# Round 3 scorecard — weighted re-grade (2026-07-01)

Three synthesis variants, same five personas (each re-read their own round-2
review for continuity). Composite weighting per Michelle: **chair (G3) and
mid-career doc (G2) count double** — weighted mean = (G1 + 2·G2 + 2·G3 + G4
+ G5) / 7, normalized to /100.

## Scores (/60)

| Variant | G1 Brooks | G2 Patel ×2 | G3 Okafor ×2 | G4 Adler | G5 Webb | Unweighted /100 | **Weighted /100** |
|---|---|---|---|---|---|---|---|
| A — Field Notes, refined | 54 | 52 | 45 | 51 | 54 | **85.3** | 84.0 |
| B — Indigo Oath, de-templated | 50 | 51 | 51 | 47 | 55 | 84.7 | **84.8** |
| C — Paper & Indigo blend | 48 | 50 | 44 | 45 | 49 | 78.7 | 78.6 |

## Per-grader ranks

| Variant | G1 | G2 | G3 | G4 | G5 |
|---|---|---|---|---|---|
| A | 1 | 1 | 2 | 1 | 2 |
| B | 2 | 2 | 1 | 2 | 1 |
| C | 3 | 3 | 3 | 3 | 3 |

## Reading the result

**A and B are statistically tied.** Unweighted, A wins by 0.6; with
Michelle's weighting (chair + mid-career double), B wins by 0.7. Both are
top-2 for every grader; C is a unanimous third. The graders converged on a
clean trade:

- **A wins on register.** Brooks gave it her highest score of either round
  (54: "every round-2 objection answered"); Patel: "grafted B's biggest
  strength onto the register Dev already trusted"; Adler kept it #1 ("fixed
  the card-lift/plate/dead-air objections in its own language").
- **B wins on structure.** Okafor: the brass index-tab catalog system "reads
  like a bound course catalog," timeline fixed outright; Webb gave it the
  round's top score (55: the anchor-linked syllabus card is "index → detail,"
  instant-render CTA). But Adler: the tab "half-earns 'signature' — a
  monogram pinned on the same template suit," and Patel still finds it
  "too formal for a reader who came to confess."
- **C validated the palette, not the page.** Best contrast discipline of the
  round (in-code WCAG audit praised by four graders), but it re-imported two
  round-2 sins (dimming, reduced-motion nuke), hides the nav CTA on mobile,
  and reads as "a collage of A's and B's devices."

**The decision is register vs structure:** the warm colleague's journal (A)
or the institutional index (B). Both now carry the same conversion machinery.

## Fix list for the chosen variant (from the reviews)

**If A:** un-clip the Wednesday scatter on ≤390px phones; replace the
hard-coded 25.7rem slip box with content-driven sizing; fix the 960px
two-column curriculum stagger scramble; remove the SaaS lift-glow from the
CTA button; add a mid-page CTA (chair asked twice); give 201–203 one degree
more presence than a dagger footnote; consider making the hero syllabus rows
anchor-link like B's.

**If B:** restore some ceremony to the private-by-design moment (Patel: the
lost indigo band was "the round's biggest demotion cost" — a quiet indigo
keyline card near How-it-works, not a full-bleed section); make the timeline
add more than titles (Webb: longest section, still repeats the hero card);
fix the double "Next step" label in the finale; warm the register one notch
(Patel: "too formal to confess to"); give 201–203 more than a footnote
(chair, twice).

**C regressions worth harvesting nowhere:** none — but its in-code WCAG
contrast-audit practice should be adopted in the final build.

Full reviews: `g1-eleanor-brooks.md` · `g2-dev-patel.md` ·
`g3-grace-okafor.md` · `g4-simone-adler.md` · `g5-marcus-webb.md`.