Every answer comes with proof.

Antananarivo.

It converts at whichever yields a lower price per share for the SAFE holder — i.e. the better of the cap or discount, not both. The conversion produces dilution to founders and prior investors in the priced round triggered by the conversion.

Two nuances often missed:

• Post-money SAFEs (YC v2) compute dilution before the new round, so existing shareholders bear the dilution; pre-money SAFEs spread it across new and existing.
• The conversion price is computed against the company valuation cap on a fully-diluted basis (including option pool) — which often surprises founders.

Likely overkill at 12 services. Defer it until you have a concrete need.

The panel reached this conclusion after three rounds. The cost of Istio (operational complexity, latency tax, training burden) is roughly constant at small scale, but its value compounds — mTLS, observability, traffic policies, canary routing — only when you have enough services that ad-hoc tooling stops scaling.

Signs you actually need it:

• You're hitting the limits of K8s NetworkPolicies for east-west security.
• Your observability gaps are between services, not within them.
• You need traffic shaping (canaries, A/B) that's expensive to implement per-service.

Lighter alternatives to consider first: Linkerd (simpler, lower overhead), application-layer mTLS via cert-manager, or just OpenTelemetry + a service catalog.