Capstone: The 3-Site Portfolio

Overview

This capstone exists because of Rule 4: launch three sites first. Three is not a compromise between one and ten. It is the smallest number that lets you compare, and the largest number you can run honestly while still learning what works. One site teaches you nothing about whether your judgement or your niche produced the result. Ten sites, launched before any gate has been passed, multiplies an unproven process by ten and calls it a portfolio.

The three Phase 1 sites are fixed: SMB Finance and Banking (Weekly Cash Flow Brief), B2B SaaS and AI Tools (Weekly Stack Brief), and SMB Cybersecurity (Weekly SMB Security Brief). Each gets a 12-asset MVP, a newsletter, and one tool — never thirty pages.

The hardest lesson here is not analytical, it is behavioural. The gates were designed in Course 0, when you had no emotional stake. By Month 9 you will have one. The entire value of a pre-committed kill criterion is that it was written by someone who was not yet in love with the site, and the only way it works is if you refuse to renegotiate it now.

12.1 How do you sequence three launches?

What is it?

A staggered launch plan that brings three sites live across roughly 90 days — deliberately not simultaneously — so each launch teaches something that improves the next one.

Why it matters

Simultaneous launches feel efficient and destroy the learning. If all three go live in the same week, every mistake in your process is baked into all three at once, and you cannot tell which outcome came from the niche and which came from your execution. Staggering converts site two and site three into beneficiaries of what site one taught you. It also protects the thing most likely to fail: your own capacity. Three newsletters starting in the same week is three weekly deadlines starting in the same week.

How to do it

1. Rank the three sites by the Course 2 displacement-adjusted score, and launch the strongest first — it deserves your least-experienced execution the least.
2. Give site one a 30-day head start: 12-asset MVP live, newsletter shipping weekly, one tool working.
3. Write a launch retrospective for site one before site two goes live. If nothing is written, site two is not permitted to launch.
4. Launch site two at day 30–45, applying the retrospective. Reuse the component library from Course 4 rather than rebuilding.
5. Launch site three at day 60–90 only if the first two newsletters have not missed an issue. A missed issue is a capacity signal, and it means you cannot yet support a third.
6. Freeze the portfolio at three. No fourth site before the Month 9 gate, regardless of how well things appear to be going.

Tools needed

Your Course 2 niche scorer output, the Course 4 component library, and launch checklists.

Lab 12.1

Build the 90-day sequence for your three sites with named dates, the 12 assets per site listed by title, and the explicit proceed-gate between each launch. Then write down what would have to be true for you to not launch site three — and check whether you would actually honour it.

Artifact

Portfolio launch plan
  Site 1: ___ | displacement-adjusted score ___ | launch date ___
    12 assets: 2 pillar ___ | 3 money ___ | 2 tools ___ | 2 data ___ | 3 support ___
    Newsletter: ___ | first issue date ___
  Site 2: ___ | launch date ___ | proceed-gate: site 1 retrospective written? ___
  Site 3: ___ | launch date ___ | proceed-gate: zero missed issues on sites 1-2? ___
  Portfolio freeze date: ___
  Condition under which site 3 does NOT launch: ___

Common mistakes

• Launching all three in one week because the build felt efficient, then discovering three weekly deadlines land on the same day.
• Launching the weakest niche first "to practise" — wasting your best-executed launch on your worst opportunity.
• Skipping the retrospective, which is the only mechanism that makes staggering worth anything.

Pro astuces

• The proceed-gate that matters most is the missed-issue check. Newsletter cadence is the earliest honest signal of whether your capacity assumptions were fiction.
• Reusing the component library should make site two roughly 40% faster to build. If it does not, your Course 4 library was not actually portable, and that is worth fixing before site three.

12.2 What does the Month 3 gate test?

What is it?

The first pruning gate. At Month 3 you judge execution only — did you ship what you said you would ship — and explicitly not results, because three months is too early for demand evidence in any niche worth entering.

Why it matters

Most portfolios die of a quiet execution failure that nobody names for a year. The Month 3 gate exists to catch the difference between "the thesis was wrong" and "we did not actually do the work". These require opposite responses: a wrong thesis means prune or pivot, while incomplete execution means finish the work before drawing any conclusion. Judging results at Month 3 produces the worst decision available — killing a sound thesis that was never given a real test.

