# Criteria sets — axis definitions and scoring anchors

Six axes, scored 1–10. The set is chosen at F0 and **locked for the run**. Changing it
afterwards means re-scoring everything from scratch — never silently, because a table
built from two different rubrics looks exactly like a table built from one.

---

## Why six, and why these

Fewer than five axes and everything collapses into "good idea / bad idea". More than
seven and the scores stop being distinguishable from each other — the reviewer starts
assigning the same number across axes, which is the same as having one axis with extra
steps.

The constant across all four sets is the shape: **two axes for upside, one for the thing
that could go wrong, one for whether you can actually do it, one for time, one for cost.**
Only the vocabulary changes, so a person can move between sets without relearning.

---

## BIZ — revenue, cost, customers, suppliers

| Axis | Question | 2 | 5 | 9 |
|---|---|---|---|---|
| **E** effect | How much does this move the number? | noise | noticeable in a quarter | changes the trajectory |
| **R** risk | What breaks if this is wrong? | recoverable in a day | costs a month | irreversible or reputational |
| **F** feasibility | Can it be done with what exists? | needs new capability | needs a new process | doable next week |
| **S** speed | Time to first result | > 6 months | this quarter | days |
| **P** strategic fit | Does it compound with what we already do? | orthogonal | adjacent | reinforces the core |
| **K** cost | Money and attention | large and recurring | one-off, meaningful | rounding error |

> **R and K are scored inverted:** 9 means *low* risk and *low* cost. Every axis points
> the same direction, so the weighted sum needs no sign juggling — the single most common
> source of ranking errors in hand-built versions of this.

## TECH — architecture, tooling, systems

Same shape, different vocabulary: **E** impact · **R** risk · **F** feasibility ·
**S** speed · **P** maintainability · **K** cost.

`P` is the axis people underrate here. A change that works today and needs a custodian
forever should not outrank one that works today and needs nobody. Maintainability is
where that difference gets recorded.

## PERSONAL — career, life, direction

**E** payoff · **R** risk · **F** feasibility · **S** speed · **P** values fit · **K** cost.

`P` carries the most weight in practice even when the numbers say otherwise. If an option
scores 9 everywhere and 3 on values fit, the ranking is telling you something the total
is hiding — this is exactly the case the bias audit exists to catch.

## DECISION — comparing options that already exist

**E** upside · **R** downside · **F** reversibility · **S** speed · **P** fit · **K** cost.

> If you are here, consider whether you need this skill at all. Brainstorming generates
> options; comparing three you already have is a different job. Use this set only when
> the options are genuinely open-ended.

`F` becomes **reversibility** — how cheaply can this be undone. It deserves its own axis
because a reversible decision at 70% confidence and an irreversible one at 70% confidence
are not the same decision.

---

## Scoring discipline

**Anchor first.** Before scoring the field, pick three ideas and score them with written
reasoning. Everything else gets scored relative to those. Without anchors, scores drift
across a long session — early ideas get scored generously, late ones harshly, and the
ranking partly encodes when something was written down.

**Score the mechanism, not the sentence.** An idea with no `mech` field cannot be scored
on feasibility, so it ends up scored on how confident it sounds. The script flags these.

**Use the full range.** If nothing scores below 4, the scale has collapsed to 4–9 and the
weights are doing nothing. Something in the field should be a 2.

**Never adjust a score to change a rank.** If the ranking looks wrong, the rubric or the
weights are wrong — fix those and re-run. Adjusting scores to reach a preferred answer
converts the whole exercise into a justification with arithmetic attached.