Where we end up
The spine is finished at the end of this session, and the thing it produces is not a plan. It is an artefact: one page, written down, that a vertical roadmap can import so the industry write-up spends its pages on the industry rather than re-deriving the same four moves. Without it, every vertical carries its own copy of the frame, and copies disagree. It is not a matter of whether, only when.
★ What you walk out with tonight. One page with seven sections filled in, a first three you can defend, a named list of what is off the roadmap and why, and a six-row score on how much of the spine your page actually carries.
One page, seven sections, nothing else
7 min liveThe constraint is the point. A page that runs to four pages stops being read, and once it stops being read the decisions inside it stop being revisited. Seven sections, each short enough to argue with.
LiveWhy section 7 exists, and why it is named rather than implied4 min▶
Most roadmaps are a list of what is in. An item that was considered and rejected simply vanishes, which means the person who asked for it has no idea it was considered at all. So they ask again next quarter, in good faith, and the whole argument is re-run from nothing by people who were not in the first room.
Naming it fixes that cheaply. Section 7 lists each rejected candidate, in the words it arrived in, with one line on why. The important entry is the popular one: on the bench in session 3 the single most-requested item on the list needed no data at all, cleared every gate in 4 weeks, and carried a revenue fit of zero. It passes everything except the gate that matters. If that is not written down with its reason, it comes back, and it comes back with momentum.
The tone matters more than the content here. "Not on the roadmap because it does not move the money this business makes, and here is the money sentence" is a reason somebody can disagree with in a specific way. "Not prioritised" is not a reason, it is a way of ending a conversation without having it, and it guarantees the conversation repeats.
LiveWhat each section inherits, and what it must not restate3 min▶
Section 1 is one sentence. Not a business model canvas. From session 5: the same use case is worth different amounts at different businesses, so this sentence is what gives every score below it a meaning. If it takes two sentences, the second one is usually describing a different business inside the same company, and that is a finding worth its own page.
Sections 2 and 3 are one table, two columns. Rung and verdict. They are separate axes and putting them in one table is what stops them being conflated. A row can read "predict it / augment the person" and that combination is the normal case, not an edge case.
Section 4 is binary and short. Clears today, or does not, with the blocking entity named. No maybes. "Held partly" resolves to does not clear, which is the whole lesson of session 4.
Section 5 lists gaps, not wishes. Each one classified as a form change, a pipeline change or a standing job, because those three have wildly different costs and a list that does not distinguish them is a list nobody can price.
Section 6 is three items and three reasons. Not a twelve-quarter Gantt chart. Three, because that is roughly what anybody can hold, and the reasons are what make the three revisable.
Self-studyThe sequence, and how session 3 changed what belongs in it4 min▶
The instinct is to sequence everything and start at the top. The bench in session 3 put twelve candidates through the two gates and the result was blunt: dropping everything that does not clear the gate today took the plan from 64 weeks to 27, with revenue fit unchanged at 12, because the same six cases cleared in both versions. Two thirds of the plan length was work that could not be built and therefore contributed nothing while still consuming the estimate.
That is why section 6 only sequences what clears. The rest is not deleted, it moves into section 5 as a gap with a named blocker, which is a different kind of object: a gap is something you can choose to close, whereas a blocked item sitting in a sequence is a promise you cannot keep.
The other half of the same bench is the reason section 7 has to name the popular item. The three most-requested new-data cases produced a revenue fit of 0 over 25 weeks. Enthusiasm and revenue fit turned out to be uncorrelated on that board, and there is no reason to expect them to correlate on yours either, in either direction.
One honest limit, the same one every figure on these pages carries: the gym chain is invented, every effort figure is stipulated, and none of it is a claim about how often real portfolios are blocked or how often the favourite is empty. The arithmetic is real. The chain is not. The mechanism is what transfers.
What makes the page survive contact with a planning cycle
5 min liveA roadmap written once and never revisited is worse than no roadmap, because it carries the authority of a decision and the content of a guess. Three habits make the difference, and all three are about writing down the reason next to the thing.
Date it, and attach the reason to every decision. Not a version number, a date, at the top, next to the money sentence. A decision with its reason attached can be noticed going stale by anybody who reads it. A decision without one can only be defended or overturned, and both of those need the original room back.
LiveTriggers: what would change this line, written on the line3 min▶
Next to every gap, what would change its classification. A standing job that somebody finds a feed for becomes a pipeline change. A pipeline change with a screen already in the build queue becomes a form change. Writing the trigger turns a gap from a permanent complaint into something with a condition attached, and conditions can be met by people who were never in the planning meeting.