How to do it

1. Check completeness against the plan: are all 12 assets live per site, or not?
2. Check newsletter cadence: issues shipped versus issues promised. This is a hard count, not an impression.
3. Check the tool: is it live, working, and does it save state for the visitor?
4. Check the legal and compliance gate from Course 4: disclosure, ownership, contact, and no unverified claims past expiry.
5. Record the real cost to date against the 12-month model, and the variance. Variance above 25% is itself a finding.
6. Issue one verdict per site — on track, behind, or not executed — and for anything not "on track", state the specific work required and a date.

Tools needed

The pruning scorecard, your 12-month cost model, and the Course 4 launch gate checklist.

Lab 12.2

Score all three sites against the five Month 3 criteria and write a one-paragraph verdict for each. Where a site is behind, name the work and the date — not an intention. Then record whether the shortfall was capacity, cost or attention, because that answer predicts Month 6 better than any traffic number.

Artifact

Month 3 gate record
  Date held: ___   Sites reviewed: 3
  Per site: assets live ___/12 | issues shipped ___/___ | tool live Y/N
            compliance gate PASS/FAIL | cost variance ___%
  VERDICT: on track / behind / not executed
  Required work + date (if not on track): ___
  Root cause of any shortfall: capacity / cost / attention
  NOTE: results deliberately NOT judged at this gate
  Next gate date (Month 6): ___

Common mistakes

• Judging traffic or revenue at Month 3 and killing a thesis that was never tested.
• Marking a site "on track" when 6 of 12 assets are live, because the remaining six are "nearly done".
• Treating a missed compliance item as a minor administrative issue rather than a hard fail.

Pro astuces

• Hold the gate on a booked date with the scorecard open. A gate held "around month 3, informally" is not a gate.
• The root-cause field is the most predictive line in the whole record. Capacity shortfalls almost never resolve themselves by Month 6 without a scope cut.

12.3 What does the Month 6 gate test?

What is it?

The demand gate. At Month 6, with execution established, you test the falsifiable thesis written in Course 0: is there evidence of return demand — an audience that comes back without an intermediary sending them?

Why it matters

This is the gate that answers the governing question with data instead of hope. Rankings and impressions are not the evidence here; they can rise while the asset becomes more fragile. What matters is whether anyone chose to return: newsletter retention, direct and branded arrivals, tool reuse. A site with growing search traffic and zero return demand is a rented audience, and the Month 6 gate is where you find that out while you still have runway to change it.

How to do it

1. Restate the Course 0 thesis for each site verbatim. If you cannot find it written down, that is the finding.
2. Measure return demand: subscriber retention after 4 issues, direct plus branded arrivals as a share of total, and repeat tool usage.
3. Measure the AI-surface position from Course 9: citation presence and the impressions-stable/clicks-falling pattern.
4. Compare against the thesis prediction. State plainly: supported, unsupported, or untested.
5. Check the four concentration ratios from Course 11 — at Month 6 they are usually worse than expected and easiest to correct.
6. Issue a verdict: fund, hold flat, or pivot. Killing is available but usually belongs at Month 9 unless compliance or thesis failure is total.

Tools needed

Your Course 9.6 dashboard, the AI-surface tracker, newsletter analytics, and the pruning scorecard.

Lab 12.3

For each site, place the Course 0 thesis and the six-month evidence side by side and mark it supported, unsupported or untested. Then answer the governing question in one sentence per site, using only measured return-demand evidence. If a site cannot be answered without hypotheticals, that is your answer.

Artifact

Month 6 gate record
  Per site:
    Course 0 thesis (verbatim): ___
    Retention after 4 issues: ___ | Direct+branded share: ___%
    Repeat tool users: ___ | AI citations on core topics: ___
    Concentration: affiliate ___% | site ___% | source ___% | sponsor ___%
    Thesis status: SUPPORTED / UNSUPPORTED / UNTESTED
    Governing-question answer (measured evidence only): ___
    VERDICT: fund / hold flat / pivot
  Next gate date (Month 9): ___

Common mistakes

• Substituting impressions or rankings for return demand, which measures the intermediary's behaviour rather than the audience's.
• Rewriting the thesis after seeing the data so that the result appears to support it.
• Marking a site "untested" as a way of postponing an uncomfortable verdict twice in a row.

