PQC-gossip QS-DAG Explorer

Live endpoint-backed testing

Headline metrics in this section are reserved for real configured endpoints, real returned ACK/receipt data, and wall-clock measurements. Modeled values are isolated below in a separate reference section.

Bridge endpoint · PQCNET_PRIMARY_ENDPOINT

LB bridge endpoint · PQCNET_LB_ENDPOINT

Custom hardware bridge endpoint · PQCNET_CUSTOM_ENDPOINT

Custom32 endpoint · PQCNET_CUSTOM32_ENDPOINT

Droplet roster

Expected distinct nodes · PQCNET_EXPECTED_NODES

Probe attempts

Batch size · PQCNET_BATCH_SIZE

Batches per run · PQCNET_BATCHES_PER_RUN

Concurrency · PQCNET_BENCH_CONCURRENCY

Timeout ms · PQCNET_TIMEOUT_MS

Dispatch mode · PQCNET_DISPATCH_MODE

Target total tx · PQCNET_TARGET_TOTAL_TX

Transport · PQCNET_TRANSPORT

Custom node label · PQCNET_CUSTOM_NODE_LABEL

Reviewer-safe mode

Hide modeled/extrapolated logical capacity values and show only measured live evidence.

target roster pending validation

not run yet · live endpoint-backed tests idle

Live proof of test basis

Audit trail showing which configured endpoints were actually contacted and what evidence was returned.

live evidence only

truth guards pending

Run history / quick trends

Expanded charts make QS-DAG endpoint throughput, latency, and convergence trends easier to inspect.

charts expanded

not run yet - demo placeholder only; sparklines will use the last 20 live runs after probes complete.

Latest run details

Endpoint inventory

Live DAG receipt explorer

Populated only from live receipts. If DAG-native fields are absent, this panel records that failure explicitly.

Live DAG receipt rows

Fault-injection %

Formalizes the existing PrivacyNet chaos sweep into a paired clean-vs-faulted artifact. Sources: chaos-module.md, large-scale-module.md, and EmailsToKenEdward.md when supplied with the run package.

artifact contract

fault-injection evidence not verified

Requirement	Panel contract
Outcome percentages	Report ok/dropped/timeouts/rerouted as x/requests with Wilson confidence intervals; preserve invalid, resp_cap, internal, and other error buckets.
Baseline pairing	Pair every faulted run to a clean run with the same seed and emit throughput_drop_pct = (clean - faulted) / clean.
New knobs	Add -byzantine-fraction f and assert malformed_committed == 0; add -resp-cap-bytes / -resp-cap-fraction feeding the existing resp_cap bucket.
One-factor sweep	N >= 30 seeds; latency +10ms on 30% links, drop-prob {0,.02,.05,.10}, timeout, slow-node, byzantine {0,.15}, resp-cap {off,1024B/40%}.
Outputs	JSON rows include outcome percentages, Wilson CIs, throughput_drop_pct, byzantine_rejected, malformed_committed, and p95; Prometheus exports the same counters/gauges.
Negative test	Disable adaptive_reroute and require success percentage to drop versus the paired reroute-enabled baseline.

verified fault-injection report pending

Gossip-under-churn delivery

Defines the missing gossipsub/Kademlia delivery harness with per-message round and hop tags, churn, link-loss injection, and delivery-ratio CDFs.

new harness

gossip-under-churn evidence not verified

Component	Implementation contract
Node instrumentation	Tag every message with msg_id, first_seen_round, and hop_count; emit PQCNET_GOSSIP_DELIVERY_PROBE_V1 extending receipt-only PQCNET_GOSSIP_RECEIPT_PROBE_V1.
Harness	Build pqcnet-gossip-churn with -nodes, -fanout, -rounds, churn fraction/interval/seed, -link-loss p, physical mesh netem/iptables support, and -jsonl.
Metrics	Measure delivery_ratio@round_k CDF, rounds_to_99pct, dissemination_ms vs log N with R^2, and mean hop count.
Sweep	Run churn {0,.1,.2,.3} x link-loss {0,.05,.10}, N >= 20, with Wilson CI on delivery ratio.
Charts	Add delivery-vs-round and dissemination-vs-log N charts; pass/fail tile gates on >=99% delivery by round 3.
Negative test	Partition the mesh and require the delivery ratio to fall below target, proving the panel catches broken dissemination.

verified gossip-under-churn report pending

Modeled reference only

These controls and outputs are not live, not endpoint-measured, and never feed the main live scorecards.

MODELED · NOT LIVE

Knobs / env-style signal

Editable runtime settings used by the run buttons.

Payload Speed Mode

Measure real payload byte throughput over the live PQC-gossip WebTransport path. This is separate from optional modeled batch capacity and can optionally verify payload receipt hashes.

real payload bytes

Payload Speed Mode measures real payload byte throughput. Use it with ACK throughput and receipt checks to distinguish benchmark batch throughput, real data-plane throughput, and integrity/receipt evidence.

Payload size bytes

Custom payload size bytes

Payload count

Concurrency

Dispatch mode

Receipt verification

Payload pattern

Expected distinct nodes

Payload speed primary endpoint

Payload speed LB endpoint

Payload speed custom endpoint

Payload speed additional endpoints

Custom plaintext

not run yet · Payload MB/s uses real payload-body bytes and excludes JSON envelope overhead.

Endpoint bakeoff + Ramp Test Mode + Soak Test

Compare primary, LB, custom, and additional endpoints one at a time, then ramp concurrency across the selected dispatch mode to identify saturation, or hold payload pressure steady for a soak window.

performance lab

Bakeoff contest endpoints

Ramp Test endpoint set

Soak Test endpoint set

concurrency steps

batches per step

payload size bytes

payloads per endpoint

payload/ramp concurrency

soak duration minutes

cooldown ms

not run yet · endpoint bakeoff, Ramp Test Mode, and Soak Test idle