Leaderboards

Ranked over every author Datatracker has ever seen.¹ Click a row to pin that person as the source node in the BC Explorer.

¹ Rankings are final, binding, and unappealable, until someone commits a new Internet-Draft, which happens on the order of hourly.

Search Rows

Authors ranked by the selected metric. Click a column header to sort; click again to reverse. Click a row to load that author as the Buscemi Centrality source.
#	Name	RFC 1st	RFC co	ID 1st	ID co	Total	Weighted	h
Loading IETF graph...

BC Explorer

Pick any author as the source node s. The tool then computes Buscemi centrality of every other author relative to that s: BC(v; s) = λ · A(v, s) + (1 − λ) · Σ α(u; s) · A(v, u), with all ingredients live-tunable in the panel on the left. All parameters are dimensionless. Any semantic interpretations attached to them by the reader are the reader's own.

Source node

Parameters

λ (direct vs. neighbourhood mix) 0.50 r (neighbourhood radius, weighted-cost) 2.0 D (path depth cap for A) 3 K (Pareto frontier width) 6

Edge types

Cost and quality factors per edge type.
Type	cost c(t)	quality q(t)
coauthored_rfc	Cost for coauthored RFC	Quality for coauthored RFC
coauthored_id	Cost for coauthored I-D	Quality for coauthored I-D
repeated_coauthorship	Cost for repeated co-authorship	Quality for repeated co-authorship

Repeated-coauthorship threshold (docs)

Top 50 near BC(v; s)

Top 50 authors by Buscemi centrality relative to the current source. Click a column header to sort; click a row to re-root the explorer to that author.
#	Name	A(v, s)	BC(v; s)

Look up any author's BC relative to the current source

About

Portrait of Steve Buscemi, for whom the measure is named. — Steve Buscemi. The canonical source node.

Companion to Buscemi Centrality: Source-Relative Centrality in Heterogeneous Affiliation Graphs (Underpoot & Flugelhorn, SIGBOVIK 2026).

Steve Buscemi is not, as far as anyone knows, an IETF author. Pick any author in the Datatracker graph as the source s (for example ekr, jari, or martinthomson) and read the centrality of every other author relative to them. Call it ekr-centrality, jari-centrality, or whichever analogue you fancy. The paper reminds us that all of graph theory is, at some level, secretly about Steve Buscemi.

Formula (from the paper)

Path quality. Score a single route from one author to another: multiply the quality q of every edge you cross (each greater than 0 and at most 1, so weaker edges drag the product down), then divide by one plus the total traversal cost. Long, weak paths lose to short, strong ones.

$$Q(P) = \frac{\prod q(\tau_i)}{1 + \sum c(\tau_i)}$$

Q(P) = (∏ q(τᵢ)) / (1 + Σ c(τᵢ))

Accessibility. How easily can u reach v? Take the single best path between them; whichever route maximises Q wins. A node is trivially accessible to itself (the empty path has unit quality and zero cost), so A(v, v) = 1.

$$A(u, v) = \max_{P \,:\, u \leadsto v} Q(P), \qquad A(v, v) = 1$$

A(u, v) = max over paths P : u ⇝ v of Q(P), A(v, v) = 1

Source neighbourhood. The local crowd around the source: every author whose weighted distance to s is at most r. Think of r as a radius drawn around s in the graph; the measure only cares about what happens inside that radius.

$$N_r(s) = \{\, u : d_w(u, s) \le r \,\}$$

N_r(s) = { u : d_w(u, s) ≤ r }

Neighbourhood weights. Each neighbour u gets a weight α proportional to how accessible it is from s. The weights sum to 1 across the neighbourhood, so α(u; s) is u's share of the source's local world: a neighbour barely inside N_r(s) gets a sliver, a well-connected one gets a big slice.

$$\alpha(u; s) = \frac{A(u, s)}{\sum_{x \in N_r(s)} A(x, s)}$$

α(u; s) = A(u, s) / Σ A(x, s) over x in N_r(s)

Buscemi centrality. A blend of two views of how central v is to s. The first term is how easily v reaches s directly. The second is how easily v reaches each of s's neighbours, averaged by how important each of those neighbours is to s. λ is the dial: at λ = 1 you get pure direct accessibility, at λ = 0 pure neighbourhood embeddedness; the default λ = 0.5 mixes them evenly. A v can score highly by being close to s directly, or by being close to lots of s's collaborators.

$$BC(v; s) = \lambda \cdot A(v, s) + (1 - \lambda) \sum_{u \in N_r(s)} \alpha(u; s) \cdot A(v, u)$$

BC(v; s) = λ · A(v, s) + (1 − λ) · Σ α(u; s) · A(v, u) over u in N_r(s)

How the graph is built

Nodes are Datatracker person records that appear in at least one documentauthor row. Between any two authors u and v there is one edge, whose type is the strongest of:

repeated_coauthorship: they share ≥ N documents (default N = 3).
coauthored_rfc: they share at least one RFC.
coauthored_id: they share at least one Internet-Draft.

Leaderboard definitions

Datatracker's documentauthor table includes order, a 0-indexed position. “First-authored” means order == 0; “co-authored” means order > 0. Editors (marked “Ed.” on the front page) are counted as authors per RFC 7322 convention. Combined metrics:

Total docs = first-authored + co-authored, across RFCs and I-Ds.
Weighted = 1.0 × first-authored + 0.5 × co-authored.
h-index = largest h such that the author has h documents each with at least h co-authors.

Caveats

Approximate path search. A(u, v) is computed via depth-limited modified Dijkstra with Pareto pruning. The exact problem (max-Q simple path) is NP-hard in general. Defaults: depth D = 3, Pareto width K = 6. Increase for more accuracy at the cost of speed.
For the BC sum over N_r(s), we cap the neighbourhood at its top 50 contributors by α to keep the computation responsive.
Datatracker person records are de-duplicated by ID, not by name. Some authors may appear twice under different person IDs if Datatracker has not merged them. This is not considered a deficiency of the measure.
All numerical results produced by this tool are strictly indicative. No load-bearing IETF decision should depend on them. The authors reserve the right to disclaim any such decision after the fact.

Future work

It has occurred to the authors that the same construction applies, unchanged, to any affiliation graph with typed edges, and that running it over arXiv or Google Scholar would likely be of broader scientific utility than running it over the IETF Datatracker. The authors are not planning to do that.

Honorable mentions

Honorable mention to The Oracle of Bacon, the original and still the canonical instrument for computing distances to a single distinguished actor. This tool is not affiliated with The Oracle of Bacon, the University of Virginia, or Kevin Bacon.