Pro astuces

• Unsubscribe rate per issue is a more honest product signal than list growth. Growth can be bought; retention has to be earned.
• "Untested" is a legitimate verdict exactly once per site. The second time, it means the test was never designed.

12.4 How do you run the Month 9 keep/prune/kill gate?

What is it?

The decision gate. At Month 9 each site receives exactly one verdict — keep, prune or kill — measured against the kill criteria written in Course 0, and executed without renegotiation.

Why it matters

Rule 2 only has value at this moment. Kill criteria are trivially easy to write in month zero and genuinely hard to honour in month nine, because by then the site has your name on it, your weekends in it, and a handful of encouraging numbers you can point to. Every failing portfolio contains a site that should have been killed at Month 9 and was instead given "one more quarter" — repeatedly. The renegotiation is the failure mode, not the site.

How to do it

1. Retrieve the Course 0 kill criteria before opening any analytics. Reading the numbers first contaminates the judgement.
2. Compute contribution margin per site — revenue minus loaded cost, from Course 7.5 and 11.1 — never gross revenue.
3. Compute cost per durable outcome: cost per retained subscriber and per returning user, not per pageview.
4. Apply the criteria mechanically and write the verdict before discussing it with anyone.
5. For prune, name exactly which assets are removed or consolidated and redirect them properly per Course 5.
6. For kill, execute the wind-down: stop publishing, preserve the owned audience, honour commitments, archive the assets, and log the decision with its reasoning.

WIND-DOWN SEQUENCE (kill verdict)
1. Stop new publishing immediately. Do not "let it run" without maintenance —
   unmaintained pages with commercial claims are a compliance liability.
2. Tell the audience plainly. One honest final issue outperforms silence,
   and offer a migration path to a surviving site's newsletter if it is genuinely relevant.
3. Honour every outstanding commitment: sponsor terms, partner agreements, refunds.
4. Export and keep the owned audience. The list is the asset that survives the site.
5. Archive the content, the artifacts and the analytics before anything is switched off.
6. Decide the domain's fate deliberately: keep, redirect only where topically honest, or let lapse.
   Never sell a domain whose authority came from your editorial reputation into unknown hands.
7. Log the decision, the evidence and the reasoning. This log is what makes the NEXT gate credible.

Tools needed

Course 0 kill criteria, the pruning scorecard, unit economics, and the risk register.

Lab 12.4

Issue one verdict per site with the evidence attached, then execute the required action for at least one site. If your verdict set contains three "keeps", audit your thresholds — write down what result would have produced a kill, and check that such a result was actually possible.

Artifact

Month 9 verdict & kill log
  Per site:
    Kill criteria (retrieved BEFORE analytics): ___
    Contribution margin: ___ | Cost per retained subscriber: ___
    Cost per returning user: ___ | Compliance position: defensible Y/N
    VERDICT: keep / prune / kill
    Evidence: ___
    Criteria renegotiated? Y/N — if yes, why, and new threshold + date: ___
  Prune list (assets removed/consolidated + redirects): ___
  Wind-down record (if killed): steps 1-7 with dates ___
  What the process taught me about my own judgement: ___

Common mistakes

• Reading the analytics before retrieving the criteria, then finding the criteria "need context".
• Deciding on gross revenue, which almost always makes the most expensive site look like the best one.
• Abandoning a killed site quietly instead of winding it down — leaving live commercial claims nobody maintains.
• Discarding the email list with the site, destroying the only asset that transfers.

Pro astuces

• Have someone else read the verdict against the criteria. Five minutes of outside reading defeats months of accumulated attachment.
• A well-executed kill returns capital, capacity and — most valuably — a documented reason. That reason is worth more than the site was.

12.5 How do you allocate capital across the portfolio?

What is it?

A written decision, made at each gate, about where the next quarter's money and hours go — driven by measured contribution margin and return demand, and constrained by the Course 11 concentration limits.

Why it matters

Capital and attention are the two scarce inputs, and attention is scarcer. Left undirected, both drift toward whichever site is currently most interesting or most stressful — which is rarely the one with the best marginal return. Explicit allocation forces the comparison that a three-site portfolio exists to make possible, and the 50% single-site cap prevents the winner from quietly becoming the whole business.

