The B4 Framework — methodology
Two axes.
Four verdicts.
X · Strategic differentiation: does owning this system make you different in a way customers pay for?
Y · AI feasibility: could a competent team, with today's models, actually build and run it?
Every category's position falls out of five dimensions, scored 1 to 5. The call shows how stable it is, and the org lens adjusts for the team’s engineering and AI capacity.
Methodology v4.0 — August 2026
Consider owning it. The category differentiates you competitively, and the evidence makes building a scoped capability realistic. Test the business case and your capacity to run it. Teams should build the software that gives them competitive advantages.
The five dimensions — scored 1 to 5
Specificity
Sets the X axis
How deeply the system encodes processes that are yours: data models, rules, workflows nobody else runs. Higher specificity means you need more customization, and buying generally costs more in integration and customization.
Strategic control
Sets the X axis
Are you going to be faster, better, or cheaper than your competitors by building a custom version of this software for your business?
AI feasibility
Sets the Y axis · evidence-graded
Buildability: can a competent team build and operate a custom alternative? Grounded in evidence-based research. AI feasibility doesn't score high unless teams have already built it in production, and a category whose only evidence is bolted-on AI features caps at 2.
Vendor value
Urgency modifier
How much value the vendor you're paying for offers. Will you use 20% of the features or 80%? It helps you determine how fast to act inside a quadrant.
Cost trajectory
Urgency modifier
Where the pricing is headed. Compare vendor cost with the full cost of building and operating the same scope. A widening gap makes investigation more urgent; it does not change the quadrant.
How the verdict is decided
Methodology v4.0 — August 2026
A score is a reading, so the verdict is a range.
Five people scoring the same category do not land on the same number. They land near it. So publishing one badge per category and stopping there says more than the evidence supports.
Here is what that cost. The X axis is the average of two whole numbers, which means it lands on 3.5 a lot — and 3.5 is exactly the line between the quadrants. Roughly one category in seven sits on it. Under the old math, which side those fell on came down to how a comparison was written. Around three quarters of the index is within a single one-point move of a different badge.
So the index stopped publishing a point. Each category now carries a scenario share across all four calls, the leading call, and a word for how stable that classification is. Same five dimensions, same 1-to-5 rubrics, same scores. What changed is the arithmetic that turns them into an answer.
Every score gets a band
Three of the five dimensions decide the quadrant. Each one is read as a range, not a point: one step either side of what it scored, equal weight on all three values. A dimension at 3 is treated as maybe 2, maybe 3, maybe 4. At the edges, values repeat: a 5 reads 4, 5, 5; a 1 reads 1, 1, 2. All three readings keep equal weight.
The band is enumerated, not simulated
Three dimensions, three values each, is 27 combinations. Every one of them gets placed on the map and its weight added to whichever quadrant it lands in. That produces a scenario share for all four calls. The shares measure sensitivity to the ratings, not the probability that a project will succeed. No sampling, no randomness — run it twice and you get the same answer to the last decimal.
An axis is high only at 4 or above
The line sits at 3.5, and a category has to clear it outright. Sitting exactly on the line reads as low. X is the average of two whole numbers, so it lands on 3.5 often — and that used to be settled by which way a comparison operator happened to point. Now it is settled by a rule you can read.
Ties fall toward the cheaper mistake
When two calls hold the same weight, the order is Buy, then Bridge, then Beware, then Build. Telling a team to build something they should have bought is the most expensive sentence this index can say, so Build is the last call a tie can produce. That is a stated principle, not an implementation detail.
The call shows how stable it is
The quadrant holding the most weight is the call. Alongside it: clear when it holds 70% or more, lean from 50% to under 70%, split below 50. A near-call flag when the runner-up is within 15 points. Most of the index is a lean — that is the honest reading, and it was invisible before.
The org-maturity lens
"Could a team build this" depends on whose team.
Start with raw feasibility for competent engineers using AI. The small lens shifts it down one step for a team without dedicated engineering; medium leaves it unchanged; large shifts it up one for an AI-mature platform team that ships reliably. Apply the lens once, then clamp the center to 1–5 before constructing its band. Choose by capacity, not headcount. Every other rating stays put.
It works like a view setting. Three pills, switch them as often as you like, and your pick stays in your browser. In the August v4 launch snapshot, 51% of categories had different primary verdicts at the small and large lenses. Pro members and the MCP get the lens; everywhere else shows the medium reading.
The band and the lens do different jobs. The band says how far a reading might drift. The lens re-asks the question for a different team, and that answer gets its own scenario tally and confidence label.
Trust is built in
01
Categories, not vendors
Vendors churn, get acquired, pivot. A category's structural position moves slowly — which is what makes a score worth publishing and re-checking.
02
It re-scores itself
A nightly job scores against the quarterly publication deadline, and the AI frontier gets re-assessed every week. Every score lands as an append-only snapshot with its full five-dimension state. 1,850 and counting.
03
No vendor money
Analysts get paid by the vendors they rate. B4's one revenue source is the buyer. Only 4% of the index says build, and a framework that says "build everything" is a marketing document.
Run it on your own stack.
Search every scored category, check the dimensions, and challenge a verdict. See plans for the full directory, filters, and exports. Or connect your AI directly to the database through MCP.