estimate_scan_hours() was pure masscan math — zero nmap time modeled.
For masscan+nmap and geo-scout modes with 10K masscan hits, actual nmap
time dominates but was completely invisible in the estimate.
Added nmap phase calculation using empirical constants:
- NMAP_TIME_PER_HOST_SEC = 10 (nmap -sV -T4 per host)
- MASSCAN_HIT_RATE = 0.01 (fraction of IPs with open ports)
- _NMAP_WORKERS = 10 (matches WEBRUNNER_NMAP_WORKERS default)
Also fixed build_estimate_table() partial-chunk overestimation: last
chunk was billed at full preset size regardless of actual IP count,
causing 1.5-2x cost inflation for non-aligned scan ranges.
n0mad1k
dec6c33897