Petri net and influence diagram, running.

Fire the net's transitions by hand or let the stochastic clock run. Then solve the influence diagram and apply the optimal policy to the net: the decision becomes routing, and routing changes the queue. Same model as the Python package stratago_review; illustrative parameters.

1 · Stochastic Petri net (GSPN)

Token game of the review flow

Light bars are immediate transitions (zero time); dark bars are timed (exponential time). A transition can only fire when every input place holds tokens.

Opened

Human queue

In review

Decision

Production

Matrix bypass

Automated track

R · revisores livres

λ · PRs per day

Hours per pass

c · reviewers

r · rework

p · routed by the matrix

Clock

In production

Analytic ρ

Analytic W / PR

Simulated L

Simulated W / PR

P-invariant R + In review = c:

Latest firings

2 · Influence diagram, solved

The optimal policy, computed

S (good or defective PR) is hidden. Risk R and the LPA₂v state of evidence E are observed before deciding track D1. For each observation, the solver picks the action with the highest expected value. The decision matrix appears as a result, not as an assumption.

D2 · batch size

P(defective), small batch

Incident impact, high risk

Cost of a reviewer hour

Benefit of delivering

Large batch: P(defective) +0.10 and 1.6× hours. Unit of value: equivalent work hour.

E = V true

E = F false

E = P paracomplete

E = I inconsistent

Low risk

High risk

Expected value / PR

VOI of evidence

EVPI

Human track

Sets p = and hours per pass = , then resets the net.

D1 action	Human hours	Defect escape	Matrix quadrant
Automated	0,05	85%	Routine / Exploratory
Rule (contracts, property tests)	0,35	30%	Consequential
Deep	1,50	8%	Strategic
Return (split, more evidence)	0,10	—	Out of the queue

P(E | S): good → V 70%, F 5%, P 15%, I 10%; defective → V 10%, F 55%, P 15%, I 20%. P(high risk) = 30%; impact with low risk = 2. All parameters are illustrative and should be calibrated with the team's telemetry.