---
name: 01-create-once-distribute-forever
description: Run the moment a piece of work is finished. Trigger - "run 01" plus the asset and where it already lives.
---

# TASK 01 — CREATE ONCE, DISTRIBUTE FOREVER

## INPUT
The finished asset. Where it already lives. Who it
was written for. Anything in it that came from a
client.

## PROCESS
1. Find the SPINE. One sentence the whole thing
   rests on, that somebody could disagree with. If
   it cannot be said in one sentence, the asset is
   not finished. Say so before planning anything.
2. Atomise. Break the asset into claims that stand
   on their own: a position, a mechanism, a number
   I published, a reframe, a mistake. Every slice
   must make sense to someone who never clicks.
3. Silence check every slice against task 05.
   Argument out, inventory in, held items logged.
4. Place, do not post. Each slice gets a channel
   chosen because that argument already happens
   there, not because we have an account.
5. Pick the risky one. One slice goes somewhere
   uncomfortable or says the unpopular half out
   loud. That is the 20 to 30%.
6. Schedule across weeks, not days. Launch week is
   the smallest part of the plan.
7. Set the recycle date, 90 days out, with the note
   on what evidence will need refreshing.

## OUTPUT
SPINE, one sentence. SLICES as a table: slice,
channel, format, week, native or link. THE RISKY
ONE. HELD BACK with review dates. RECYCLE DATE.

## RULES
- No slice that only makes sense if you clicked.
- "Check out my new post" is not a format.
- Reddit and forums: answer the question, no link,
  unless the thread asked for one.
- Never a client result the client has not
  published, in any slice, on any channel.

## EDGE CASES
- The asset is a client case study. The strongest
  slices are the ones we cannot use. Slice the
  mechanism, never the account. If nothing
  survives, call it what it is, a sales asset, and
  route it to sales enablement instead.
- The asset is a talk I have not given yet.
  Distributing it early gives away the talk.
  Distribute the question before the date, stack
  the answer slices for the week after, and hold
  the best line for the room.

## GOOD LOOKS LIKE
"The spine is 'the unit of retrieval is a passage,
not a page.' Everything else is support. Nine
slices, six weeks, and one of them goes on Reddit
with no link, because you just wrote a piece about
earning citations. Earn one."

---
name: 02-discovery-to-scoped-proposal
description: Run after a discovery call. Trigger - "run 02" plus the call notes, who was in the room, and what they run today.
---

# TASK 02 — DISCOVERY CALL TO SCOPED PROPOSAL

## INPUT
Call notes or transcript. Who was on it and who was
not. What they asked for. What they are running
today. What they think the problem is.

## PROCESS
1. Separate the ASK from the PROBLEM. The ask is
   usually "more content" or "we need GEO". Write
   the problem in one sentence in their words, then
   the constraint in mine.
2. Name their constraint honestly. If the
   constraint is production, say we are not the
   cheapest way to solve it. Production is free
   now. Say that in the proposal.
3. Map the room. Who signs, who blocks, whose job
   changes if this works. If someone was missing
   from the call, name them and ask for them.
4. Scope to one outcome, not a channel list. The
   channels are how, not what. Every channel in
   scope carries its own distribution plan or it
   comes out.
5. Sequence it. What has to be true first, and what
   we deliberately do not start in month one.
6. Price from [RATE CARD]. Ask once. Never infer a
   number from scope, hours or another deal.
7. Write the disqualify paragraph. Who this is
   wrong for, in plain words. It closes more than
   it costs.

## OUTPUT
THE PROBLEM in their words. THE CONSTRAINT in mine.
THE OUTCOME. SCOPE by sequence, with the
distribution plan attached to each line. WHAT WE
ARE NOT DOING. PRICE from [RATE CARD]. WHO THIS IS
WRONG FOR.

## RULES
- Never invent, imply or anchor a price.
- Never promise a ranking, a citation or an AI
  Overview placement. Promise the work and the
  measurement.
- No named client result unless that client
  published it themselves.
- No channel in scope without a distribution plan
  on the same line.

## EDGE CASES
- They want AI visibility and do not rank at all.
  Retrieval still runs through the index, so the
  floor comes first. Put the boring sequencing in
  the proposal even though it makes the sale slower
  and smaller. Selling them the ceiling first is
  how we lose the account in month six.
- Procurement asks for a rate card or a rate
  comparison. There is no rate card. Send scope and
  outcome, offer a call, and say plainly that we
  price the problem. Do not manufacture a sheet to
  satisfy a form.

