At 10:47 on a Tuesday night, a chief financial officer you have never met opens an email forwarded by your sponsor with the subject line "worth two minutes." Attached is one page. You are asleep. You will never be in the room where this page is read, never get to add the caveat, never get to explain what the asterisk means or why the recommendation says PILOT and not something safer. The page has to do all of that alone: state the verdict, show the score, price the fix, name its own kill condition, and ask for a decision, in front of a skeptical stranger, at close to midnight, with no author present. Every artifact you have built across this level, the process map, the standard operating procedure, the baseline pack, the data audit, the people scorecard, the selection score, exists so that this one page can survive that moment. This lesson teaches you to build it.

The Verdict, Not the Lab Result

Back in Chapter 2.3 you built the One-Page Data Readiness Report: the four data artifacts compressed to a single page a steering committee would actually read. That page taught you the core discipline, compression with a chain, every number on the page recomputable from an exhibit behind it. Keep the discipline. But understand clearly what that page was and what this one is, because the difference is the difference between a lab result and a verdict.

The data readiness report answered one question about one dimension: is the data ready for this pilot? It was a specialist's finding, and like a blood panel, it informed a decision without being one. The document you build in this lesson, the One-Page Readiness Report, is the verdict. It integrates everything the level produced: the selection score that says this process deserved assessment at all, the three readiness dimensions (process, data, people) with their conditions attached, the remediation economics that price the path to ready, and a recommendation with a verb in it that someone can approve, defer, or kill from. A committee should be able to make the funding decision from this page without reading anything else, while everything else stands one tab behind it, ready for inspection the moment anyone asks.

This is the assessor's signature artifact. Not the most sophisticated thing you have built this level, but the one everything else exists to feed, the way an architect's hundreds of drawings exist so that a planning committee can approve one elevation. It is also, professionally, the artifact with your name on it in the largest sense: when the pilot succeeds, this page is the record that you called it; when the assessment kills a bad design, this page is the documented kill that counts as a win. The program has told you since Level 1 that a documented kill is a win. This page is where the documenting happens.

And the stakes of the craft went up for one specific reason: this page will be forwarded. The data report was mostly read by people who knew you and could ask you questions across a table. The readiness report travels. Sponsors forward it upward, CFOs forward it sideways, a board member's chief of staff prints it for a flight. It will be read by people who never met you, stripped of every scrap of hallway context, with nobody present to defend it. A page that needs its author in the room is a page that fails the moment it succeeds enough to be forwarded, which is exactly when it matters most.

Write the page to survive being forwarded: it will travel to a reader who never met you, carrying no context but its own, and it must win alone.

Two Reports, Same Assessment: How the Honest Analysis Lost

Here is the failure story this lesson is built against, and notice that it is not a story about a bad assessment. It is a story about a good one, badly delivered.

An operations lead at a mid-size logistics firm spends six weeks running a genuinely rigorous readiness assessment of a customs-documentation process. The evidence is strong, the scoring is honest, the remediation path is priced. She writes it all up as a twelve-page deck: methodology first, findings in the middle, recommendation on slide eleven. It is, by any analytical standard, excellent work. It goes to the steering committee in April.

The committee defers it. Not rejects: defers. "Good work, a lot here, let's give it proper time next month." In May it is deferred again, this time with the request that stings precisely because it is reasonable: "can you come back with a summary?" While the summary is being scheduled, a regional vice president presents a competing AI initiative on one confident slide: a hero metric, a vendor logo, an implied timeline, no baseline, no conditions, no kill criteria. The committee, relieved to finally approve something, funds it in eleven minutes. Nine months later that pilot is quietly shut down, roughly $310,000 spent (an illustrative figure, but a familiar shape), having joined the statistics you know: the 42 percent of companies S&P Global found scrapping most of their AI initiatives in 2025, the 95 percent of enterprise GenAI pilots MIT found delivering no measurable return. The rigorous assessment, meanwhile, never got its slot. The organization got the worst of both worlds: it paid for honesty and bought enthusiasm.