How to do it

1. Compute contribution margin and cost per durable outcome for all three sites on the same date, using the same definitions.
2. Rank by marginal return: where does the next unit of spend produce most? This is rarely the same ranking as total revenue.
3. Allocate money and hours separately. Hours are the binding constraint, and a site can be cash-cheap and attention-expensive.
4. Apply the 50% single-site revenue cap. If the leader is approaching it, fund the second-strongest instead — build the alternative, never damage the leader.
5. Reserve a fixed share — 10–20% — for maintenance of the existing assets: claim re-verification, corrections, tool upkeep. Unfunded maintenance is deferred failure.
6. Write what you are explicitly not funding this quarter, and revisit at the next gate rather than mid-quarter.

Tools needed

Unit economics, the revenue calculator, your Course 9.6 dashboard, and the Course 11 concentration report.

Lab 12.5

Write the next quarter's allocation: money and hours per site, the marginal-return ranking that justifies it, the maintenance reserve, and an explicit not-funding list. Then check the allocation against the 50% cap and state what you would do if the leading site breached it next month.

Artifact

Allocation decision record
  Date: ___   Gate: Month 3 / 6 / 9
  Per site: contribution margin ___ | cost per retained subscriber ___
            marginal-return rank ___ | share of portfolio revenue ___%
  Allocation: money ___% / ___% / ___%   hours ___h / ___h / ___h
  Maintenance reserve: ___% (claim re-verification, corrections, tool upkeep)
  NOT funding this quarter: ___
  50% single-site cap: current leader ___% | action if breached: ___
  Next review: at the next gate (not mid-quarter)

Common mistakes

• Allocating by gross revenue, which funds the most expensive site rather than the most efficient one.
• Allocating money but not hours, then wondering why the funded site did not move.
• Leaving maintenance unfunded, so claims expire and corrections queue up until a compliance problem forces a freeze.
• Reallocating mid-quarter every time a site has a bad week, which produces three half-tests instead of one answer.

Pro astuces

• The not-funding list is the most useful line in the record. Without it, every plan silently expands to include everything.
• When all three sites look ambiguous, fund one fully rather than all three partially. Ambiguity usually means underinvestment, and three underfunded tests answer nothing.

12.6 What makes the portfolio transferable — and when do you expand?

What is it?

A documented asset pack per site — audience, content, tools, data, process and governance history — that could be handed to a buyer, a partner or a successor. Plus the evidence test that must pass before the portfolio grows beyond three sites.

Why it matters

Transferability is the honest measure of whether you built an asset or a job. If the site cannot operate without your undocumented knowledge, its value is your continued attention, and it cannot be sold, delegated or paused. The asset pack also answers the governing question in its most durable form: an owned audience and a governance record are exactly what survive when the intermediary changes its mind. And expansion is where most portfolios fail — copying an unproven process ten times only produces ten copies of the same mistake.

How to do it

1. Document the owned audience: list size, retention, consent basis, export format, and proof that an export→import round trip actually works.
2. Document the content and tool inventory with the claim register state, so a successor knows what is verified and what has expired.
3. Document the process: publishing SOP, review SOP, screenshot SOP, and the release checklist — enough that someone else could ship an issue this week.
4. Attach the governance history: gate records, the trigger history, corrections, and the kill log. This is what makes the numbers believable.
5. Run the succession test — could someone else operate this for a month using only the pack? Fill every gap it exposes.
6. Apply the expansion test before adding site four, and expand by one at a time, never in batches.

Tools needed

Your claim register, risk register, checklists, gate records, and your IndexedDB export from the dashboard.

Lab 12.6

Assemble the asset pack for your strongest site and run the succession test on the riskiest part — have someone else attempt one newsletter issue using only your documentation. Then complete the six-requirement expansion test honestly and record the verdict, including "not yet" if that is the truth.

