Skip to content
Glossary

Every term used inside SongForgeAI.

Songwriting is full of vocabulary that means something specific to practicing writers and something vague to everyone else. Here’s the full list, defined plainly, cross-referenced to the Berklee technique guides and the scoring rubric.

A/B verse structure
A verse that introduces two contrasting ideas (A and B) before the chorus ties them together. Common in country + singer-songwriter. Both halves of the verse are external (scene, action); the chorus resolves them internally.
Anti-inflation rules
Four rules built into the Lyric Scoring Standard that prevent the eval panel from drifting toward praise: Gravity Rule (default score is 50, not 75), Burden of Proof (claims need evidence), Antagonist Ceiling (weak metric caps composite), Historical Context Anchor (measure against what has actually been released).
Arrangement directive
A bracketed performance cue inside the lyrics that tells the singer or generator how to perform a section. Examples: [WHISPER], [FULL BAND DETONATION], [DEAD SILENCE]. Arrangement directives encode the Hayes/Brindell sparse-to-dense lift into the lyric itself.
Bridge
The section (usually single, near song-end) that shifts perspective — from external to internal, from narrator-A to narrator-B, from past to present. Bridges get a free pass on the Stolpe detail rule because the shift IS the point.
Burden of Proof
Anti-inflation rule: every high score must be justified by specific evidence in the lyric. The eval panel can't give Imagery a 90 without quoting the images.
Chorus-internal rule
Stolpe teaching: choruses lean internal (thought, emotion, narrator state). A chorus that reads all-external (scene only) feels like a report, not a payoff.
Cold reader
An adversarial re-read mode that reads the lyric as if hearing it for the first time, with no context on what the writer meant. Surfaces lines that DO NOT land with a stranger.
Composite score
The weighted total (0-100) from the 12-metric Lyric Scoring Standard. Craft 25% + Expression 40% + Impact 35%. A 50 is average. A 90+ is historically rare.
Consonant family
A group of consonants that substitute cleanly in rhymes because they share articulation. Labials (p/b/m/f/v), alveolars (t/d/n/l/s/z/r), velars (k/g/ng). Rhymes that match consonant family but not exact consonant count as 'family rhymes'.
Content word
A noun, verb, adjective, or adverb — words that carry meaning. Stressed in natural English speech. Should land on strong beats in a lyric and at line endings.
Craft Diagnostics
The collapsible panel on the forge result screen that surfaces meter variance, POV stability, and detail balance. Three axes the 12-metric rubric deliberately doesn’t cover.
Destination
Andrea Stolpe’s term for the single line an entire song walks toward. Named in a 4-part destination brief BEFORE writing: destination phrase, angle, sensory anchor, temporal frame.
Detail Balance
A diagnostic axis that reads every section and flags when a chorus leans external (should be internal per Stolpe) or a verse leans internal (should be external). Pre-choruses are checked for transitional balance; bridges pass freely.
External detail
Stolpe's term for concrete, image-evoking lyric content — action + object + scene. 'She stood at the kitchen sink.' Grounds the listener. Verses lean external.
Family rhyme
Two words that share a vowel + a consonant family but not exact consonants. "cat" and "cad" are a family rhyme (alveolar consonants). Stronger than slant, weaker than perfect.
Fingerprint
A song's unique stylistic signature — the combination of narrator voice, vocabulary, imagery choices, and rhythmic instincts that make it identifiably one writer's work. Personal Style Memory (PSM) captures this across sessions via starred lines.
Fire line
A line so strong it can't be forgotten. Josh Ritter's informal term for lines that transcend the song — the ones listeners quote back. The goal of Transcendent Line detection in the eval panel.
Fix Wounds
A refinement mode that targets only the specific lines the eval flagged as weak. Preserves everything else. Use when a song is mostly right but one or two lines drag it down.
Function word
An article, preposition, conjunction, pronoun, or auxiliary verb — words that carry grammar, not meaning. Unstressed in natural English speech. Should recede, not dominate the line endings.
Gauntlet
The automatic refinement pass that runs after every forge. Severity-routed (Polish at 85+, Targeted Surgery at 70-84, Structural Rebuild below). Takes eval wounds + prosody warnings + structure wounds + Stolpe violations as its fix roadmap.
Ghost collaborator
One of the 50 named voices in the war-room panel. Each has a distinct creative fingerprint; the model pulls 5-7 into an active-round bench based on what the song needs.
Gravity Rule
Anti-inflation rule: the default score for any metric is 50, not 75. The eval starts from 'average' and earns its way up, not from 'excellent' and negotiates down.
Hook
The musical + lyrical payoff line the rest of the song supports. Usually in the chorus. The strongest hooks state the song's title.
Internal detail
Stolpe's term for thought, emotion, or abstraction — lyric content that does NOT provoke an image. 'I wondered if love was enough.' Names the narrator’s state. Choruses lean internal.
Kintsugi protocol
A wound-classification rule for the gauntlet: lines marked VULNERABLE are the song's power — repair gently; lines marked WEAK are safe to rebuild from scratch. Named for the Japanese art of repairing pottery with gold.
Line-rewrite
A micro-iteration action on a single line — click any line on the result screen to open a rewrite popover that sends just that line to a Haiku-routed rewriter. Preserves surrounding context; zero full-song re-forge cost.
Listener Seat
