Technology 02
Operating model
Agentic AI
Not a single chatbot, but a team of eight with distinct roles.
Each agent owns one stage of the review and hands off to the next. An orchestrator manages sequencing and re-runs, and every output is traceable down to the source document.
01 — Reasoning chain
Eight specialists in a line, passing the review along.
The orchestrator manages sequencing and re-runs. Each stage's output becomes the next stage's input, and the final record returns as the first input of the next cycle.
OUTPUTS · ARTIFACTS
Signal map
Dimension matrix
Scenario options
Verification gate
Decision pack
Planning memory
ONE REASONING CHAIN · EIGHT SPECIALISTS
Anneal evidence feeds every stage · every output traces back to its source
1
Framer
Question framing
2
Scout
Signal detection
3
Matrix
Dimension scoring
4
Scenario
Scenario design
5
Leverage
IP leverage
6
Watcher
Change monitoring
7
Writer
Document drafting
8
Archive
Record keeping
02 — Orchestration
Thin evidence sends the work back a stage.
The orchestrator puts each stage's output through a verification gate. When evidence is insufficient or contradictory, only that stage is re-run, and anything needing human review is separated out and flagged as open.
Verification gate
✓
Every sentence bound to its source text
PASS
✓
Measured and claimed values tagged
PASS
!
Single-source evidence — cross-check requested
RETRY 02
◷
Awaiting human judgement — budget and contract terms
HOLD