Artifact

Transferable asset pack
  Site: ___
  1. Audience: size ___ | retention ___ | consent basis ___ | export/import round trip TESTED ___
  2. Content & tools: inventory ___ | claim register state (verified/expired) ___
  3. Process: publishing SOP ___ | review SOP ___ | screenshot SOP ___ | release checklist ___
  4. Governance: Month 3/6/9 gate records ___ | trigger history ___ | corrections ___ | kill log ___
  5. Succession test: operated by someone else for 1 month? ___ | gaps found ___
  EXPANSION TEST (all 6 must pass):
    [ ] One site passed Month 9 as KEEP
    [ ] One kill or prune actually executed
    [ ] Return demand proven on the keeper
    [ ] All four concentration limits within caps
    [ ] Newsletter cadence held 6 months (>=90%)
    [ ] Asset pack passed the succession test
  VERDICT: expand by ONE site / not yet — reason: ___

Common mistakes

• Treating the asset pack as an exit document rather than an operating one — it is what makes the site survivable, not just saleable.
• Claiming a tested export when only the export half was ever run. An untested import is not a backup.
• Expanding to sites four through ten in one batch because the process "feels repeatable now".
• Skipping the succession test because you already know how everything works — which is precisely the problem it detects.

Pro astuces

• Keep the pack current as you go. Written at the end, it is fiction; written continuously, it is the operating manual.
• The governance history is what a serious buyer or partner examines first. Gate records and a kill log prove the numbers were produced by a process rather than by luck.
• Expanding by one site at a time preserves the thing that made three sites valuable: the ability to tell which variable produced the result.

Scenario assessment: the Month 9 gate you do not want to hold

Month 9. SMB Finance: contribution margin positive, newsletter retention stable at 41% after four issues, direct and branded arrivals up from 4% to 19%, all concentration limits within caps. B2B SaaS and AI Tools: highest gross revenue of the three and growing, but 71% of it comes from one affiliate programme, margin is thin after review costs, and direct arrivals are flat at 3%. SMB Cybersecurity: your favourite site, best-written, two industry awards for the newsletter — 380 subscribers, 12% retention after four issues, negative contribution margin for six straight months, and its Course 0 kill criteria read "kill if contribution margin is negative at Month 9 and retention is below 20%." You have already moved that threshold once, at Month 6. A broker has offered $14,000 for the cybersecurity domain, which is more than it has ever earned. Runway is seven months.

Final project: the portfolio you can defend

This is the terminal deliverable for the whole curriculum. It is not a plan — it is a record of decisions you actually made, with the evidence that produced them and at least one verdict that cost you something.

Deliverable
1. Portfolio launch plan: 3 sites staggered across 90 days, 12 assets each, proceed-gates between launches
2. Month 3 gate record: execution scored per site, root cause of any shortfall, results deliberately excluded
3. Month 6 gate record: Course 0 thesis verbatim vs measured return demand — supported / unsupported / untested
4. Month 9 verdict & kill log: one verdict per site, criteria retrieved BEFORE analytics, wind-down steps if killed
5. Allocation decision record: money AND hours per site, maintenance reserve, explicit not-funding list
6. Transferable asset pack: audience, content, process, governance history, succession test result
7. Expansion verdict: the 6-requirement test completed honestly, including "not yet" if that is the answer
8. One page answering the governing question per site, using measured evidence only

Pass standard: every site has three dated gate records; at least one prune or kill was executed as written rather than renegotiated; capital is allocated by contribution margin rather than gross revenue; the asset pack has survived a real succession test; and the governing question is answered per site with measured return-demand evidence — no hypotheticals.

Course 12 checklist

• Three sites ranked by displacement-adjusted score before launch order was set
• Launches staggered across 90 days, not simultaneous
• Site 1 retrospective written before site 2 launched
• Missed-issue proceed-gate honoured before site 3
• Portfolio frozen at 3 sites until the Month 9 gate
• Month 3 gate held on a booked date with the scorecard open
• Month 3 judged execution ONLY — no traffic or revenue verdicts
• Root cause recorded for every shortfall (capacity / cost / attention)
• Cost variance vs the 12-month model computed at Month 3
• Course 0 thesis restated verbatim at Month 6
• Return demand measured: retention, direct+branded share, repeat tool use
• Thesis marked supported / unsupported / untested without rewriting it
• Four concentration ratios checked at Month 6
• Kill criteria retrieved BEFORE opening analytics at Month 9
• Verdicts decided on contribution margin, never gross revenue
• Cost per retained subscriber and per returning user computed
• One verdict issued per site: keep / prune / kill
• Any threshold renegotiation recorded explicitly, with the count
• At least one prune or kill actually executed, not deferred
• Wind-down sequence followed: audience preserved, commitments honoured, assets archived
• No unmaintained pages with live commercial claims left running
• Allocation records money AND hours separately
• Maintenance reserve funded at 10–20%
• Explicit not-funding list written for the quarter
• 50% single-site cap checked, with an action if breached
• Export→import round trip tested, not just exported
• Asset pack includes the governance history and the kill log
• Succession test run by someone else for one month
• All six expansion requirements assessed honestly
• Expansion limited to ONE site at a time, never a batch
• Governing question answered per site with measured evidence only