A civilian voice on the war-room panel — not a writer, a receiver. Five permitted responses: 'I lean in,' 'I stop believing,' 'I recognize myself,' 'I admire but don’t feel,' 'I would replay this.'
Lyric Scoring Standard
The 12-metric open-standard rubric (CC-BY-4.0, v1.0) that powers SongForgeAI's eval panel. Craft tier + Expression tier + Impact tier with anti-inflation rules. Available at /scoring.
Meter variance
A diagnostic axis measuring how consistent syllable counts are within a section. Coefficient of variation under 0.15 = tight; above 0.30 = draft. Not a rubric metric; a check the rubric doesn’t cover.
Percentile anchor
The 'Top X%' label attached to every score chip — tells the user where their score sits in the full population of forges. Makes composite scores interpretable without knowing the rubric.
Perfect rhyme
Two words with identical rhyme units (same vowel + same trailing consonants). "cat" and "hat". "fire" and "desire". The tightest form of end-rhyme.
Personal Style Memory (PSM)
Star any line on any song and it becomes a style anchor in future forges. Top 8 anchors inject into the forge prompt; the model is instructed NOT to quote them verbatim. Capped at 50 stored lines with automatic decay for unused anchors.
POV stability
A diagnostic axis that tracks first-, second-, and third-person pronoun drift across a song. Stable POV within a section is the default; mid-song pronoun drift almost always reads as a draft, not a choice. Bridges get a pass.
Pre-chorus
The section between verse and chorus. Stolpe teaches pre-chorus as transitional — it should mix external + internal detail, tipping the narrator from scene toward claim. Not a verse, not a chorus; an explicit bridge between them.
Prosody
The match between the rhythm of the language and the rhythm of the music. Stressed content words want strong beats; function words want weak beats. Pat Pattison's Berklee teaching: 'Preserve the natural shape of the language.'
Rhyme family
The classification of a rhyme as perfect, family, slant, or none. Perfect is identical; family shares vowel + consonant class; slant shares vowel OR consonant but not both; none misses both.
Rhyme unit
The end-of-word phonetic signature a rhyme needs to match. Usually the last stressed vowel + all following consonants. 'late' = 'ayt', 'wait' = 'ayt' — same unit, perfect rhyme.
Scheme (rhyme scheme)
The pattern of rhymes across line endings. AABB (couplets), ABAB (alternating), ABCB (hymn meter), AAAA (monorhyme). SongForgeAI detects the scheme from the first 8 lines of each forge and grades rhyme tightness from adjacent-pair matches.
Silent-e rule
English spelling convention: a word ending in consonant + 'e' has a long vowel before the consonant. 'cake' = 'ayk'. The rhyme analyzer upgrades silent-e vowels to their long form so 'late' rhymes with 'eight'.
Singability
The ease with which a line can be sung without fighting the melody. Long lines (>14 syllables), very long (>20), stress clusters (5+ consecutive stressed words), and weak endings all reduce singability. Flagged client-side by the prosody lint.
Slant rhyme
Two words with a similar but imperfect match — either same vowel with different consonant family, or same consonants with a slightly different vowel. Pro songwriters use slant rhymes deliberately to avoid predictability.
Stolpe rule
The core Stolpe teaching baked into SongForgeAI: verses lean external, choruses lean internal, pre-choruses transition, bridges flex. Violations get flagged as wounds the gauntlet fixes.
Stress cluster
Five or more consecutive stressed content words — 'dark cold hard stone walls.' Reads as stomping; resists melodic shape. The prosody lint flags these as singability warnings.
Structural Architecture
Metric #2 on the 12-metric rubric — does the song have verse/chorus/bridge scaffolding, and does each section do a job the others can't? Deterministic structure validator feeds this metric as wounds.
Suno-ready
Lyrics formatted with bracketed performance directives + section markers that Suno's AI music generator reads as arrangement cues. [VERSE 1], [CHORUS], [WHISPER], [DEAD SILENCE].
Super Boost
An iterative refinement mode targeting composite 90+. Runs up to 5 additional gauntlet rounds with plateau detection, keeping the highest-scoring version across all iterations.
Transcendent line
A line so strong the eval panel quotes it in the summary — the fire-line candidate the song built toward. Bolded on the forge result screen and promoted into the Strongest Lines callout.
Voice panel
The 50-voice war-room catalog + 20-voice Infinity Mode panel. Active-round bench pulls 5-7 voices per round based on what the song needs. Pattison (Prosodist) and Stolpe (Destination) are the Berklee-grounded additions from Build 938.
Voltage
A 0-100 creativity dial on the forge. Low voltage = conservative, expected choices; high voltage = aggressive, experimental. Maps to Claude API temperature internally (0.85 at voltage 0 → 1.0 at voltage 50+).
War room
The metaphorical single round of the forge — 50 legendary songwriters arguing across multiple phases (Forge, Crucible, Critical Eye, Industry + Final) until the song locks. Produces war-room notes visible on the result screen.
Weak ending
A line that ends on a function word ('in', 'of', 'the'). The pause after a line naturally emphasizes the last word; landing that emphasis on an unstressed word drains impact. Flagged as a prosody warning.
Wound
A specific line or section the eval panel + prosody lint + structure validator flagged as weak. Wounds are fed to the gauntlet as its fix roadmap. Each wound names the diagnosis and points at the line.

Put the vocabulary to work.

Every term above is an instrument the forge pipeline uses on every song. Write one, see which axes light up, which warnings fire.

Forge a song