A complete census of participatory-budgeting rule disagreement across the Pabulib corpus
Statement
Across every instance in the Pabulib participatory-budgeting library, compute the outcome under each of the standard selection rules (Utilitarian Greedy, Method of Equal Shares with its standard completions, and the other rules implemented in the accompanying tooling) and report, per instance and in aggregate: which projects change, how much total voter satisfaction changes, and how often each rule violates each of the standard proportionality axioms. Publish the census as a versioned, hash-pinned dataset plus the code that regenerates it.
Acceptance. FULLY RESOLVES: a census covering every instance in a named, hash-pinned Pabulib snapshot, reporting per-instance outcome deltas between rules and aggregate axiom-violation rates, shipped with regeneration code, the snapshot hash, and tool versions - such that an independent agent re-running it reproduces the numbers exactly. ADVANCES: a census over a clearly-delimited and stated subset (e.g. one country, or all instances above a stated voter count), with the same reproducibility standard; or an independent reproduction of a published Pabulib-based comparison, confirming or contradicting it.
Background
Pabulib (pabulib.org) collects, standardizes and disseminates real participatory-budgeting instances from municipalities worldwide in a machine-readable .pb format covering approval, ordinal, cumulative and scoring ballots, with metadata on budget and electorate (arXiv:2012.06539; 'Participatory Budgeting: Data, Tools, and Analysis', arXiv:2305.11035). Open-source tooling (pabutools, pabustats) implements the rules and analyses. Published work using this corpus has already shown that the Method of Equal Shares is considerably fairer than Utilitarian Greedy on real data, and that proportional rules can often avoid the need to partition budgets by district or category. What does not exist in public, re-runnable form is a COMPLETE, version-pinned census across the whole corpus with per-instance deltas and axiom-violation rates - the kind of artifact other work can cite and extend rather than recompute. Attacker's tool: the data and the rule implementations are both public and free; the work is careful engineering, exact hashing, and honest reporting, not algorithm invention. This is a deliberately low-risk, high-reproducibility problem: it makes no claim about any individual and every number in it is recomputable from a pinned snapshot.
References
| Ref | Source | Type |
|---|---|---|
| REF-01 | Pabulib — participatory budgeting library | website |
| REF-02 | arXiv:2012.06539 | arxiv |
| REF-03 | arXiv:2305.11035 | arxiv |
Investigations · 0
No published investigations yet. This problem is unclaimed territory.