← Pilot Hub
Week 6 · Run an Audit

AUDIT PICKER

Three modes, one Skill. Picked on purpose — for the artifact in front of you, at the stakes that actually matter.

TL;DR — the one rule

Light every time. Standard before anything leaves the building. Full for board, funder, external publication. That's the whole picker. The rest of this page tells you when each rule applies — and which panel to ask for.

01The Three Modes

The adversarial-audit Skill comes in three modes. The names sort by depth and time — not by quality. A light-mode audit on a counselor email is not a worse audit than a full-mode pass on a Gates LOI. It's the right audit for the work in front of you.

Fast

Light

3 agents · Stage 1 only

Rapid triage. Verdict, score, critical flags. 600–900 words back. Runs in roughly 90 seconds. The mode you run every time before something leaves your hands.

Use for

  • Counselor and partner emails
  • Internal Slack updates and team memos
  • LinkedIn drafts and short blog posts
  • Anything you'd previously have sent unchecked

Don't use for

  • Anything board, funder, or executive-facing (upgrade)
  • Multi-page documents (upgrade)
Default

Standard

3–4 agents · Stage 1, then Stage 2 on request

The default for anything about to leave the building. Stage 1 triage first. Full panel breakdown only if the triage indicates work needed. Five to seven minutes when both stages run.

Use for

  • Quarterly partnership updates
  • Briefing decks and weekly status writeups
  • First drafts of grant LOIs and concept notes
  • Board prep memos (not the board memo itself)
  • Op-eds and longer thought-leadership posts

Don't use for

  • A 90-second light pass would have caught it (downgrade)
  • The single biggest funder proposal of the year (upgrade)
Deep

Full

4 agents · all three stages · literal rewrites

The whole skill, end to end. Triage, full panel breakdown, and per-agent literal rewrites of flagged sections. Ten minutes per artifact, sometimes more. Reserved for the work where being wrong is costly.

Use for

  • The Gates LOI or large federal proposal
  • Board-facing strategy memos
  • External publication carrying DC CAP's name
  • Anything signed by Eric
  • Any artifact a regulator or auditor might read

Don't use for

  • Routine work — full is wasted there
  • Anything you'd rather just send (run standard first)

02How to Pick in 30 Seconds

Five questions, asked in order. The first "yes" gives you your mode. If you reach the end without a yes, pick Standard.

1

Is this a routine internal artifact — Slack, email to a teammate, calendar invite description?

If yes → Light. Examples: a counselor follow-up, a partner check-in, a LinkedIn draft, a short blog post. Run light every time something is about to leave your hands.

2

Is this board-, funder-, or executive-facing, where being wrong is costly?

If yes → Full. Examples: the Gates LOI. The Q3 board memo. A funder report. The capstone deliverable. Anything signed by Eric.

3

Does this carry the org's name into public — external publication, op-ed, press, regulator-facing material?

If yes → Full. The downside risk of an unverified claim in public is higher than the time cost of a full-mode pass.

4

Is this everyday work about to leave the building — partnership updates, briefing decks, grant first drafts, board prep memos?

If yes → Standard. This is most of what passes from your desk to someone else's. Standard is the right answer roughly 70% of the time.

5

Still not sure?

Pick Standard. If the triage comes back clean, you've spent 90 seconds. If it surfaces a convergent flag, you've avoided a problem before it shipped. The picker improves with reps.

03The Convergence Rule

When two lenses flag the same line cold, the line is the problem.

Convergence is the single most important signal in any audit triage. A single lens flagging a line is a recommendation. Two lenses flagging the same line, for different reasons, with no knowledge of each other's notes — that's the math.

Read the triage for convergent flags first. They go to the top of the fix list. Everything else is recommended polish. If there are no convergent flags, the verdict can be "safe to send" even with single-lens flags still on the page.

04Which Panel for Which Artifact

The Skill picks the panel automatically based on the artifact. The reference below shows you what to expect. If your artifact doesn't fit cleanly, ask for the Adversary panel — it works on any artifact and runs the general argument, logic, and language checks.

