Work
Work that is ready, not just scheduled
A date is a prediction. Readiness is a fact. This module is built around the difference. A task is not startable because the bar chart says so, but because the permit exists, the access is granted, the isolation is in place and the parts are on site.
A breakdown that reaches the real work
Most planning tools stop at a task and hand execution to something else, which is where the trail breaks.
A node is a task or a work order
Lightweight phases and milestones where that is all you need; a real work order where the work happens, carrying its own parts, hours, failure coding and signatures. The plan and the execution record are the same object, not two that have to be reconciled.
Dependencies with cycle detection
Finish-to-start relationships across the breakdown, with the graph refusing to accept a cycle rather than discovering it later as an unschedulable plan.
Earned value that closes
Periods that close, lock and take explicit corrections, with rules of credit, cost-to-date and cost-of-delay. A closed period stops moving, which is what makes the number worth reporting.
Scenarios you can compare
Branch a plan, change it, and put the two side by side before committing, rather than editing the live plan and losing what it used to say.
The rule that matters
Everything else in this module follows from one decision about constraints.
A constraint clears only from an authoritative record.
Permits, access and isolation never clear automatically. No amount of sensor data, inference or convenient assumption will mark them satisfied. A person or a system of record has to say so, and that act is logged.
Telemetry works in the other direction: it can raise a conflict, flag a spatial clash, or contradict a claim that something is ready. It is allowed to stop work. It is never allowed to start it.
Made-ready evaluation
Each node carries its constraints and reports whether it is genuinely startable, so a look-ahead shows what can actually be done next week.
Percent plan complete
Measures how much of what you committed to actually happened. The honest counterweight to a plan that always looks green.
Spatial conflict detection
Two crews scheduled into the same space at the same time is a constraint, surfaced before the second one travels.
Evidence, held rather than assembled
Sealed snapshots
Seal the state of a project, verify the seal later, and diff two snapshots to see exactly what changed between them. Legal holds prevent anything under them from moving.
Obligations and gates
Commitments tracked with versions and their own metrics, and gates that a project has to pass rather than a milestone somebody can drag.
Scope freeze with addenda
Freeze the scope, then change it only through an addendum that carries an approval. The record of what was agreed survives the change.
Competency register, fail-closed
Skills with a level, a validity window and evidence behind them. Grants are verified and revocable, and an expired competency stops being valid on its own rather than when someone notices.
Dates you can interrogate
A forecast is only worth having if you can see where it came from and when it refuses to answer.
Float, and what is actually driving the date
A forward and backward pass across the network gives total float and free float per node, and marks what is critical by the relationship that drives it, not by whichever bar happens to be longest.
Critical-path drag, priced
For anything on the critical path, how many days it is personally costing the finish date, and what those days are worth. It turns “the project is late” into a ranked list of what to shorten first.
A forecast built from your own history
P50 and P80 completion dates drawn from your closed work orders, thousands of real start and finish times, bucketed by category and asset type. Every schedule-risk tool asks a human for three-point estimates, which is the weakest part of the method. We already hold the distribution.
Schedule health that refuses to flatter you
A DCMA-14 subset run as an advisory report. A check the data model cannot fail reports “not applicable” rather than a green pass, a ratio over too few nodes reports “insufficient basis”, and the score is a plain count of passes over eligible checks, never a weighted composite. It never blocks a save, a gate or a date.
When the history is thin, the answer is “no reference class”, not a number.
A P80 computed from a range somebody invented in ten seconds is less truthful than a typed date, because it looks computed. If no bucket reaches a usable sample the forecast declines to produce one, and every figure it does produce carries the sample size it came from.
Worth saying plainly: the technique is not novel, InEight and Deltek Acumen Fuse have shipped versions of this for years. What is different is that the inputs are free, because they are your own completed work.
Where the assistant helps, and where it stops
The assistant will draft a breakdown from a scope, revise one you already have, review a draft for gaps, and enrich individual nodes. Every one of those arrives as a proposal that a person accepts or dismisses. Nothing it produces enters the plan on its own, and the decision is recorded either way.
What this is not
Dependencies are finish-to-start, with no lag and no other relationship types. There is no optimiser and no resourceleveller, and nothing here draws a Gantt chart. If any of those is the requirement, this does not do it, and we would rather you knew before a pilot than during one.
Projects is an optional module. It is off by default; turning it on is a commercial decision, and turning it off never interrupts work already in progress.
Bring a plan that went wrong
The useful conversation is about a job that slipped, and whether readiness would have caught it. That tells you more than a feature list.