Now run the counterfactual, because it is the entire lesson. Version B of the same assessment: identical evidence, identical scoring, identical integrity, formatted as one page with a four-verb recommendation on the first line and the ask dated at the bottom. That page takes eleven minutes to decide on, which means it fits exactly the slot the VP's slide took. It would have been read in the room, forwarded after the room, and decided at the meeting it named. Same truth, different container, opposite outcome.

The uncomfortable conclusion: the market for executive attention is brutally efficient, and it does not grade on analytical effort. A twelve-page deck and a one-page verdict are not two formats of the same product; in a steering committee they are different products, and only one of them is a decision instrument. This is the Chapter 2.3 archaeology lesson (unread evidence is not evidence) escalated to the full assessment: format is not vanity, and it is not a concession to lazy executives. Format is the delivery mechanism for truth. An honest analysis that loses the attention contest does not merely go unread; it leaves the slot open for the one-slide enthusiasm pitch, which is how good organizations end up funding exactly the pilots their own evidence warned them about.

The Anatomy, Part One: Header, Verb, Scores, Evidence

The One-Page Readiness Report has seven blocks in a fixed order. We walk them in two passes, with the copyable language patterns as we go. The first four blocks establish identity, verdict, and proof.

Block 1: The header block

Four lines of small print at the top: the process name, the assessment window, the assessor's name, and, critically, the artifact versions the report relies on. For the running example this level has been building: process map v1.1, SOP v1.0 (SOP: standard operating procedure, the written instruction set for the work), baseline pack 2026-06, data readiness report of Aug 21, people readiness scorecard v1.0. One row of small print, and it does something no amount of confident prose can do: it turns "trust me" into "check me." Any reader can pull map v1.1 and verify. Committees have been burned by reports resting on nothing, and they can smell the difference; they fund "check me." The version line also protects you across time: when someone quotes your report in March against a process that changed in January, the header states exactly which world you assessed.

Block 2: The recommendation line, first line of the body

Not the last line. The first. One of exactly four verbs, each of which is only complete with its required attachment:

• PILOT, with conditions and dates. "Pilot the exception-routing design, go-live Oct 27, conditional on the two funded fixes below clearing by their dates." A PILOT without conditions from a first full assessment should worry everyone; real processes are almost never unconditionally ready.
• REMEDIATE-FIRST, with the priced fix list and a re-score date. Not "the process needs work" but "fund these three items, $62,000, and re-score on Nov 3." A remediation verb without a re-score date is a polite never.
• DEFER, with the trigger that reopens it. "Defer until the ERP migration completes in Q1; reopen automatically at cutover plus 30 days." A DEFER without a trigger is a graveyard with better manners.
• KILL THIS VERSION, with what would change the answer. "Kill this version of the pilot; a redesign that removes the manual re-keying step, or a data source with resolution codes attached, would reopen the question."

Spend a moment on the fourth verb, because it is the one assessors flinch from and the one that most builds a career. "Kill this version" beats "no" in three ways. It is more accurate: assessments almost never kill goals, they kill specific designs, and the phrase says so. It preserves the salvageable parts of the idea, which are usually real. And it preserves the proposer's dignity, which is not sentimentality but operational realism: the proposer will be in the room, and next lesson is entirely about surviving that room. A "no" makes an enemy; a "kill this version, and here is what would change the answer" makes the proposer the potential author of version two. This is the program's "a documented kill is a win" principle, institutionalized into the report's own vocabulary.

Block 3: The score block

Three items, compressed: the process's position on the selection 2x2 (the value-versus-feasibility grid from the start of this chapter), the weighted selection total against the threshold ("3.9 against a pilot threshold of 3.5"), and the three readiness dimension scores: process, data, people. And here is the rule that separates honest compression from quiet fraud: the conditions travel with the scores. The conditional 3 you produced in the scoring session, the data score that is a 3 only if the dedup and retention fixes are funded and land by their dates, appears on the page as "3*" with its footnote right there: "*conditional on the two funded fixes clearing by Oct 10; unfunded, this dimension scores 2 and the recommendation reverts to REMEDIATE-FIRST." Dropping the asterisk and printing a clean 3 is how reports lie while telling the truth: every individual number technically defensible, the overall impression false. The asterisk costs you a little polish and buys you the only thing the page actually runs on, which is being believed the second time.

