The deduced tasks do not carry planned dates: the plan is recomputed from their expected durations and the flows, from the start of the supervision.
Earned value (SPI, CPI)
Earned value management, a classic of project control, applied to the deduced process. It compares three amounts measured with an indicator of the tasks: the work the process should have done by now (planned value), the planned value of the work actually done (earned value), and what this work cost (actual cost). Their ratios tell whether the process is late (SPI) or over budget (CPI).
implemented byEarnedValueStrategy · gind-npm-strategy-supervision-earned-value
01 · Three amounts
Planned, earned, actual
| Amount | Computed as |
|---|---|
| Plan | Each task takes its expected duration, starting once the tasks before it end (critical path method along the nominal flows, from the start of the supervision). The tasks of the branches of an exclusive gateway the process did not take, and the contingent tasks never reached, are left out. |
| PV planned value | Sum of the expected value of each task × the share of its planned duration already elapsed. |
| EV earned value | Expected value of the tasks done + for a task going on, half of it (50/50 rule) or the share of its duration spent (at most 90 %). |
| AC actual cost | For the tasks over: their measured value, otherwise their expected value scaled by the time they really took (cost, strain and duration grow with the time). |
| BAC budget at completion | Sum of the expected values of the tasks which will run. |
02 · Indices
SPI and CPI
SPI = EV / PV below 1: late
CPI = EV(over) / AC below 1: over budget (on the tasks which are over)
Wrong = max( (1 − SPI) / (1 − SPI threshold), (1 − CPI) / (1 − CPI threshold) ) when an index is below 1
CPI = EV(over) / AC below 1: over budget (on the tasks which are over)
Wrong = max( (1 − SPI) / (1 − SPI threshold), (1 − CPI) / (1 − CPI threshold) ) when an index is below 1
Not judged too early — nothing is computed before the plan has done a share of the whole work (10 % by default): in the first minutes a few seconds of delay would give a meaningless SPI.
CPI on the tasks over — with the 50/50 rule a task going on earns half of its value while it costs its full time: counting it would show a CPI of 0.5 for a task on time. Its cost is judged once it is over.
03 · Parameters
In the supervision box
| Parameter | Default |
|---|---|
| Value measured with | cost (or duration, strain, distance) |
| Watch the schedule (SPI) · threshold | yes · 0.8 |
| Watch the cost (CPI) · threshold | yes · 0.8 |
| Judge from | 10 % of the work planned |
| A task going on counts | half of its value (50/50 rule) |
04 · Example
IMT Albi Crisis: the first task takes too long
“Confirm gas leak” is expected to take 3 minutes. Emily Hernandez accepts it and does not finish it. Five minutes after the start of the supervision, the plan expected it done and the next tasks started: PV = 16.6, while only half of the first task is earned, EV = 3.3. SPI = 0.20, far below 0.8: Wrong = (1 − 0.20) / 0.2 = 4.0, “SPI 0.20: 20 % of the cost planned by now is done”.
05 · Watch points
Good to know
It is robust but jumps at the end of a task; choose the elapsed share for long tasks that progress regularly.
gind-detection-engine-earned-value-strategy/…/EarnedValueStrategy.javaPlan, PV, EV, AC, SPI and CPI.gind-detection-engine-abstract/…/ProcessGraph.javaNominal flows, discarded branches, propagation along the critical path.