v2 analysis and visualization coverage

Wiki home

Documentation index · Source repository

This inventory covers 40 parser fields/families and 34 Go, Python and shell sources from the supplied v2 revision 74b71090410cac1315ae08f479a95c086feeaaf6. The manifest records hashes. Run python3 scripts/audit-v2-analysis.py --v2 ../v2 to detect changes or omissions; this is not a numerical equivalence test.

The following visualization families are available through Saved results → Images and its comparison tools. Main Metrics keep their session/delivery-window definitions; additional analysis uses research.definition = v2-corrected-observations-v1. Different source observations and statistical definitions can produce different values from v2. Missing measurements remain N/A, and unresolved propagation origins remain unknown.

Parser outputs

v2 family	v3 representation and conditions
node_count, average_degree, average_degree_excluding_leaves	Protocol/group timelines and summaries. The non-leaf denominator statistic is ALL degrees divided by the number of nodes with degree > 1
diameter, shortest_path_length	Computed from saved edges. Excludes self/unreachable pairs and reports connected-pair coverage
clustering_coefficient, assortativity, modularity	Mean local clustering, degree correlation and seeded Louvain modularity; undefined statistics remain N/A
Five centralities: betweenness, PageRank, degree, closeness, eigenvector	Normalized global/group means for saved graph observations
degree_distribution-*	Mean per-snapshot degree probability, PDF/CDF and weighted Student-t fit
frt, reachability, drc	Per-message first-reception latency, reach and duplicate counts; summaries over message means. DRC/graph-node and DRC/target are separate
eager_count, lazy_count, eager_frt, eager_reachability	Estimates from GRAFT/IHAVE/IWANT and message IDs. Unclassified counts remain visible; eager FRT requires classified receipts with timing evidence
pub_map, propa_map, reach_map, eager_propa_map, eager_reach_map	research.messages: publication, receipt, parent, hop, population and evidence; propagation/estimated-eager curves; JSON/CSV export
dup_map, time	Per-message duplicate cumulative curves and time axes; raw duplicate events remain in events.jsonl
send_ihave, recv_ihave, send_iwant, recv_iwant	Separates RPC occurrences containing a control type, entries and referenced IDs
Logical IHAVE/IWANT send/receive fields	Advertised/requested ID counts, distinct from unique IDs or delivered bytes
send_graft, send_prune, logical_send_graft, logical_send_prune	Mesh transitions and reciprocal logical-edge creation/removal. Wire RPC counts have separate _rpc names
v3 additions	IDONTWANT, control drops, unknown paths, lifecycle/score summaries, measured total/per-protocol libp2p stream bandwidth

Python and batch visualization

v2 source	v3 tool
basic_graph/*	Series/Case repeat means and sample-SD bars, arbitrary metrics and D conditions, degree probabilities
compare_graph/*	Matched-case difference, target/(target+baseline), target/baseline and error propagation
trade-off_graph/*	FRT–DRC scatter, reach color, reference arrows, two/three-objective Pareto, weighted score, i/f groups and p filter
time_vs_metric_plot.py	Propagation/duplicate cumulative curves, linear/log time and combined panels
calc_lazy_metric.py	Eight observational contribution/overlap/efficiency estimates against a matched lazy-off baseline
dup_regression.py	All four duplicate regression designs, coefficients/rank/in-sample R², observed/fitted comparison
old/graphic.py	Metric JSONL and x/y imports; arbitrary metric, axis and color scatter
old/propagation.py, old/prop_plot.py	Per-message first-reception paths, latency/hop distributions, mean cumulative receiver counts and increments over time/hops
old/prop_plot_graph.py	CSV/JSON curve overlays at raw or peak-normalized scale
old/mean_degree_ratio.py	Total/Count imports and mean snapshot degree probabilities
old/clustering_vs_eigenvector.py	Two-metric scatter and reference-case arrows
old/frt_drc_graph.py	Three fixed historical/ER reference charts from v2 constants, accessed separately from selected experiments
old/fitting.py	Weighted Student-t MLE; df/location/scale trends; univariate/multivariate quadratic models and predictions for supplied D values
image_gen.sh, total_image_gen.sh, run*.sh	Submit selected server analyses and export all PNG/CSV/chart definitions as ZIP. Does not execute the old Python CLI/path layout

Imports support up to 32MiB/file: metric JSONL, propagation tree JSON/JSONL, duplicate-time maps, x_case/y/yerr or x/y JSON, numeric CSV, and v3 analysis JSON. Arbitrary v2 internal data shapes and original directory/file naming are not automatically reproduced. See visualization for operation.

Definitions and source evidence

Exact formulas, aggregation, graph sampling and Eager/Lazy inference belong in saved-result research metrics. Retain these distinctions when comparing v2:

• Main session-window Metrics and the research definition for v2 chart families are separate. Research reachability conditions its population on receipt times; per-message statistics and cumulative curves also use different weighting.
• v2 reciprocal GRAFT graphs and v3 sampled neighbor graphs are different observations. Saved v3 graphs merge topics into undirected edges; analysis uses at most 1,440 observation samples. Historical summaries without edges cannot recover new centrality/path metrics.
• Distinguish estimated eager/lazy, unclassified origins, model predictions and v2's fixed historical/ER references. Even new message-ID logs do not directly establish sender-queue causes.
• Research reducers correct the membership union and even median. Student-t fits observed probabilities directly instead of reconstructing 60,000 artificial samples or substituting arbitrary fallback parameters. Exact numerical, file-layout and plot identity with the old Python output is not asserted.

New detailed IDs and graph edges exist only in logs collected after deploying that version. See API for fields/omission limits, retention for source/analysis/image boundaries, and bandwidth for the additional measured stream-usage definition.

Validation is in research metrics, origin inference, Peer tracing, research UI, and result image tests. See development for the source-inventory check.

---

Korean translation: 한국어.