SOVAL SPARK
Bring one unfinished thought.
Choose what you need, add any constraints, and leave with a reusable prompt or a starting point for your workroom.
YOUR SOVAL MAP
A first shape for the work
What Soval protected
Your exact words stayed separate. The interpretation is labeled. No project was formed, no content was accepted, and nothing was published.
Choosing to keep it holds this draft only in this browser tab while sign-in finishes. It is removed from temporary handoff storage on arrival and is not saved until you submit it in the private workroom.
Your starting point for useful work
Keep the human intent. Move the work.
Use AI without surrendering the thread. Soval keeps your original words, your boundaries, and every decision distinct from what a model suggests.
- Source snapshot
- 38a72622
- Digest-bound sources
- 233 files
- Current capability
- Prompt shaping · saved workflows
- Authority effect
- none
- Audience
- public guide · personal workroom
Start hereShape a prompt without an account. Sign in to save work and use AI drafting. Internal references require an invitation.
YOUR DATA. YOUR GUARDRAILS.
An LLM can give you an answer. Soval helps you control what that answer becomes.
Ordinary chat is excellent for generation. The hard part begins afterward: Which words were yours? What came from the model? What evidence supports it? Which boundaries still apply? Who decided it was ready?
What an LLM does
It responds to a prompt with useful language, analysis, plans, or code. The output can be strong, but confidence and repetition do not make it true or authoritative.
What Soval adds
It preserves the source, identifies derived work, carries boundaries forward, records provenance, survives interruption, and gives you a reviewable checkpoint before anything moves on.
Your first words remain a separate, digest-bound object. A suggestion cannot silently replace them.
Model output remains candidate material until a person deliberately decides otherwise.
Classification, lifecycle, evidence and authority effects remain visible as the work develops.
Request a Markdown checkpoint containing the source, developed objects, provenance and integrity hashes.
The model is not the product promise. The controlled record around the model is.
Bring Soval an ideaA utility for work that cannot afford drift.
Katra Forge carries an idea from deliberate capture to an approved Product Definition and typed handoff. Soval makes that path legible to people and usable by explicitly delegated agents.
Start with what you mean
Express the outcome naturally. Soval resolves the mechanics without silently inventing scope, answers, or project authority.
Know what supports each claim
Sources, decisions, candidate artifacts, repository coordinates, and verification stay distinguishable and inspectable.
Resume without rewriting history
Exact state, one primary line, explicit forks, stale-state detection, and idempotent retry make interrupted work recoverable.
Two views. One consistent system.
Soval sits above Katra Forge in the user-facing product stack while remaining a bounded application inside the complete logical system. Presentation position never transfers authority.
What people encounter
Soval is the front door.
A human or delegated agent meets Soval first. Soval sends typed governed requests into Katra Forge and presents the result with its provenance, state, and authority boundary intact.
Human or delegated agentattributable actor
Soval site applicationvisible surface
Katra Forgeservice and control layer
What the architecture contains
Katra Forge is the whole assembly.
Soval, Smusma, Sarek, Diftor, Kolinahr, The Telling, reusable modules, adapters, evidence, and release proof are bounded parts of Katra Forge—not peer systems that silently share authority.
Katra Forgecomplete governed system
Soval + control componentsbounded responsibilities
Applications + modulesdomain-specific authority
Adapters + evidencetyped external boundaries
From a thought to a handoff—without skipping the human.
The system separates inquiry, formation, approval, execution, verification, admission, and release so that progress remains real and recoverable.
Capture
Preserve the exact source and resolve related inquiry work.
Clarify
Develop evidence, requirements, unknowns, and decisions.
Form deliberately
Keep selection separate from verified formation.
Define
Compile a traceable Product Definition and candidate PRD.
Handoff
Bind human approval, admission, readback, and implementation lineage.
One surface. Bounded owners.
Soval coordinates the system but never inherits the authority of the component it invokes.
SmusmaInquiry continuity
SarekProduct control
DiftorConsequential disposition
KolinahrKnowledge and cognition
The TellingNarrative and canon
ModulesReusable capabilities
AdaptersTyped external boundaries
RepositoryDurable semantic record