Document metadata
type: session
status: established
date: 2026-09-12
topic: Voice Key authorial decision fingerprint
authority: direct author decision for product direction; implementation mechanics proposed
scope: conversational authorship system and Seeds Composition Engine
Voice Key Authorial Decision Fingerprint
Author-established direction
The product feature is named Voice Key.
Voice Key must not primarily imitate the way a user speaks. Most people do not speak in a form that should be translated directly into finished prose. Natural transcripts remain valuable because they reveal how a person notices information, connects causes and consequences, evaluates success and failure, allocates attention, handles uncertainty, and chooses what to explain or leave open.
Voice Key uses those deeper patterns to influence how AI writing presents information. It should shape authorial selection while a capable writing model supplies the craft required for clear, high-quality prose.
Surface voice imitation remains a possible secondary control. It is not the core value.
Four-layer functionality locked
The author established Voice Key as a connected four-layer system:
- Authorial Decision Fingerprint controls attention, causality, evidence, judgment, ambiguity, surprise, emotional distance, compression, correction, and closure.
- Prose Topology Profile controls the physical shape of the writing: sentence-length distribution, fragments, paragraph shape, list structure, transition density, thesis placement, and paragraph closure.
- Residue Filter detects inflated language, generic metaphors, unnecessary summaries, formal transition crutches, AI-favored vocabulary, and explanations that repeat what the scene already demonstrated.
- Pattern Memory tracks recently used structures and rhetorical devices so a useful technique does not become a new repetitive formula.
Voice Key stores ranges, distributions, evidence, and recent-use history rather than universal commands. Fragments, long-short-short rhythm, delayed theses, asymmetrical lists, and unresolved paragraph endings are available choices, not mandatory signs of human writing. The system must vary them according to the author, project, passage purpose, and surrounding prose.
The final check evaluates excessive regularity without automatically demanding more disorder. Readability, meaning, canon, scene function, and author approval remain higher priorities than irregularity.
Working product description
Most AI writing tools try to copy the way a person sounds. Voice Key goes deeper than that. It learns how someone naturally chooses, organizes, and presents information, then uses those decisions to shape better AI-generated writing. Voice Key turns natural conversations and creative choices into a unique writing fingerprint. Instead of simply copying how someone speaks, it learns what they notice, what they consider convincing evidence, how they connect causes and consequences, how they judge success or failure, what they prefer to explain, and what they intentionally leave unresolved. The system uses that fingerprint to choose between different high-quality ways of structuring a passage, revealing information, introducing constraints, creating earned surprises, and communicating emotion. The result is writing shaped by the user’s creative judgment and way of understanding the world, while still benefiting from the clarity, structure, and technical ability of a professional writing model.
Initial interview experiment
Eight natural dictation prompts tested narrative memory, correction, system explanation, observation, persuasion, speculation, practical judgment, and product criticism. The experiment showed that surface phrasing was the least interesting output. The useful repeated tendencies were decision-level patterns:
- present conclusions through visible operation and evidence;
- treat constraints as causes that redirect action;
- judge success according to the real purpose;
- compare formal systems with lived behavior;
- let meaning emerge through accumulated behavioral changes;
- close the main obligation while preserving purposeful openness;
- separate the enjoyment of production from the quality of the result;
- judge technology by whether it strengthens human capability;
- qualify criticism through experience and evidence;
- distribute detail asymmetrically rather than explaining everything equally.
These are provisional extracted tendencies, not permanent claims about the author. Creative-choice exercises and project-specific comparisons must validate or revise them.
Proposed system placement
Voice Key belongs inside the Composition Engine after canon and scene-contract validation but before final prose rendering. It receives several structurally valid presentation choices, filters out weak or canon-breaking options, and ranks the remaining choices using the general author fingerprint, the active project fingerprint, and the purpose of the passage.
The selection can change information order, focal detail, degree of explanation, timing of recognition, use of constraints, emotional distance, and closure. It cannot silently change story facts or settle unresolved author decisions.
Seeds proving-ground use
For Seeds of the Throne, the initial project fingerprint favors operational evidence, constraint-driven turns, competing definitions of victory, formal-system versus lived-reality gaps, behavioral proof of Luminai integration, delayed recognition, and earned closure with wider implications left open.
This is a writing-system direction, not new story canon. The first proposed test remains the already approved Book One outcome-presentation vertical slice.
Quality and detector boundary
Voice Key should create meaningful non-default choices at the narrative level. It should not add random errors, damage grammar, maximize statistical perplexity, or promise detector evasion. Quality, canon integrity, source grounding, and author approval remain hard gates.
At the prose level, it may break unwanted symmetry, vary sentence and paragraph rhythms, remove generic vocabulary, change thesis placement, and trust readers to infer demonstrated meaning. Those changes must serve the passage rather than imitate a fixed collection of supposed human tells.
Next validation
Run a blind three-way comparison on one bounded scene contract:
- strong project-aware baseline;
- surface speech-profile revision;
- Voice Key authorial-selection reconstruction.
Evaluate which version is best, most specific to Seeds, most causally convincing, least generically polished, and strongest in its presentation decisions. Do not ask only which version sounds most like the author.
Later guided-transcript workshop
The author requested a dedicated workshop for gathering more Voice Key evidence through guided transcripts. The planned 07 Coordination/Authoring System/Voice Key Guided Transcript Workshop contains eight resumable 10–15 minute modules spanning natural reasoning, attention, causality, evidence, reveal order, ambiguity, prose topology, residue, and Seeds-specific blind calibration.
The workshop is deliberately scheduled for later and does not replace the active Book One Architecture Workshop. Its start command is Open the Voice Key workshop.
