Case 003 — Watching the Anti-Platitude rule fire on a borderline draft
A draft that scored 71 on rubric v1.0.1 will drop into the C-band on v1.1.0 because of two closing platitudes the older rubric scored mid-band on Specificity. The Anti-Platitude rule should reveal the underlying voice problem rather than punish it arbitrarily.
Starting point
I built the porch railing myself last summer Mixed concrete in a wheelbarrow my dad left me All I really need is love, all I really need This is my truth, told slant
Lines 1-2 carry strong Specificity (named tools, inherited object). Lines 3-4 are textbook platitudes that v1.0.1 scored at mid-band on Specificity (because "love" reads as concrete-adjacent). v1.1.0 Anti-Platitude rule rebuckets them.
Result
(same lyric, no edits — just a rubric version bump)
v1.1.0 Anti-Platitude rule fires on lines 3-4. Specificity drops from 68 to 22; Voice from 70 to 30. The composite collapse exposes a real voice problem the older rubric was masking. The takeaway is not "rubric got harsher" — it is "the rubric now names what was already wrong." The fix is to replace lines 3-4 with concrete anchors; that work is documented as case 004 (next).
Pipeline
Lessons
- A rubric MINOR bump that drops a borderline-B into the D-band is the rubric working correctly — it identifies the underlying weakness rather than averaging it away.
- Platitudes are the most-common reason a draft scores in the high-60s on v1.0.1 instead of clearing 75 — the rule names what was always true.
- The reproducibility seal carrying rubricVersion is load-bearing here: a third party comparing scores across the bump knows exactly which rubric scored each.
- Surface specificity (named tools) cannot rescue a closing platitude; the close is the load-bearing position.
Want a case run on your own lyric?
Email support@songforgeai.com with your lyric + the hypothesis you want tested. Selected cases ship as their own public entry under the same strict format.
Or run it yourself in the forge