Next to every off-the-roadmap item, what would move it on. Usually one of two things: the money sentence changes, or a gate it failed gets cleared by work done for another reason. For the zero-fit favourite, the honest trigger is the first kind. It is not "we will get to it", it is "if the business starts making money in a way this item touches, it moves, and here is the sentence it would have to change".
And the discipline from session 5: do not quote a failure-rate statistic in the business case. Not one of the three circulating numbers survives inspection. One expired by its own stated terms, one traces back to an unattributed remark on a conference panel, and one is called misguided by a study that went and read its sources. The temptation is real, because a scary percentage makes a slide land. Resist it, because the case does not need it: the argument from the two gates is that two thirds of the plan length cannot be built at all, and that is a statement about your own board that anybody in the room can check.
Self-studyReading somebody else's roadmap with the same instrument4 min▶
The four moves work in reverse. Handed a roadmap you did not write, from a vendor, a consultancy, a previous post-holder or another division, run four questions down the items and stop. It takes about twenty minutes and it needs no access to anybody's systems.
What question does this item answer? Items that name a technology and a location answer none. From session 1, those belong in a "no question yet" pile rather than being assigned a rung to be polite, and on a document you have inherited that pile is usually the largest column. That is a finding about the document, not about you.
Which rung is it on? Then look at the rung below it and ask whether that one is trusted here. A skipped rung does not error, it ships and becomes confidently wrong, so an inherited roadmap with three rung-four items and a contested rung-one metric is a specific, sayable diagnosis.
What does it need that we do not hold? The fastest of the four. Most inherited roadmaps were written without anybody checking the entity list, and "held partly" is the state that kills, because the query runs and the answer looks like an answer.
Does it move the money this business makes? Last, because it is the one that requires the money sentence. This is where a plan built for a business with a different revenue model shows itself, which is the normal case for anything arriving from outside and is nobody's bad faith.
Two notes on manners. Running this on somebody else's document produces a lot of findings fast, and delivering all of them is a way of being right and useless. Lead with the rung diagnosis and the entity gaps, which are factual. And keep the verdict axis out of the first conversation entirely: from session 2, a verdict is passed on a task, never on a job or a department, and an inherited roadmap almost always states things at the level of a department.
The seam, said out loud
3 min liveThese six sessions are a spine and deliberately nothing more. They do not know your industry, your product, your architecture or your experiment design, and pretending otherwise is how a general frame turns into bad specific advice. Here is exactly where the spine stops and what picks it up.
| If what you need next is | It is on this shelf as | And it starts from |
|---|---|---|
| One business model mapped end to end, with this frame applied to it | learn-ship-ecommerce-enabler-with-phoebe | Imports the spine. Sections 1 to 7 are assumed, not re-derived |
| A five-year lookback on a real platform, with what actually happened | learn-ship-platform-diagnosis-with-phoebe | Imports the spine, then runs it backwards over real history |
| Deciding what to build for one product rather than a portfolio | learn-product-thinking-with-phoebe | A different unit of analysis: one product, not a candidate list |
| Choosing what to build it on | learn-system-design-with-phoebe | Starts after the what is settled. The spine never chooses a technology |
| Reviewing a data product that already exists | learn-data-heuristics-with-phoebe | Judging a thing that shipped, which is not what a roadmap does |
| Working out whether what you shipped actually worked | learn-experimentation-with-phoebe | The measurement the spine assumes somebody else is doing |
The import is one-directional and that is on purpose. The two vertical roadmaps read this page; this page knows nothing about them. That is what stops the frame quietly acquiring an industry's assumptions. If a vertical needs a change to the spine, the change belongs here, once, and both verticals get it.
Write your page, then score it
★ 22 min · bring sessions 1 to 5Five passes, in order, each using an output you already have. Then the scorecard below, which is the real test: not whether the page looks finished, but whether it carries the five things the spine was for.
Write the money sentence at the top, then date the page. One sentence on how this business makes its money, from session 5. Everything below is scored against it, so if it is wrong or vague nothing below means anything. The date is not decoration: it is what lets somebody in nine months know whether the reasons on the page were written before or after the thing that changed.
Write the sorted list with a verdict against each task. Rung from session 1, verdict from session 2, one table, two columns, so the two axes cannot be conflated. Pass the verdict on a task, never on a job or a department. Then count your augment verdicts: if there are none, that is worth a second look, because there is a published argument that the field carries excess incentives for automation over augmentation, so a portfolio with no augment verdicts is a signal about incentives rather than about the work.
