121 official pins are captured exactly; 17 apply inside this projection.
24 / Recorded history -> future admissibility
The resolver can branch.
The release remembers one route.
Airflow publishes a tested, versioned constraint set. This lab compares it with every solution in one frozen dependency projection, while keeping resolver search effort separate from solution-path cost.
01 / Finite universe
A complete census inside an explicit boundary
The counts below describe the frozen five-requirement Airflow Core projection. They do not describe every Airflow dependency or every PyPI release.
02 / Historical Load
One filter, three declared costs
Choose a cost. Both optima are computed over the same 64-solution universe; only the official constraint predicate changes admissibility.
03 / Optimum witnesses
Where the constrained environment changes
Only projects with more than one candidate are shown. Fixed transitive dependencies remain part of every complete 17-project solution.
| Project | Baseline / free witness | Official constraint | Wheel-byte delta |
|---|
04 / Counterfactual filters
Relax a pin without rewriting upstream evidence
Every ablation is an Onto2D analysis-only filter. The official constraint bytes and their source identity stay unchanged.
05 / Shared dependencies
Reuse is structural, not duplicated
One target project can constrain several independent branches at once.
06 / Resolver diagnostic
Rejected work is not path cost
Enumeration reveals conflicts and branching. Its work count does not populate dH.
07 / Scientific boundary
A real upstream filter, a deliberately bounded claim
Complete exhaustive enumeration under the declared project and candidate inventory.
Local analysis views that never become Airflow recommendations.
No runtime or backtracking count is interpreted as Historical Load.