A task already accepted keeps its actor: only the tasks nobody started yet are tried.
Resource substitution what-if
The deduction chose the actors and goods it knew. Since then, a firefighter may have arrived on site, a colleague may have become available. For each task nobody started yet, the strategy tries every such newcomer in place of an assigned resource and measures the gain. A large enough gain makes the solution Improvable: it still works, a new deduction would do better. It never raises a red alert.
implemented byResourceSubstitutionStrategy · gind-npm-strategy-supervision-resource-substitution
01 · Candidates
Who could replace whom
For each task of the solution which will run and that nobody started yet, for each resource assigned to it (Assigned_To) still on the field, a field instance is a candidate when it:
| Condition | Why |
|---|---|
| has the same concept as the assigned resource, or a sub-concept | A person replaces a person, a vehicle a vehicle. |
| is not used by a remaining task of the solution | It is free: the plan does not count on it already. |
has a role in common with the assigned person (Has → Role), when the person has roles | The deduction assigned the task through a role. |
| is Available (presence), when it has a presence | Busy, Offline, Do not disturb or Be right back cannot take a task. |
| is a newcomer: absent from the baseline, or Available now and not at the start | The deduction already chose among the resources it knew. |
02 · Gain
Nearer, or less strained
place = center of the problems the solution treats
distance = (d(assigned, place) − d(candidate, place)) / d(assigned, place) × 100
strain = (strain(assigned) − strain(candidate)) / strain(assigned) × 100
Improvable = max over the substitutions of best gain / minimal gain (only a gain ≥ minimal gain counts)
distance = (d(assigned, place) − d(candidate, place)) / d(assigned, place) × 100
strain = (strain(assigned) − strain(candidate)) / strain(assigned) × 100
Improvable = max over the substitutions of best gain / minimal gain (only a gain ≥ minimal gain counts)
For instance, John arrives 100 m from the gas leak while Emily, assigned to “Cut off the school gas supply”, is 5 km away: a 98 % gain, Improvable = 3.3, “John could replace Emily on "Cut off the school gas supply": 98 % nearer to the problems (100 m against 5.0 km)”. The radar button gets a blue badge and a light banner suggests to find solutions to compare.
03 · Parameters
In the supervision box
| Parameter | Default |
|---|---|
| Nearer to the problems | yes |
| Less strained | yes |
| Minimal gain | 30 % |
| Only newcomers | yes |
| Only available persons | yes |
04 · Watch points
Good to know
An assigned resource absent from the field is a failure (Weighted Graph Distance, Failing), not an improvement.
The IMT Albi Crisis timeline brings no newcomer: the behaviour is covered by tests with a person arriving near the problem, known from the start, of another role or busy.
gind-detection-engine-resource-substitution-strategy/…/ResourceSubstitutionStrategy.javaCandidates, gains, contributions.gind-webapp-r-iota/…/supervision/SupervisionEngine.javaRoles of the persons (Has) read for the context.