What we scored and chose not to build

Four candidate shapes for scaling. Every one failed a constraint check. None got built.

This note is the sibling of what we deleted while shipping. Where that one is the deletion log for mechanisms we removed from unbin’s runtime after failed A/Bs, this one is the decline log for product shapes that never became code because they failed a scoring pass.

Toward the end of unbin’s build runway we ran a product-shape exercise: could the engine underneath unbin be scaled beyond it into something larger? Four candidates went into the exercise:

• Ship the whole runtime as a distributor-facing SaaS — a Würth-competing product with the same enrichment engine wrapped in a self-serve tier and a sales motion.
• Pivot the enrichment engine to a non-ETIM vertical — take the plumbing that works on electrical wholesale and re-point it at fashion, industrial supplies, or another domain where product-data enrichment carries the same underlying difficulty.
• Extract the LLM-evaluation and calibration layer as an audit-grade evaluation harness — sell the calibration monitor and the self-verify layer as their own tool for other teams’ runtimes.
• Leave the runtime where it is and monetise the applied-AI depth as consulting — no scale ambition, just the engine plus paid engagements on adjacent problems.

Each shape was scored across six constraint axes — Demand, Distribution, Legal, Technical, Positioning, Resource — with two independent readings per cell: a default lens (does this pass on its own merits?) and a deliberately hostile contrarian one (what if the framing is wrong?). Disagreements between the two lenses were preserved rather than smoothed, and load-bearing disagreements were escalated for a human tiebreak.

None of the four shapes cleared the constraint gate.

The pattern is: on every one of the four shapes, at least one axis scored a critical zero under the contrarian reading, and the tiebreak either confirmed the zero or produced an “unclear” verdict that the composition rule counts as unresolved. A scaling proposal that leaves a critical constraint unresolved does not get built.

Take the SaaS shape as the worked example. Two axes each scored a critical zero, and either alone was enough to fail it. Distribution: BARGO has no distributor channel into electrical-wholesale buyers and no realistic route to building one on the timeline the shape would require. Resource: the two-founder capacity ceiling is already the binding constraint on the client work that pays the bills; a self-serve tier and sales motion would push net-negative against it. Neither axis recovered under the contrarian reading, and no interior variant of the shape resolved both zeros at once. The reasoning for each of the other three shapes runs the same shape, against different axes, and is on file.

What this closed. unbin.io stays at its current shape, live, and the engine underneath continues to be maintained rather than expanded into a competing SaaS. What we declined is a packaged product line, not client work — production-AI engagements and bespoke enrichment against other dataset shapes remain how we operate; what does not happen is the engine expanding into a competing SaaS.

Why this is worth publishing. Scoring work that concluded “don’t build any of these” is a real output. It is also the kind of output that never gets a launch post, which is exactly why it is worth publishing — the shape of the discipline that keeps us from shipping the wrong thing is more useful evidence than a scaling announcement would be. What we disclose is the exercise’s outputs: the four shapes, the six axes they were scored against, the two-lens discipline, and the four “no” verdicts that resulted.

Same discipline as the sibling note: delete what fails a receipt, decline what fails a constraint check. Nothing gets built on hope.

Correction (2026-08-20): an earlier version of this note called this “a month of scoring work” — we cannot ground that duration in the record, so it is gone; the verdicts stand.