## GOOD LOOKS LIKE
"You asked for 40 posts a quarter. Your problem is
not production, production is free now. You have
[N] pages and nothing cites any of them. We are not
the cheapest way to buy 40 posts and I am not going
to pretend otherwise."

---
name: 03-rfp-teardown
description: Run when an RFP or a client brief lands. Trigger - "run 03" plus the document, the deadline, and who sent it.
---

# TASK 03 — RFP AND BRIEF TEARDOWN

## INPUT
The RFP or brief. Deadline. Who sent it and how it
arrived. Anything known about the incumbent or the
consultant running it.

## PROCESS
1. Pull the STATED criteria. The rubric, the
   weightings, the mandatory requirements, exactly
   as written.
2. Pull the REAL criteria. Who wrote each section.
   Which requirement is lifted off a vendor's
   website. What the deadline says about how long
   this has been decided.
3. Find the wired clause. The requirement only one
   vendor can meet. Name it internally, never
   externally.
4. Find the question they did not ask. It is almost
   always distribution. They scoped production
   because production is what they know how to
   count.
5. Decide BID or NO BID with the reason, before a
   word of the response gets drafted.
6. If bid, three points of difference, and one of
   them reframes the RFP itself.

## OUTPUT
STATED CRITERIA. REAL CRITERIA. WIRED CLAUSE, if
any. THE MISSING QUESTION. BID or NO BID with one
sentence of reason. THE REFRAME, in the first 100
words of the response.

## RULES
- No bid is a real output. Say it out loud, do not
  bury it in a hedge.
- Never bid to fill a gap in the pipeline.
- Never name the suspected incumbent in anything
  client facing.
- Never answer a requirement for work we do not
  actually do.

## EDGE CASES
- The scoring rewards deliverable volume. That is a
  production frame and we lose on it by design.
  Either bid the reframe and eat the scoring hit,
  or no bid. Padding the deliverable count to score
  well wins a contract we will hate.
- An agency search consultant wrote it and the real
  buyer never reads our words. Answer the rubric in
  the body, because the consultant is the filter,
  and put the reframe in the first 100 words, which
  is the only part a human still reads.

## GOOD LOOKS LIKE
"Requirement 4.2 asks for a named GEO platform
partnership. That is not a requirement, that is
somebody's incumbent with a logo on it. No bid. I
would tell them why if they asked."

---
name: 04-argued-post
description: Run for the long-form piece that has to carry a position. Trigger - "run 04" plus the claim, the evidence, and who it annoys.
---

# TASK 04 — THE ARGUED POST

## INPUT
The claim. The evidence I actually have. What
changed to make this worth saying now. Who this
annoys.

## PROCESS
1. Write the claim in one sentence before writing
   anything else. A topic is not a claim. If nobody
   can disagree with it, it is a summary and it
   does not need me.
2. Name who this is wrong for. If the answer is
   nobody, go back to step 1.
3. Evidence: my own data, my own test, or a source
   with a link. No borrowed stat without a link and
   a date.
4. Build for passages. Answer first in every
   section, each section self-contained, no
   pronouns pointing backwards. If a model lifts
   one 200 word chunk out of this, it should still
   be a complete argument with my name attached.
5. Include the part that cuts against me. A post
   with no cost to the author is an ad.
6. Silence check. Argument out, inventory in, and
   name what is held.
7. Close on the imperative, not the summary.

## OUTPUT
The post. The one sentence claim, stated separately
so it can be reused. The slice list, handed
straight to task 01.

## RULES
- No listicle scaffolding, no "in today's
  landscape", no bait question at the end.
- No statistic without a link and a date.
- Never dress a prediction as a finding.
- Never soften a claim to fit weak evidence. Get
  better evidence or cut the claim.

## EDGE CASES
- The honest version undercuts something we sell.
  Publish it, then hand sales the sentence that
  turns it into the reason to hire us. The
  alternative is a sales page with a blog URL,
  which is what everyone else publishes.
- The claim is right and the only proof is client
  data. Rebuild the proof on our own properties or
  from public data. Do not publish a claim that
  only holds up in a room under NDA.

## GOOD LOOKS LIKE
"Here is the part that costs me money to write:
half of what agencies sold as on-site SEO for
twenty years now competes for 49% of the page.
Mine included. Go rebuild for the other 51% or go
explain the invoice."

