How Do You Handle a Software Project That Goes Off Track?

CEM handles off-track projects through a three-level graduated escalation chain -- Stop, Pause, Reset (light fix), Stop and Recap (medium fix), and Stop. Run It Back (nuclear reset) -- deployed in order of severity so the lightest effective intervention is used first. During validation, this chain recovered a project from an 8-commit crash day to a 68-commit recovery day (8.5x multiplier), and recovered a seasonal product through two major breakages to ship on a hard December 24th deadline with 0% final-phase issues.

The Standish Group CHAOS Report consistently finds that approximately 66% of software projects experience partial failure (challenged) and 19% fail outright. The primary recovery response in traditional development is to add resources, extend timelines, or cut scope. CEM replaces all three with systematic context management: the root cause of off-track projects in AI-native execution is typically poisoned context, not insufficient resources.

Stop, Pause, Reset is the first-line interrupt. Three words, three actions, deployable under cognitive load. Stop all execution immediately. Pause to let accumulated momentum dissipate and engage deliberate thinking. Reset by approaching the same problem from a new angle. Cost: seconds. Foundation preserves all state, so there is zero penalty for deploying it. The bias should always be toward deployment, not restraint. In the validation portfolio, the 12.1% product bug rate (against 20-50% industry norms) suggests liberal deployment of this lightest interrupt prevented compounding divergence.

Stop and Recap deploys when the light fix proves insufficient. The operator forces the AI to externalize its understanding through three questions: What have we done so far? What is the current task? What is the next immediate step? The gap between the AI's answers and the operator's intent is the diagnosis. If the gap is bridgeable, bridge it and resume. If the gap reveals fundamental corruption, escalate.

Stop. Run It Back is the nuclear option for poisoned context. Close everything. Delete the thread. Walk away if needed. Return fresh. Foundation catches what matters -- templates, code, decisions. Only the corrupted conversation context is destroyed. The January 27, 2026 event validated this directly: an AI tool deleted a landing page instead of modifying it. The operator escalated through the full chain. Next day: 68 commits, 3rd highest in the entire 116-day dataset. Foundation made deletion safe. Deletion made the fresh start possible.

PRJ-06 validated the chain under deadline pressure. Two major breakages (checkout integration failure, regional deployment failure) during a 37-day build with a hard December 24th deadline. Each breakage followed the pattern: stop, contain, fix root cause (not symptoms), resume. Between breakages, a 5-day peak sprint produced 113 units of work at controlled 15% issue rate. Final phase shipped clean.

The core principle: the cost of an unnecessary interrupt is seconds. The cost of a missed interrupt is compounding divergence. In AI-native execution, each misaligned instruction becomes context for the next, so continuing through problems produces exponentially more expensive rework than stopping early.

Related: FAQ #48 (Preventing Burnout), FAQ #52 (Environmental Control)

References

1. Standish Group (2020). "CHAOS Report." Project success and failure benchmarks (66% challenged, 19% failed).
2. Keating, M.G. (2026). "Case Study: The Recovery Build." Stealth Labz. Read case study