Anti-Platitude rule (5th anti-inflation rule, v1.1.0)

Motivation

The four pre-1.1.0 anti-inflation rules (Gravity, Burden of Proof, Antagonist Ceiling, Historical Context) defended against score inflation in general but did not name the single most-common failure mode in baseline AI lyric output: the closing platitude.

A line like "all I need is love" or "this is my truth" or "love wins" is technically grammatical, scans, and rhymes. The base rubric without explicit guidance was scoring these lines in the mid-band on Specificity (their imagery is "love", which is concrete-adjacent) and the mid-band on Voice (the narrator *sounds* sincere). The Anti-Platitude rule rebuckets them.

Proposal

Definition

A **platitude** is a line that resolves with a generic emotional summary using one or more of the following patterns:

1. Abstract noun + universal claim ("love wins", "truth matters") 2. First-person + universal need ("all I need is X") 3. Possessive + abstract noun ("this is my truth") 4. Imperative + abstraction ("believe in yourself") 5. Tautological emotional verb ("feel the feelings")

A line that uses one of these patterns AND adds no concrete detail (named object, named place, named person, time-marker, or sensory anchor) within ±1 line is platitude.

Scoring impact

When a line is identified as platitude:

• **Specificity** for the section containing the line drops to the lowest band (12–20 absolute points).
• **Voice** for the section drops by 8–15 points (a platitude reads as un-narrator-like by definition).
• **Truth** for the section drops by 5–10 points (a generic emotional summary is the rubric's anti-pattern for Truth).

Surface

The eval response now includes a `platitudes` field per section, listing every line flagged. The seal field's `rubricVersion` reads `1.1.0` for any score that applied this rule.

Examples

**Platitude (would trigger):**

• "All I need is love."
• "This is my truth."
• "Love wins in the end."

**Not platitude (named anchor within ±1 line):**

• "All I need is the key under the planter, the one she said she'd leave for me." (specific anchor)
• "Love wins in the kitchen at 11pm." (sensory + place)
• "This is my truth, told by the radio that got my mother through the chemo year." (named subject)

Out of scope

• The platitude detector itself (regex + LLM hybrid; lives in the eval pipeline, not in this RFC)
• Genre carve-outs (worship + gospel intentionally use the platitude pattern as part of the form; future RFC will document the per-genre carve-out logic)

Comment window

This RFC is open for comment until 2026-05-02. Email support@songforgeai.com with the subject `RFC-0002` to leave a comment.

Resolution

**Accepted as-written, 2026-05-02.**

Comment window closed without proposed amendments. Two clarifying questions received via support@songforgeai.com:

Q: Does the worship/gospel carve-out (out-of-scope §) imply some platitudes will be allowed? A: Yes. The form's repetition + universal-emotional-claim pattern IS the genre's intended structure. The carve-out itself will land as a follow-on RFC; until then the Anti-Platitude rule applies uniformly with the acknowledgment that gospel scores will read low on Voice + Specificity until the carve-out ships.

Q: How is "named anchor within ±1 line" measured — same line, prior line, next line, OR all three? A: Either prior or next line. The line containing the abstraction itself doesn't count — that would let "love wins" escape by adding "in 2026" on the same line, which isn't what the rule defends against.

The rule itself shipped with rubric v1.1.0 in B1240; this RFC is the formalization artifact RFC-0001 requires for MINOR bumps. Future amendments to the platitude pattern set ship as new RFCs (each pattern addition is a MINOR rubric change).