Grant · LOI · Funding Proposal

Funding Panel

  • Program OfficerWould I advance this or decline it?
  • Evidence ReviewerCan I verify every claim?
  • Budget AnalystDoes the budget prove they can do this?
  • Persuasion EditorReads like a winner, or a compliance exercise?
Strategy · Plan · Memo

Strategy Panel

  • Logic AuditorIf I challenge every "why this not that," do answers hold?
  • Evidence CheckerWhat's the evidence this strategy will work?
  • OperatorCan they actually pull this off with what they have?
  • Stakeholder ReaderDo partners and funders see themselves in it?
Board · Executive · Funder Deck

Decision Panel

  • Board VeteranCan I make a decision from this?
  • Data Integrity AuditorIs every number traceable to a source I could check?
  • Strategy TranslatorDoes this tell a coherent story?
  • Ask EvaluatorAm I ready to say yes, or do I need more?
Thought Leadership · Op-ed · LinkedIn

Content Panel

  • EditorIs there an argument, or just observations?
  • Fact-CheckerWhich claims survive a challenge?
  • Audience ProxyDid I learn something, or read someone thinking out loud?
  • Originality CriticHave I read this under a different byline?
Any artifact · default fallback

The Adversary

  • LogicianIf I challenge every "therefore," which survive?
  • Claim AuditorWhich claims is the author hoping I won't check?
  • Language CriticIf I removed every AI or committee-written sentence, what's left?
  • Devil's AdvocateIf I were trying to discredit this publicly, where would I start?

05Mode by Data Tier

The data tier sets the audit floor, not the ceiling. Tier 2 work can run full mode if the stakes warrant. Tier 4 work can run standard if a single-lens flag matters. But the floor is the floor.

Data tier Examples Floor What goes in the record
Tier 4 — Public LinkedIn drafts, website copy, public press releases, conference remarks Light Verdict only. No record kept beyond your own notes.
Tier 3 — Internal Strategy Internal strategy docs, board prep memos, donor briefings, planning materials Standard Verdict and convergent flags noted. Triage saved with the draft.
Tier 2 — Sensitive Partner-specific operational details, individual scholar mentions, sensitive comms Standard or Full Full triage in the record. Convergent flags resolved before send.
Tier 1 — Restricted Cannot be loaded into the audit without scrubbing — extract or summarize first Full on scrubbed version Triage saved with the scrubbed version. Original never enters the Skill.

06Two Anti-Patterns to Avoid

Anti-pattern 1 — Skipping light because it "sounds basic"

"It's a short email, I don't need an audit." Light mode takes 90 seconds. It catches the convergent flags you stopped seeing three drafts ago — the AI-style opener, the unsourced stat, the soft close that asks for nothing. The team that skips light ships those flags every time.

The fix: light every time. It's a habit, not a decision.

Anti-pattern 2 — Running full mode on everything

"I want to be thorough." Full mode takes ten minutes per artifact and produces literal rewrites you didn't ask for. Run it on routine work and the audit feels like overhead. Within a week, the team stops running audits at all.

The fix: match the mode to the stakes. Full earns its time on the board memo, the funder proposal, the external publication. Light earns its time on everything else.

07Your Capstone Audit Cadence

One sentence in your capstone scope. Fill the blanks. That sentence is the Diligence stub for Section 1 — and the peer review in Week 7 pulls from it directly.

Template

"The audit cadence for this build is ______ mode, run by ______, before ______ ships."

Three filled examples

  • Counselor outreach Project: "The audit cadence is light mode, run by the success coach drafting, before each batch of counselor emails ships."
  • Quarterly partner update Skill: "The audit cadence is standard mode, run by the comms lead, before the quarterly update goes to all 13 partners."
  • Grant proposal pipeline: "The audit cadence is full mode, run by the Strategic Auditor agent, before any proposal goes to Eric."