---
name: 05-lasagna-check
description: Run before anything goes public, or when a win happens and the instinct is to post it. Trigger - "run 05" plus the thing I want to post.
---

# TASK 05 — THE LASAGNA CHECK

## INPUT
The thing I want to post. What it would prove. Who
it involves.

## PROCESS
1. Sort it. ARGUMENT is a position, a mechanism, a
   reframe, a test that failed, a mistake I made.
   INVENTORY is a live client result, a scope, a
   channel that is quietly working, a prompt chain,
   a margin, a headcount, a deal. Argument
   publishes. Inventory holds.
2. Ask what it recruits. Argument recruits clients.
   Inventory recruits competitors into the exact
   channel that is working for us right now.
3. The screenshot test. If the post is a number
   going up, it fails. Ask what a reader does
   differently on Monday. If the answer is nothing,
   hold it.
4. Ask what it costs the person in it. A client
   result belongs to the client until they say it
   does not.
5. Time it. Inventory becomes argument once the
   edge is gone or the client publishes first. Log
   the item with a review date, do not just drop
   it.
6. If holding, write the one line in the piece that
   says a version is held. Silence with no
   acknowledgement reads as having nothing to say.

## OUTPUT
PUBLISH or HOLD, with the reason in one sentence.
THE HELD ITEM logged with a review date. THE
ADJACENT POST that survives the filter.

## RULES
- Never publish a client's result before the client
  does.
- Never publish the channel that is currently
  working.
- A held item is logged, never quietly dropped.
- This rule is a filter, not an excuse to go
  quiet. The lead engine is public and it has to
  keep running.

## EDGE CASES
- The win is the strongest proof we have and there
  is a live deal that needs it. It goes in the
  proposal, not the feed. Sales asset, not
  distribution asset, and the two are not the same
  thing even when the words are identical.
- Somebody else publishes the tactic we were
  holding. The edge is gone, so the hold expires
  the same day. Publish the deeper version, the one
  that says why it works, and do not claim we were
  first. Nobody believes that claim anyway.

## GOOD LOOKS LIKE
"Hold. That is a screenshot of a line going up, and
it teaches a competitor exactly which subreddit to
go stand in. The post that survives is the one
about why that channel works, with our name off
it."

---
name: 06-client-qbr-serp-reset
description: Run before a QBR where the numbers moved because the page changed. Trigger - "run 06" plus the account, the commitments, and who is in the room.
---

# TASK 06 — THE QBR ON A SERP THAT MOVED

## INPUT
The account. What we committed to. What the numbers
did. What changed on the page itself. Who is in the
room and what each of them is measured on.

## PROCESS
1. Open on the decision they have to make, not the
   dashboard. The dashboard is evidence, not an
   agenda.
2. Name our miss first. Whatever we got wrong this
   quarter goes before anything about the market.
   Leading with the market reads as an alibi and
   they will hear it that way.
3. Then the structural part, plainly and with the
   source: organic is 49% of the SERP, AI Overviews
   take 23%, and the rest belongs to features
   nobody writes their way into.
4. Reset the metric honestly. From rankings and
   sessions to share of AI recommendation and
   citations. Say out loud what we can measure,
   which is prompt panel testing, AI referral
   segments, crawler activity in logs, and what
   buyers say on sales calls. Say what we cannot,
   which is a position number.
5. Make the reallocation ask with the tradeoff
   named. What gets less money so this gets more.
6. Name one thing we are stopping.

## OUTPUT
THE DECISION. WHAT WE GOT WRONG. WHAT MOVED UNDER
US, with sources. THE NEW MEASURE and its limits.
THE REALLOCATION with its tradeoff. THE THING WE
ARE STOPPING.

## RULES
- Never present a redefined metric as a win.
- Never promise a citation or an Overview
  placement.
- No benchmark borrowed from another client, ever,
  named or unnamed.
- Measurement caveats go in the room, out loud, not
  in a footnote on slide 30.

## EDGE CASES
- Traffic is down and our work was good. Say both,
  in that order. The number is down, here is what
  we controlled, here is what moved. Reversing that
  order costs the account even when the argument is
  correct.
- The champion sold this internally as rankings and
  is compensated on sessions. Changing the metric
  costs them personally. Write them the internal
  memo version, in their language, that lets them
  change the story upward without looking like they
  bought the wrong thing.

## GOOD LOOKS LIKE
"You bought a page that no longer exists. Organic
is 49% of it now. Before I explain that, here is
what we got wrong ourselves this quarter, because I
am not going to blame Google for our Q2."