Block 4: Evidence highlights

Three to five one-liners, each with the same skeleton: a finding, a number, a source. These are the level's greatest hits, chosen for decision relevance, not drama: the exception-reason field that is free text in 61 percent of records (data profile, Exhibit C); the blame-fear theme raised by 11 of 14 interviewees (interview synthesis, Exhibit E); the roughly $428,000 annual run rate of the process (baseline pack 2026-06, Exhibit A). Each line is traceable to a tab; none is an adjective. The test for whether a finding earns one of these slots: would a stranger reading only these lines understand why the recommendation says what it says? The highlights are the recommendation's visible reasons. If your verb and your highlights do not obviously connect, one of them is wrong.

The Anatomy, Part Two: Money, Kill Condition, Ask

Block 5: Remediation economics

One short block that prices the conditions and sets them against the benefit. For the running example: $34,000 of data fixes plus $28,000 of role-based training equals a $62,000 one-time readiness investment, set against an illustrative annual benefit of $130,000 to $180,000, with the basis of the range stated on the page: the low end assumes 55 percent of eligible exception volume auto-routes at target accuracy with the slower adoption the people scorecard predicts, the high end assumes full eligible volume, both computed against the $428,000 baseline.

Notice the shape: a range, not a point, with the range's basis stated. This is deliberate and it is doing credibility work. You learned in Level 1's ROI lesson that the single-point benefit projection, the "$540,000 in year one" with no stated assumptions, is the hockey-stick tell, the signature of every pitch that later joined the 95 percent. Committees that have been burned notice, and by 2026 most committees have been burned. A stated range with a stated basis says: this author knows the difference between a forecast and a wish, and has told you which assumptions to attack. Paradoxically, the humbler number is the more fundable one, because it is the one that survives the CFO's first question.

Block 6: Risk and kill condition preview

Two lines. First, the top residual risk, the one that survives all the funded conditions: here, the blame-fear pattern, 11 of 14 interviewees afraid of owning the AI's mistakes, which no data fix touches and which the rollout design must absorb. Second, the pre-committed kill condition the pilot will carry into its charter: "if verified extraction accuracy is below 85 percent at the week-6 checkpoint, revert to manual routing and remediate" (an illustrative threshold; yours comes from your own baseline and tolerance).

Understand what this block does, because it is the single strongest credibility move available to an assessor: a report that names its own kill condition before funding proves its recommendation is falsifiable. Every committee has sat through advocates whose projects could not fail, only be underfunded, undermined, or measured too soon. A recommendation that arrives pre-loaded with the evidence that would kill it is a different species of document: it reads as assessment, not advocacy, and it converts the committee from adversaries checking your work into partners in a test you designed. It also quietly forecloses the zombie-pilot future, because the kill condition is now on the page the committee approved, with a date. This block is a preview of the pilot charter you will write in this chapter's final lesson; here it appears as two lines, planted early, where they buy the most trust per word.

Block 7: The ask

The page ends with the specific decision, at the specific meeting, on the specific date: "At the Sept 4 steering committee: approve the $62,000 readiness investment and the pilot charter with the conditions and kill condition above." Not "for discussion." Not "recommended next steps." A named decision with a calendar slot. The ask is what converts the page from a memo into an instrument; a report without an ask produces appreciation, and appreciation funds nothing. Write the ask last when drafting, place it last on the page, and make it the second thing you polish after the recommendation line, because the verb opens the reader and the ask closes the meeting.

