Product architecture
Define Libraries, permitted source catalogues, knowledge boundaries, human authority and continuity rules.
Live product testing · persistent AI GM · governance under play
First Fracture
The operational testing model for Briana's larger AI-guided roleplay architecture: a persistent campaign used to test continuity, player agency, evidence separation, recovery and bounded access to governed lore.
Approved original Campaign-character visual used here as evidence of exact-asset canon control. The image identifies an approved file; it does not override written Campaign state.
Operational evidence
A design tested against a living world.
First Fracture is not presented as a generic concept demo. Its local runtime was checked read-only on 25 August 2026 and reported healthy at event 156. The figures below describe this Campaign only.
156established events in the persistent Campaign log
35persistent actors represented in the runtime
141adjudicated outcome records
485knowledge-ledger records supporting bounded character knowledge
Read-only runtime census: 25 August 2026. Database quick-check returned OK; foreign-key errors: 0. Counts are different record classes and are not additive.
Architecture-to-operations loop
The Campaign is the proving ground.
The broader architecture defines how governed knowledge should become playable context. First Fracture supplies the operational pressure that reveals whether those controls survive continuity, recovery and unpredictable player action.
First Fracture runtime
Exercise the rules through a persistent campaign with live actors, decisions, consequences and recovery needs.
Observed failures
Expose source confusion, canon drift, identity errors, recovery gaps and unsafe assumptions as concrete defects.
Governed correction
Append corrections, strengthen acceptance criteria and return evidence-backed lessons to the product design.
Capabilities under test
What persistent play actually exercises.
Persistent world state
Actors retain motives, limits, relationships and off-screen activity while consequences and world clocks continue beyond a single exchange.
Player agency
The GM may adjudicate the world, but it may not invent the player character's actions, consent or decisions.
Epistemic separation
Character knowledge, GM-sealed information, technical state and shared Campaign facts are kept distinct rather than collapsed into one context.
Append-only correction
Established history is not silently rewritten. Corrections supersede earlier records while preserving what changed and why.
Recovery discipline
After interruption, the instance verifies the authoritative runtime edge before resuming play and checks the append after every write.
Bounded lore access
Reference material remains separate from Campaign canon until a specifically authorised, fail-closed adapter action permits a narrow read.
Evidence maturity
Three layers. Three different claims.
The case study does not use one successful check to imply that every planned capability exists.
Persistent Campaign runtime
The live local database reached event 156 and passed a current read-only integrity check. It contains persistent actors, outcomes, consequences, knowledge records and world clocks used to run the Campaign.
Library of Cybertron adapter
The corrected v1.2.1 release passed 53 of 53 executable tests, including fail-closed, isolation, injection-shaped input, concurrency and continuity-separation checks. It was not connected to the live Campaign. A separately authorised read-only v1.2.0 handshake at event 142 imported zero claims and changed no Campaign state.
Red Leader Stage One
A forty-test acceptance specification covers authority, privacy, hidden adjudication, transaction recovery, player agency and lifecycle controls. It remains a working design specification: implementation, execution and deployment are not claimed.
Briana's role
Product authority and test direction.
Briana directs the system as Campaign authority, product architect and adversarial reviewer. AI collaborators provide implementation, narration, drafting and analysis within the boundaries she sets and approves.
- Defines the human-authority and player-agency boundaries.
- Tests whether proposed controls survive real Campaign use.
- Challenges overreach, source confusion and unsupported canon claims.
- Separates proposed, implemented, verified and deployed states.
- Maintains final authority over Campaign facts, exact visual canon and publication.
Rights and presentation boundary
Original systems work inside an unofficial fan project.
This is an unofficial, non-commercial fan-project case study. Transformers and related intellectual property belong to their respective rights holders. This page documents Briana's original system architecture, AI direction, governance, Campaign operations and original contributions—not ownership of the underlying franchise.