Score card
A score card reports how a piece of content scored against a model. It gives an overall score, a breakdown per trait, and — for each — a tier, a distance to the next tier, and a confidence reading.
§1What the card reports
Every score card carries the same four things. Each is a separate decision input and is meant to be read independently.
- An overall score on the 0–100 scale — the composite across the model's traits.
- A per-trait score on the same 0–100 scale for each trait declared on the model.
- A tier for each score — Strong, Solid, Developing, or Weak — and the headroom to the next tier.
- A confidence level for each score, rendered as dots: ●●● high, ●●○ moderate, ●○○ low.
The same fields appear in the JSON response from the REST API and the MCP tool output. The card is the visual rendering; the data is the same.
§2How to read a card
0
51
65
79
100
clarity
72
Solid
●●●
specificity
58
Developing
●●●
structure
83
Strong
●●○
Overall — harmonic mean across all traits, penalized by the weakest. The bottleneck drives improvement potential.
Bottleneckspecificity · +10 to Solid
Figure 1. An example card. The content scored Solid overall, held back by specificity; structure is already Strong.
Signals are identified below. The top block is the overall score; each row below is one trait.
1
Overall score
A single number summarizing the content. It is the composite across all traits and is penalized by the weakest trait, so it answers "is this good on every axis?" not "is this good on average?"
2
Position and tier
The marker shows where the overall score sits on the 0–100 axis. The tier label names the band — Strong, Solid, Developing, or Weak — and the number next to it is the headroom: how many points remain to reach the next tier.
3
Confidence
How reliable this score is, as a three-dot glyph: ●●● high, ●●○ moderate, ●○○ low. Low confidence means the model cannot discriminate the content precisely and the score should be read loosely. See confidence.
4
Per-trait row
One row per trait. Each row carries the same four readings as the overall block: a numeric score, a position on the axis, a tier label, and a confidence glyph. The row's color matches its tier.
5
Trait bar and tiers
The bar shades the four tier bands along the 0–100 axis. Hairlines mark the three breaks where one tier becomes the next. Each trait's breaks come from its own training data, so the bands can be wider or narrower from row to row.
§3Going further on any signal
Click any row to open its detail: the trait's description, its polarity labels (which end of the axis is which), and the headroom from the current score to the next tier on that trait. Click the overall block to return to the summary, which names the bottleneck trait — the weakest trait and the one the overall score is most sensitive to.
§4Edge cases
§4.1Single-trait models
A model with one trait still shows an overall block; the overall score equals that trait's score. The shape of the card does not change with the number of traits.
§4.2Low-confidence rows
When a trait's confidence is low, the numeric score is still shown but the tier label and headroom are withheld. The confidence glyph makes the reason explicit. Treat the score as directional, not exact.
§5Related concepts
- Composite — the overall score at the top of the card.
- Tiers — the four bands each score falls into.
- Breaks — the thresholds between tiers.
- Headroom — the distance from a score to the next tier.
- Confidence — the reliability reading shown as dots.