Propagate
Know what Propagate does, what it skips, and why.
The tree recalculates itself constantly — this button is for forcing the question of what got left behind.
Propagate pushes driver values downstream through the tree's causal arrows and recomputes outcomes, running in dependency order — driver before outcome. It finishes with a toast summarising the result, in the shape "Propagated 12 nodes — Skipped 7 (2 missing baseline, 4 missing source values, 1 in cycle)".
Three skip reasons, and two are fixable. Missing baseline — the target node has no Actual or Plan to build on; add data to that node and run again. Missing source values — an upstream node it depends on is itself empty. In cycle — the edges form a loop, and nodes inside a cycle are skipped rather than left to recalculate forever; break the loop by removing one of the edges.
Propagation isn't only a button — it also runs silently after any inline edit or side-panel change, so the tree stays internally consistent without you asking. What the manual button is for is forcing a full pass and seeing the summary, particularly after a change that could ripple a long way.
Two behaviours worth knowing before you rely on it. Time lags are respected — a two-month-lag edge reads its driver's value from two months earlier, not the current month. This works because edges themselves are editable, not just nodes: opening Edit edge on a connection exposes a Probabilistic toggle, a Weight %, and a Time delay in months, with a mode badge on the canvas (wire, ratio, or delta) showing how that edge currently combines with its target. This is what Apply actually writes, and it's the real mechanism behind the lag behaviour above. And manual KPIs are never overwritten; only formula and expression-mode nodes get recomputed, so a figure someone typed in by hand stays exactly as they left it — this is the same "Computed from upstream" flag covered in 7.4.
A ≠ formula badge can appear on a node whose stored value has drifted from what its formula would now compute — the sign of a stale manual override sitting on a node that should be live. The exact drift threshold that triggers the badge isn't confirmed; treat the badge itself as the reliable signal rather than any specific percentage.
- 1Tree tabs, each with its node count, and the tenant's total tree count at the right
- 2Views, and the month navigation with its edit summary
- 3Data layers — Actual, Plan, Forecast, VAR
- 4View modes — Default, Focus, Impact — here set to Impact
- 5Card colouring — Variance and Direction are one toggle, only one active at a time; Confidence and Sparklines are separate overlays
- 6Add Node, Add Formula, Generate via AI
- 7Propagate, Auto-layout, Inspect, Search, the orientation toggle, Calibrate, Delete Tree
ValueWorks 2026.08
Takeaways
Propagate recalculates in dependency order and reports a toast: nodes propagated, and skipped with a reason.
Skip reasons: missing baseline, missing source values, in cycle — the first two are fixed by adding data, the third by breaking the loop.
Edges are editable too (Probabilistic, Weight %, Time delay) — the actual mechanism behind lag. It also runs silently after every edit; manual KPIs are never overwritten, and a ≠ formula badge flags a stale one.
Surfaces covered
- Dependency order
- Skip reasons: missing baseline
- Missing source values
- In cycle
- ≠ formula-drift badge
Also taught in the Academy
Module 7 · Digital Twin: the Driver Tree
Take it as lesson 7.9 in Business Analyst & FP&A