The Worked Example: The Invoice-Exception One-Pager, In Full

Here is the complete page for the assessment this level has been running: the accounts-payable invoice-exception process. Every figure is illustrative, carried forward from the artifacts you built in earlier chapters. This is the block to copy, block for block.

Read the page the way its three hardest readers will. The CFO lands on ECONOMICS and RISK: a $62,000 ask against a stated-basis range, payback math visible, a kill condition already priced in; nothing here requires believing the author. The chief operating officer lands on the RECOMMENDATION and SCORES: named owners, dated conditions, a defined reversion if funding slips. The stranger the page gets forwarded to lands on the first line and the last: a verb at the top, a dated decision at the bottom, and every claim in between wearing a tab letter. Three readers, one page, no author required. That is the standard.

And notice what the page never does. It never editorializes ("exciting opportunity"), never rounds the asterisk away, never presents the benefit as a point, and never asks for trust it has not wired to a tab. The page is confident everywhere and unsupported nowhere. That combination is rarer than it should be, which is exactly why it wins.

The Forwarded Reader, the Appendix, the AI Passes, and the Verb You Sign

Writing for the reader you will never meet

Because the page travels, it must define its own terms, briefly, without condescending. The practical rules: one parenthetical each for the instruments ("the selection scorecard (the weighted value-versus-feasibility rating from the assessment)" and "dimension scores (1 to 5 readiness ratings for process, data, and people)"), and no acronym uncontextualized, ever. This is the Level 1 rule about interrogating other people's artifacts, now applied to your own: you taught yourself to distrust any document that assumes context; do not ship one. The test is mechanical: hand the draft to someone outside the assessment and ask them to state the recommendation, the cost, and the kill condition in under two minutes. If they stumble on a term, the page failed, not the reader.

The appendix architecture

One page, then tabs: the scoring record with the anchor definitions and the weighted math; the three dimension reports (the data readiness report, the people scorecard, the process baseline and map); and the disagreement log from the scoring session, the record of where assessors differed and how it resolved. Committees almost never open the tabs, and their existence is precisely why: a page backed by a visible, organized appendix reads as the tip of an iceberg, and a page backed by nothing reads as a guess wearing formatting. Completeness signaled is completeness trusted. Include the disagreement log even though it feels like showing your seams, especially because it feels like that: a record that says "we disagreed on people readiness and here is how the evidence settled it" is worth more credibility than a unanimous score, because unanimity is what fabrication looks like too.

AI's four jobs on this page

AI drafted half the artifacts behind this page. On the page itself it has exactly four jobs, all assistive.

• Compression passes. Findings arrive as paragraphs; blocks need lines. "Compress to 30 words; the number, the source tab, and the owner must survive." Then verify what fell out, because compression is where models quietly drop qualifiers, and the asterisk on a conditional 3 is exactly the kind of thing that falls.
• The hostile-reader red team. "Read this page as a skeptical CFO who has personally scrapped AI projects before. List every challenge you would raise, in order of likely damage." Then rehearse the answers out loud, each one ending at a named tab. The three challenges the invoice-exception page should expect: where does the $130k to $180k range come from (Tab A unit costs times the eligible-volume assumptions, stated on the page), why believe the people score (Tab E, 11 of 14 speakers, counted voices not vibes), and why not start now and train in parallel (Tab E again: the blame-fear theme makes untrained go-live the specific failure mode the kill condition would catch six expensive weeks later).
• The consistency sweep. Hand the model the page and the appendix and ask it to recompute every number on the page from its appendix source, flagging orphans and mismatches. Then check its checking. A page that says $62k while the tabs sum to $67k is the most embarrassing error this document can carry, because it is the one error any reader can find without expertise, and one caught mismatch retroactively poisons every clean number. AI sweeps for it in minutes; there is no excuse left for shipping one.
• Drafting the four-verb language. Models are genuinely good at generating the disciplined phrasing of a REMEDIATE-FIRST with dates or a KILL THIS VERSION with reopening conditions, and reading three drafted variants sharpens your own. Draft with it freely.

