Problem
Strata Cloud Manager's Policy Optimizer and Config Cleanup need firewall traffic logs in Strata Logging Service (SLS). Before moving logs there, engineers need a defensible number for log volume and the egress path. The public SLS Estimator takes one log rate and one retention period, so it can't model several Panoramas or per-firewall egress.
What I built
A 17-tab Excel runbook generated from Python. Each number and command carries either a source ID from 91 numbered official sources or an assumption label, and 14 open questions are logged. One researched fact shaped the model: firewalls send logs to SLS directly, and neither Panorama nor a Log Collector relays them, so egress is per firewall.
The working workbook is not published. A sanitized sample rebuild (HTML views of the key tabs plus the sample workbook, on a fictional fleet with sample values only) opens from the Open app button at the top of this page.
How AI was used
A multi-agent research pass over public vendor documentation and knowledge-base articles, then an adversarial review across 5 lenses: 58 verified findings (math 9, sources 13, usability 8, credibility 19, stress 9), including two high-severity formula bugs, all applied before v1.0. Verifier agents checked 24 network and CLI command facts: 5 held, 13 were corrected and 6 could not be confirmed.
Results
0 Excel error cells after full recalculation, in both the blank and the sample workbook. The public estimator and the sizing documentation each state a bytes-per-log assumption; the runbook documents and brackets both, each tied to its source, so a sizing can be read against either.
Status and limits
v1.0 is built and the project is paused; it has not been used in the field. Key inputs such as the peak factor stay assumptions until a pilot measures them.
What I learned
The adversarial review mattered more than the research. The credibility lens alone raised 19 of the 58 findings.