Schedule risk management for federally funded R&D

Federal R&D programs are scheduled against milestones with money and follow-on funding attached: a DARPA program's phase exit, an ARPA-E performance target, an NIH aim's go/no-go, an FFRDC's annual program review. The recurring question from the sponsor is not "when is the deterministic date" but "how confident are you, and what is your basis." Answering that requires a probabilistic schedule, an explicit contingency basis, and traceability from the estimate to the committed date.

That is what separates a Level 4 risk-driven schedule from a padded Gantt. The schedule has to show the risk drivers, the simulated finish distribution, the contingency derived from it, and the decision points where the program can pivot or stop — all reproducible and reviewable.

AACE International's Recommended Practice 132R-23 defines maturity levels for risk-adjusted scheduling, with Level 4 being full risk-driven analysis: a quantitative schedule risk analysis where identified risks drive the simulation and the resulting contingency is built into the committed plan rather than appended as a flat buffer. It is the standard sponsors and review boards increasingly expect on programs where the schedule carries real consequence.

CritPath AI implements Level 4 directly. Risk events are injected into a Monte Carlo run against the live dependency network, the simulation produces criticality and sensitivity output, and the contingency is sized from the distribution — giving PIs and program managers a basis they can put in front of a program office and defend line by line.

Flat percentage contingency — "add 20% to be safe" — is exactly the kind of number a sponsor will push on, because it has no basis. CritPath AI replaces it with a Monte Carlo engine that runs PERT-Beta distributions over task durations, injects discrete risk events from the register, and reports a full finish-date distribution with P50, P80, and P90 dates, a criticality index showing which paths drive the outcome, and a tornado chart ranking the biggest risk drivers. Optional duration correlation captures the reality that related tasks tend to slip together.

The output is a contingency figure with a documented derivation: this much schedule reserve protects this milestone to this confidence level, because these specific risks drive it. That is a number a program review can interrogate and accept, rather than a guess that invites a fight.

Federal programs are gated by design — phase exits, aim transitions, annual reviews — and those gates carry Go, No-Go, Pivot, or Defer outcomes that change everything downstream. In most tools a gate is just a milestone diamond; the replan after a decision is manual. CritPath AI models gates as first-class schedule objects, and a gate decision triggers retroactive rescheduling: the engine re-cascades dates, buffers, and the critical path automatically when the program pivots or a milestone slips.

For a multi-PI grant or a DARPA seedling deciding whether to advance, that means the consequence of a decision is visible immediately and on the record, not reconstructed by hand a week later. The schedule stays honest through the program's actual decision logic.

Every schedule-affecting change in CritPath AI is captured in an append-only audit log, so a reviewer or program officer can reconstruct how a committed date was reached and what changed it. Combined with Level 4 risk-driven scheduling, that gives federal programs the documented, reproducible basis their oversight expects — without the cost and friction of a legacy desktop risk tool that one analyst runs and exports as a static report.

Pricing is deliberately grant-fundable: $10 per user per month for every standard feature and unlimited projects, with AI copilot usage billed separately by metered token usage so a program only pays for the AI it actually uses. That fits comfortably inside a grant's tooling line, and the whole program team — PIs, co-investigators, program managers — can work in the same browser-based plan.

It is worth being precise about compliance posture, because federal buyers ask. Shipped today: AACE 132R-23 Level 4 risk-driven scheduling, Monte Carlo contingency, decision gates with retroactive rescheduling, and an append-only audit log. During the public beta, AI assistance — the schedule-aware copilot, AI work-breakdown decompose, and reporting — is available; fully autonomous AI-agent runs are turned off.

On the roadmap and not yet shipped: 21 CFR Part 11 electronic-signature attestation, which some federally regulated programs require, is an Enterprise roadmap item rather than a current capability. SOC 2 Type II, SAML SSO, HIPAA BAAs, and on-prem deployment are Enterprise (custom) offerings. We flag these plainly so a program office can plan around what exists now versus what is coming.