What AI never does is pick the verb. The human owns the verb. PILOT versus REMEDIATE-FIRST is not a computation; it is a judgment call with a career attached, made under uncertainty, and the assessor signs it. The signature is not a formality: it is the whole difference between an assessment and a model output, and it is why the committee is buying your page instead of prompting for their own. The scores inform the verb; the assessor chooses it and answers for it. That is BCG's 10-20-70 arithmetic (10 percent algorithms, 20 percent technology and data, 70 percent people and process) applied to your own workflow: the model did the compression; the accountable judgment is the 70, and it is yours.

Zoom out once before the checklist. The playbook behind this program promises one scene as the payoff of the entire readiness discipline: walking into a CFO's office and saying, here is the scorecard we selected with, the audit we ran, the workflow we redesigned, the baseline we measured, the result we got, and the decision memo at each gate. The One-Page Readiness Report is that speech's first chapter, the earliest document in the chain that proves the whole thing was a method and not a streak of luck. Build it like it will be read again in two years, because if the pilot matters, it will be. Next lesson: delivering this page in the room, to the people whose project it judges, without making enemies.

What to Do Monday Morning

You have a scoring record, or you are one lesson away from one. Here is the sequence that turns it into the page.

1. Assemble the one-pager from the scoring record using the seven blocks in order: header with the artifact version line, recommendation line with one of the four verbs and its required attachment, score block with every asterisk intact, three to five evidence highlights (finding, number, source), remediation economics as a range with its basis stated, the two-line risk and kill condition preview, and the ask with a meeting and a date.
2. Run the hostile-CFO red team. Prompt the model to read as a skeptic who has scrapped AI projects before and list every challenge in damage order; rehearse an answer to each, ending at a named tab. Any challenge with no tab behind it means the page is carrying a claim it cannot defend: fix the page.
3. Run the consistency sweep. Every number on the page recomputed against its appendix source, by the model first and by you second. Ship no page with an orphan number.
4. Get the verb challenged. Give the page to one colleague who disagrees with your recommendation and let them argue the alternative verb for ten minutes. If your verb survives, you now have the rehearsal transcript for the meeting; if it does not, the colleague just saved you from signing the wrong one.
5. Attach the appendix tabs: scoring record, the three dimension reports, the disagreement log. Label them on the page even though almost nobody will open them; the labels are the trust.
6. Book the decision slot, with the ask stated in the calendar invite itself: "Decision requested: $62k readiness investment and pilot charter." A page waiting for a meeting to happen to it is the twelve-page deck with better margins.

Key Takeaways

• Build the One-Page Readiness Report as the level's signature artifact: the single decision surface integrating selection score, the three readiness dimensions, remediation economics, and a recommendation a committee can approve, defer, or kill from without reading anything else.
• Write for the forwarded reader: the page travels to people who never met you, so it must define its own terms, contextualize every acronym, and win with no author in the room.
• Open with one of four verbs and its required attachment: PILOT with dated conditions, REMEDIATE-FIRST with a priced list and re-score date, DEFER with a reopening trigger, or KILL THIS VERSION with what would change the answer; prefer "kill this version" to "no" because assessments kill designs, rarely goals.
• Keep every asterisk: display conditional scores as "3*" with the footnote on the page, because hiding the condition is how reports lie while telling the truth, and the version line in the header is what turns "trust me" into "check me."
• Price remediation against a benefit range with its basis stated, never a single point: the point forecast is the hockey-stick tell, and committees that have been burned notice.
• Name the kill condition before funding: a recommendation that arrives falsifiable, with its own pre-committed kill trigger and checkpoint date, is the strongest credibility move an assessor can make.
• Use AI for compression passes, the hostile-CFO red team, the page-versus-appendix consistency sweep, and drafting the four-verb language, then sign the verb yourself: the recommendation is a human judgment with a career attached.
• End with a specific ask at a specific dated meeting and attach the visible appendix tabs, because completeness signaled is completeness trusted, and a report without an ask is a memo.