Split the list into clears the gate today and does not, with the gap kinds. From sessions 3 and 4. Held partly resolves to does not clear, because that is the dangerous state: the query runs and the answer looks like an answer. For every blocked item, name the entity and classify the gap as a form change, a pipeline change or a standing job. On the bench, six of twelve entities were held in full, and the half-filled field was worth more than any new source.
Write the first three, and the reason each of the three is there. Only from what clears today. Three, not twelve. The reasons are what make the three revisable later, and they are the thing you will be asked for first. Remember what dropping the blocked items cost on the bench: the plan went from 64 weeks to 27 and the revenue fit stayed at 12, because the same six cases cleared either way.
Write what is deliberately off the roadmap, with what would move each item on, then score the card. Name the popular one explicitly. On the bench the most-requested item needed no data, cleared every gate in 4 weeks, and carried a revenue fit of zero. Next to each omission, the trigger that would move it on. Then take the six-row scorecard below and be harsh, because a 0 is the next thing to write rather than a mark against you.
If your page will not fit on a page. The commonest cause is section 2 carrying twenty candidates. That is not a formatting problem, it is the list not having been through the gate yet: run pass 3 and most of them move into section 5 as gaps, which is where they belong and where they are cheaper to argue about.
Six rows, one per session
10 minOne row for each of the six sessions. Score the page you just wrote, not the page you intend to write. The wording of each row is the claim from that session, so a low score points at a specific session to go back to.
After the course
45 min- Show section 7 to the person who asked for the zero-fit item, before you show anybody the sequence. Not as a decision handed down, as a reason offered for disagreement. You will either keep the item off the page with their agreement or learn that the money sentence is wrong, and both of those are worth more than the meeting costs.
- Write the trigger next to your three most annoying gaps, including the ones you privately think will never close. A gap with a condition attached can be closed by somebody who was never in the room; a gap without one is just a complaint with a date on it.
- Take one roadmap you did not write and run the four questions down it. What question does each item answer, which rung, what does it need that you do not hold, and does it move the money this business makes. Stop there and write nothing else. Twenty minutes, and lead with the rung diagnosis, which is the factual part.
- Put a date in the calendar to re-read the page, one planning cycle out, with one instruction to your future self: check the reasons, not the items. Items going stale is normal and visible. A reason going stale is invisible, and it is what turns a good page into a page that is quietly wrong.
Three questions
Clear all three and session 6 gets stamped in your passport.
1 · Why does the one-page artefact carry a named section for what is deliberately not on the roadmap?
The entry that earns the section is the popular one. It passes everything except the gate that matters, so without a written reason it returns with momentum. D is close but wrong in a specific way: it frames omission as deferral, which is exactly the "not prioritised" non-reason that guarantees the conversation repeats.
2 · On the bench, what did dropping every candidate that does not clear the data gate today cost the plan?
Work that cannot be built contributes no fit while still consuming the estimate, so two thirds of the plan length was carrying nothing. Those items are not deleted, they move into the gaps section with a named blocker, which is a different kind of object: a gap you can choose to close rather than a promise you cannot keep. C is the answer that comes from confusing this row with the separate finding that the three most-requested new-data cases produced a fit of 0 over 25 weeks.
3 · Your business case needs to explain why the roadmap is short. What belongs in it?
None of the three survives inspection: one expired by its own stated terms, one traces to an unattributed remark on a conference panel, and one is called misguided by a study that read its sources. Quoting three unsound numbers is not triangulation, it is three times the exposure. The gate argument is stronger anyway, because it is about your board and anybody in the room can check it.
What this session teaches, and where it came from
Every fact on this page, and its verification tier, is recorded in the course's source map.
Session 6 cheat sheet · pin this
The spine is done. What it deliberately does not know is your industry, and that is what the two vertical roadmaps on the same shelf are for: learn-ship-ecommerce-enabler-with-phoebe maps a specific business model end to end, and learn-ship-platform-diagnosis-with-phoebe runs a five-year lookback on a real platform. Both import these seven sections rather than re-deriving them. Past that, the shelf splits by question: deciding what to build for one product rather than a portfolio is learn-product-thinking-with-phoebe; choosing what to build it on is learn-system-design-with-phoebe; reviewing a data product that already exists is learn-data-heuristics-with-phoebe; and working out whether what you shipped actually worked is learn-experimentation-with-phoebe.