Reality Interpreter
Embedded in LeapKeep Core

Dorra is the
Reality Interpreter
inside LeapKeep.

Dorra turns software change into consequence-aware intelligence by combining live system context, historical memory, and operational boundaries — so teams understand what changed, why it matters, and what to do next.

Open DORRA StudioDORRA in System Space
DORRA V-1000Reality Interpreter · Context Assembly Active
● Processing
Change Inputmigration · config · code deltaContext Graph9 nodes · 11 edges · live stateMemory Stackepisodic · semantic · proceduralPolicy / Doctrineclearance rules · escalation gatesEvidence Engineretrieval ID · confidence · flagsInterpretationrole-aware · consequence-mapped
context graph: live
memory: 14 episodes
policy: v2.4 active
confidence: 0.82
Positioning

Most AI tools answer prompts.
Dorra interprets reality.

Instead of responding from a blank context window, Dorra assembles live system state, dependency relationships, prior outcomes, and policy boundaries before it answers. That makes its output grounded in the software world it serves — not just the text it receives.

The result is interpretation with memory, context, and consequence — not a sophisticated autocomplete.

Generic AI model
Blank context window at query time
No memory of prior decisions
No access to live system state
Cannot trace blast radius
No governed output boundaries
Dorra V-1000
Live context graph assembled first
Episodic + semantic + procedural memory
System Space as reasoning substrate
Consequence-mapped blast radius tracing
Governed by explicit policy doctrine
Core capabilities

The five capabilities

01

Live Context Assembly

Dorra pulls together the right system context at the moment of use: services, dependencies, configs, environments, deploy targets, and change surfaces. It does not guess from training data — it reads from the live graph.

Every interpretation starts with a context assembly pass across System Space. Dorra identifies which nodes, edges, and metadata are relevant to the query before forming any response.
02

Consequence Interpretation

Dorra explains what a change means, not just what changed. It traces likely blast radius, risk concentration, and operational consequence across the live stack — with specificity, not generality.

Risk is mapped to actual services, actual query paths, and actual ownership boundaries. Output reads like an expert explaining a system, not a model explaining text about a system.
03

Memory-Backed Intelligence

Dorra retrieves past incidents, prior clearances, rollbacks, overrides, and similar changes so today's decision benefits from yesterday's outcomes. Every answer is informed by what actually happened.

Four memory tiers: working memory (current session), episodic memory (past clearances and incidents), semantic memory (normalized facts), procedural memory (investigation sequences and doctrine).
04

Role-Aware Translation

Dorra explains the same change differently for builders, operators, founders, and technical leaders. One system serves multiple decision-makers without collapsing into jargon or over-simplification.

A builder sees affected query paths. An operator sees rollback options and service exposure. A founder sees business impact and timeline. A technical leader sees architectural consequence and policy alignment.
05

Bounded Action Guidance

Dorra recommends the next safest move within explicit constraints, approval gates, and policy boundaries — then knows when to escalate instead of improvising. It suggests. It does not act alone.

No silent execution. No direct doctrine changes. No production-level mutation without approval. Every suggested action is logged, attributable, and reviewable before it becomes anything real.
Architecture

The Dorra architecture stack

Seven layers that make Dorra a context-and-consequence engine — not a prompt wrapper. Each layer is proprietary, versioned, and governed.

1
Context Graph Layer
Dorra's reasoning substrate — not documentation.
servicesAPIsdatabasesrepossecrets / configenvironmentsdeploy targetsrisk zonesownership maps
2
Memory Stack
Stronger than vector RAG — relational + episodic.
Working memory: current session + active evidenceEpisodic memory: clearances, incidents, rollbacks, overridesSemantic memory: normalized system facts, patterns, policiesProcedural memory: investigation sequences, escalation doctrine
3
Policy / Doctrine Layer
Workers read doctrine. Only governed flows write it.
clearance policiesescalation rulesapproval requirementsenvironment restrictionsconfidence thresholdsnever-do-this boundaries
4
Role Engine
Same reality. Different narrative. No jargon collapse.
builder framingoperator framingfounder framingtechnical leader framingauditor / security reviewer framing
5
Evidence Engine
Outputs are inspectable. Not mystical.
evidence pointersretrieval set IDmemory references usedpolicy version at query timeconfidence leveluncertainty flags
6
Bounded Action Layer
Dorra suggests or queues. It does not act alone.
no silent executionno direct doctrine changesno production mutation without approvalall actions logged + attributable
7
Governance Log
Trust layer and training asset — versioned, tamper-evident.
prompts + retrieved contextmemory items usedpolicy set appliedtools calledoutputs givenhuman overridesfinal outcomes
The non-replicable principle

Dorra does not just answer from data. It answers from structured, governed, living operational memory.

That is what makes it more than a wrapper — and what makes it increasingly hard to displace once it is installed inside a clearance workflow.

The moat

Why Dorra is different

Dorra is not a chatbot layered onto software. It is a context engine wired into software reality. Generic models reason over prompts. Dorra reasons over a living context graph, persistent memory, and a governed access model designed specifically for software change.

A competitor can copy
Your UI
Your prompt templates
Your model provider
Parts of your feature list
Your landing page copy
They cannot copy
Context graph — live, proprietary, yours
Event + decision memory — versioned, owned
Evidence-linked governance log
Role-aware translation layer
Policy doctrine — governed, non-rewritable
Installed position inside clearance workflow
Dorra moat
Proprietary context graph
Event and decision memory
Evidence-linked governance log
Role-aware translation layer
LeapKeep OS moat
Workflow placement inside clearance
Clearance authority — not AI authority
Decision history and rollback record
Policy doctrine evolution over time
Trust & boundaries

Dorra assists interpretation.
LeapKeep decides clearance.

Dorra does not quietly override explicit rules, invent authority, or act beyond bounded permissions. It works inside a system built for explicit checks, audit trails, and reviewable outcomes. Clearance authority is never delegated to AI — it is owned by the team and recorded permanently.

Governed output

Every Dorra response carries evidence pointers, confidence levels, and policy version. Nothing is mystical.

Audit trail

Prompts, retrieved context, memory items used, tools called, outputs, and overrides — all versioned and tamper-evident.

No autonomous action

Dorra suggests or queues. It does not execute against production without explicit human approval at every gate.

Get started with Dorra

Give your software a memory,
a map, and a consequence engine.

Dorra is the first reality interpreter for software change — built on context graphs, decision memory, and bounded operational intelligence.

Open DORRA StudioAnalyze a project in LeapKeep