---
name: 07-curiosity-screen
description: Run on every candidate, at every stage of the loop. Trigger - "run 07" plus the role, the materials and any work sample.
---

# TASK 07 — HIRING FOR CURIOSITY

## INPUT
The role. The candidate's materials. Any work
sample. Notes from whoever has spoken to them.

## PROCESS
1. First pass with the credential line covered.
   Agency logos and degrees predict very little in
   a market that reset this year.
2. Look for digging. Did they get into the data
   before they pitched the idea. Do they know where
   the audience actually is, forums, Reddit, Slack
   communities, or are they quoting a secondhand
   summary of it.
3. Run the "why did it work" test. Ask about
   something that worked for them and listen for
   the mechanism. "It went viral" is not an answer,
   it is a description.
4. Look for self-directed learning. Something they
   learned that nobody asked them to learn.
5. Look for the disagreement signal. Have they ever
   told a manager the plan was wrong, and what
   happened next.
6. Give a live problem, not a take-home essay.
   Score the questions they ask before they answer.
7. Taste check. Show two pieces of work, ask which
   is better and why. Production is free. Taste is
   what we are hiring.

## OUTPUT
A scorecard on the five signals, each with verbatim
evidence from the candidate. The one question for
the next round. A hire or no-hire lean with the
reason in a sentence.

## RULES
- A credential is context, never a signal. It never
  scores.
- An outcome is not evidence without the mechanism.
- No proxy that quietly penalises someone for
  having the wrong logos or no logos.
- Compensation stays out of the screen. The band is
  [PRIVATE] and it is a separate conversation.

## EDGE CASES
- Strong mechanism, thin portfolio, career changer.
  This is the exact profile the screen exists to
  catch. Weight the live problem heavily and do not
  let a short work history quietly override what
  they just demonstrated.
- Excellent credentials and fluent answers that are
  all outcomes. Go one level down twice, politely.
  If the mechanism never arrives, it is a no, and
  it will be the least comfortable no in the loop
  because everyone else in the room liked them.

## GOOD LOOKS LIKE
"She could not name an agency on her CV that would
impress anyone. She could tell me which subreddit
the buyers actually live in, and why the top
comment outperformed the post it was under.
Advance her."

---
name: 08-week-plan-two-companies
description: Run Sunday night. Trigger - "run 08" plus the calendar, both pipelines, publishing commitments and anything on fire. Reads on a phone.
---

# TASK 08 — WEEK PLAN ACROSS TWO COMPANIES

## INPUT
Calendar. Both pipelines. Publishing commitments.
Speaking dates and travel. Open hiring loops. Board
commitments. Anything on fire.

## PROCESS
1. Fixed points first. Talks, recordings, client
   calls, publish dates, board obligations. They do
   not move and everything else fits around them.
2. Three outcomes, one each: Foundation,
   Distribution.ai, and the public engine. An empty
   lane gets named as the risk, never padded to
   look full.
3. Write the FOUNDER ONLY list. Discovery calls,
   proposal scope and price, the final call on a
   hire, and the position inside an argued piece.
   Nothing on that list gets delegated this week.
   Anything not on it should be.
4. Travel and speaking weeks: writing moves
   earlier, never later. Name what gets drafted
   before wheels up.
5. Decline and defer, with the script written out.
6. The avoided thing. It is usually a pricing
   conversation, a client conversation that is
   overdue, or the piece that would commit us
   publicly to the reposition.

## OUTPUT
Half a page, phone readable. FIXED POINTS. THREE
OUTCOMES. FOUNDER ONLY. DECLINE AND DEFER, with the
script. THE AVOIDED THING.

## RULES
- Outcomes state what is true on Friday, not what
  gets worked on.
- Maximum three. Name what got parked.
- Never a week with no publishing and no selling.
  One of the two happens or the plan is wrong.
- Never plan a conference week like a normal week.

## EDGE CASES
- Both companies have a live deal cycle in the same
  week. One gets the founder, the other gets a
  date. Say which out loud in the plan. Splitting
  the week evenly is how both of them slip.
- A conference collides with a publish date. Move
  the publish earlier or drop it on purpose with a
  note saying so. The lead engine never goes quiet
  by accident. Quiet is a decision here, not a
  default.

## GOOD LOOKS LIKE
"AVOIDING: the pricing conversation on [ACCOUNT].
Third week now. Every week it stays open you are
selling last year's scope at last year's number,
and you are the only person who can close it."