← the graph explorer

How scoring works.
The whole system, simply.

You score a few dozen nodes. The system covers all ~380 — honestly, with every derived number traceable to a human or AI judgment. Five ideas explain everything below: five reads per node · scores flow in the direction of their authority · “don’t know” is an answer · the map wears the scores · priority is computed, never opined.

1 · A node takes five reads

One weight, then three dimensions — each dimension read twice: where you are today, and the target for this company (not “max everything”). The gap between the two reads is the potential, and each gap type has a different fix.

Importance · 0–3
Does this matter for this company's motion? Weights everything downstream.
0 not relevant · 1 supporting · 2 core · 3 critical
Effectiveness
Does it produce its intended outcome, reliably?
gap = build it BETTER
Automation & AI
How is it executed — manual → AI-native?
gap = build it SMARTER
Scale
What volume does it reliably sustain?
gap = run it BIGGER
In the panel: tap a track once to set today, again to set the target — like picking hotel dates. Each 0–4 level has a written anchor, so a “3” means the same thing to you, to Avero, and to the AI. That shared language is what makes disagreement between scorers meaningful later.

2 · Scores flow in the direction of their authority

Nobody scores 380 nodes. You score where you have judgment; the system fills the rest with two currents and one precedence rule.

IMPORTANCE FLOWS DOWN MEASUREMENTS ROLL UP 2.4 derived — rolled up 3 scored — pins 2 3 3 inherited — a prior, not a fact —

Importance flows down. Say “Outbound is Critical” once — every capability inside inherits it until you override one. Say “Events: not relevant” and the whole subtree leaves the map, the math, and the roadmap.

Measurements roll up. A parent's Effectiveness, Automation & AI, and Scale are the importance-weighted average of its children — how built a channel is is how built its parts are, with the parts that matter counting more.

Your scores are pins. A hand score is a fact: it feeds the parent's roll-up upward, and acts as a starting assumption for everything below it. If your pin and the children's roll-up later disagree, nothing is overwritten — the node gets a tension flag, and a human decides. Every number on the map is one of four kinds:

1scored here 2rolled up from below 3inherited from above 4no signal
The rule: at the node > below the node > above the node > nothing. Evidence beats belief; belief beats blankness.

3 · “Don’t know” is an answer

Every track starts with a dashed ? cell. Tapping it is a real vote — “we examined this and can’t answer” — and it behaves completely differently from just not scoring:

? voted unknown

  • Counts as assessed — off the to-score list
  • today ? = a visibility gap → instrumentation & discovery work · target ? = an ambition gap → strategy work
  • Propagates: the subtree shows inherited unknown — an examined-but-unmeasurable zone, drawn on the map
  • Parent roll-ups get flagged partially known
  • Disagreement-ready: Client says 4, Avero says ? = “you claim a 4 — show us”

— no vote yet

  • Counts as unassessed — still on the list
  • Generates scoring work, nothing else
  • Propagates nothing, flags nothing
  • Silently absent from every average
  • Any first vote fills it

4 · The map wears the scores

No dashboards to open — the graph itself is the readout.

Tank fill
fill = built today · dashed waterline = the target
Halo = importance
strong = critical · soft = core · none = supporting
Dashed = tucked
supporting node folds its subtree · click to open
Events
Ghost = not relevant
washed out; its whole subtree leaves the map
The payoff: triage the ~11 channels first. Irrelevant wedges vanish, supporting ones fold, and what's left is the client's actual machine — with fill levels showing where it's weak and halos showing where it matters.

5 · Priority is computed, never opined

Nobody scores “priority.” It falls out of the reads above — which is what keeps the roadmap honest.

Priority
the roadmap rank
=
Importance
0–3 weight
×
Σ Gaps
target − today,
all three dimensions
÷
Effort
shortlist only
With unknowns it becomes an honest floor: Priority ≥ 6 · Effectiveness gap unknown — never a fake total. And when nothing is measurable yet: Priority unknown — which is precisely the case for a discovery engagement.

6 · Next: three scorers, one map

Today you score a single layer. With the backend, Client, Avero, and AI each score independently — the AI drafts first, humans correct what they disagree with — and nothing is averaged away. The distance between scorers becomes the headline: “You believe your Ads scoring is a 4. The audit sees a 2. The AI can't find it at all.” Each layer propagates on its own, so disagreement is visible even on derived nodes.

Status: explainer for the system as designed 2026-07-12. Live today: the five reads, the ? vote, tank/halo/tuck/ghost treatments, priority with floors. Designed, building next: roll-ups, inheritance, tension flags, multi-scorer layers. Full contract: the scoring & propagation spec · deeper rationale: the framework explainer — both internal working documents, not published here.