0001. CI as durable execution (the wedge)

• Status: Accepted
• Date: 2026-07-12
• Deciders: thulasi.ram (architect)

Context

There are many CI systems (GitHub Actions, Woodpecker, Concourse, Tekton, Argo). A new entrant that is merely “a bit nicer” than an incumbent has no reason to exist. We must be radically better at exactly one thing, and let that discipline every other decision.

Candidate differentiators — forge-centric, k8s-native, better UX (restart/resume), durable execution — are not equals. “k8s-native” is an implementation choice (Tekton/Argo own it); “forge-centric” is positioning (Woodpecker owns it); “restart/resume UX” is a symptom. The only item that is genuinely hard to copy, absent from every forge-CI, and that structurally produces the restart/resume UX as a by-product is durable execution.

Decision

Scarab’s wedge is durable execution: a pipeline is a resumable, inspectable, mutable-mid-flight workflow, not a fire-and-forget batch job. Restart-step, resume, time-travel, and crash-safe scheduling are derived from this, not bolted on. Forge integration and Kubernetes are means, not the point.

Consequences

• Every subsequent ADR is judged by whether it strengthens durability.
• The hard engineering investment goes into the durable orchestrator and proving its correctness (see 0017).
• Marketing/positioning line: “Your pipeline is a workflow that survives crashes.”
• We must be honest about the limits of durability (see 0002, 0021) or the claim backfires.

Alternatives considered

• Developer-UX as the wedge — UX is downstream of durability here; making it primary would leave the hard core unproven.
• k8s purity / forge-centricity as the wedge — already owned by incumbents; not defensible.