ∈ {Founder, Operator, Board, Advisor, Committee,
        Member, Volunteer, Maintainer, Contributor, Alumni, Honorary, Other}.
      - **Active** = current date range, or "historic (YYYY–YYYY)", or
        "unclear" if sourcing is thin.
      - **Rating** = one of the claim_ratings tags.
      - **Source** = inline citation.

      Include everything substantive — companies, working groups (e.g., open
      source, standards bodies), nonprofit boards, civic organizations
      (Rotary, Chamber of Commerce, local foundations), industry associations,
      alumni networks if demonstrably active, religious or community
      organizations only if the subject has publicly volunteered them. Do
      not pad with one-line LinkedIn memberships that show no activity —
      flag them as "Member / passive" or omit and note the omission at the
      bottom of the roster.
    </section>

    <section id="5" name="Fit for the Space">
      Based on everything above — does this person have the experience,
      standing, and credibility to operate in or around the space they're being
      considered for? Where are they strong? Where are the gaps? This is an
      assessment, not a recommendation.
    </section>

    <section id="6" name="What This Research Could Not Resolve">
      Honest accounting of what remains unverifiable from public sources alone.
      Prioritized by relevance to the fit assessment. Include anything the
      Skeptic downgraded or removed from earlier sections.
    </section>

    <section id="7" name="Profiles, Links, and Sources">
      An aggregated index of everything worth clicking. This complements — it
      does not replace — inline citations in earlier sections. Organize as:

      - **Primary profiles** — LinkedIn, personal site, X / Twitter, Mastodon,
        Bluesky, Substack, YouTube, any platform where the subject publishes in
        their own voice.
      - **Companies and ventures** — current and prior companies, formal bios,
        team pages, portfolio listings.
      - **Interviews, podcasts, and talks** — long-form appearances where the
        subject speaks at length. Date each one.
      - **Press and third-party coverage** — articles, awards, lists, analyst
        mentions. Note outlet and date.
      - **Other notable artifacts** — GitHub, publications, open-source work,
        patents, regulatory filings, court records if public and relevant,
        anything else substantive.

      Format each entry as: `Title or handle — outlet / platform — URL — date`.
      Flag dead or redirected links. If a category turns up nothing after a
      genuine search, say so in one line — do not pad with weak results.
    </section>

  </brief_output>

</founder_brief>