AI-agent prompt for Course 12

Act as an independent portfolio gate auditor for a publisher running exactly three sites: SMB finance and banking, B2B SaaS and AI tools, and SMB cybersecurity. Your job is to hold me to criteria I wrote BEFORE I had any emotional stake, and to refuse to help me renegotiate them.

I will paste: my Course 0 thesis and kill criteria per site, my 12-month cost model, revenue and loaded cost per site, arrivals by source, newsletter retention per issue, tool usage, AI citation presence, and my cash runway.

Do this:
1. Reconstruct the launch sequence and tell me whether the sites were staggered and whether each proceed-gate was honoured. Flag simultaneous launches as a learning failure.
2. Hold the Month 3 gate on EXECUTION ONLY: assets live out of 12, newsletter issues shipped vs promised, tool live and saving state, compliance gate, cost variance vs model. Explicitly REFUSE to judge traffic or revenue at this gate, and say why.
3. Hold the Month 6 gate on RETURN DEMAND: newsletter retention after 4 issues, direct plus branded share of arrivals, repeat tool usage, AI citations. Compare against my thesis VERBATIM and mark it supported, unsupported or untested. If I appear to have rewritten the thesis after seeing data, say so plainly.
4. Hold the Month 9 gate: compute contribution margin per site and cost per retained subscriber and per returning user. Apply my kill criteria MECHANICALLY and issue one verdict per site: keep, prune or kill.
5. If a site meets its kill criteria, say KILL. Do not soften it, do not propose "one more quarter", and do not weight awards, writing quality, sunk effort or my attachment. If I have already moved a threshold once, tell me I am moving it again and count it.
6. For any kill, output the wind-down sequence: stop publishing, tell the audience honestly, honour commitments, export and KEEP the owned audience, archive assets and analytics, decide the domain deliberately, log the decision.
7. Allocate the next quarter's MONEY and HOURS separately by marginal return, enforce the 50% single-site cap by building the alternative rather than damaging the leader, fund a 10-20% maintenance reserve, and write an explicit NOT-FUNDING list.
8. Assess the transferable asset pack and run the six-requirement expansion test. Answer "not yet" whenever it is the honest answer.

Rules:
- Never recommend expanding beyond three sites unless all six requirements pass, and never by more than one site at a time.
- Never use gross revenue for a keep/kill recommendation. Contribution margin only.
- Never treat high gross revenue as strength when it is concentrated: flag affiliate >30%, single site >50%, traffic source >60%, single sponsor >25%.
- Never advise leaving a killed site's pages live and unmaintained while they carry commercial claims.
- Never advise discarding the email list with the site, and never advise migrating subscribers without honest, permission-respecting disclosure.
- Domain sales: raise the reputational risk of selling a domain whose standing came from my editorial work, especially in YMYL-adjacent niches.
- Do not invent benchmarks, "typical" retention rates, RPMs or valuations. Mark anything unmeasured as UNMEASURED.
- Compliance, tax, consumer-protection and data-protection matters vary by jurisdiction: state that you are not giving legal, financial or tax advice.
Output one table per gate, then a plain-text verdict block per site with any renegotiation attempts, breached caps and already-met triggers called out ABOVE the tables.

Primary sources

• FTC endorsement, influencer and review guidance — obligations that continue to apply during a wind-down and to any surviving pages.
• Redirects and site moves — the correct handling for pruned and consolidated assets.
• Search spam policies — relevant to expired domains, redirect abuse and consolidation done badly.
• AdSense programme policies — publisher obligations that persist while pages remain live.
• Web Vitals — the experience baseline the surviving portfolio must still meet.

Platform policies, affiliate terms, disclosure duties and tax treatment of asset sales change frequently and vary by jurisdiction. Every claim here is a dated claim with a 90-day expiry — log it in your register, re-verify against the primary source, and take professional advice before selling, winding down or transferring any asset.