Skip to content

superpios/node-scorecard

Repository files navigation

Sentinel Node Scorecard

Network intelligence & historical diagnostics for Sentinel dVPN nodes.

The Scorecard answers the question that matters most to a node operator: how has my node behaved over time, and how does it compare to the rest of the network?

It samples every node on the Sentinel network on a schedule, builds a historical record measured in hundreds of thousands of samples, and turns it into an actionable reliability score, a country ranking, time-series trends, and a node-by-node diagnostic — all from data collected on-device, with no backend.


What it does

🗺️ Network Explorer

The whole network at a glance. Every node, sortable and filterable by:

  • Score (computed with real variance — see below)
  • Status (active / inactive) — inactive nodes hidden by default, toggle to show
  • Protocol (V2Ray / WireGuard / OpenVPN)
  • Price (P2P/hr and P2P/GB)
  • Peers, historical uptime %, country / city
  • ASN / hosting provider — which network each node runs on, with saturation flags

Search by moniker, address or city. Click any node to open its full diagnostic inline.

🔬 Node diagnostic

Click any node (or paste an address) to deep-dive:

  • Historical uptime timeline — a visual active/inactive bar for every sample collected over time
  • Measured uptime % and stability (how often it flaps offline)
  • Country ranking — "#12 of 47 in Italy", not a meaningless absolute number
  • Detected problems — 16 rule-based diagnostics: inactive, low uptime, unstable, API unreachable, stale heartbeat, zero peers, price outliers (per hr and per GB vs network median), very low bandwidth, outdated version, saturated ASN, crowded country, young node, missing speedtest — plus a historical uptime trend check (declining / recovering, computed from the node's own timeline), a lease check that catches the most frustrating case — a perfectly healthy node that gets no client traffic because it isn't whitelisted for plans (peers ≈ 0 across recent history) — and an elite badge for top-20 ranked nodes
  • Reliability grade — an A+/A/B/C/D/E letter from the historical score, with a global rank ("#12 of 1,594 active nodes")
  • Score breakdown — how each weighted component contributes
  • Generate support report — a ready-to-paste summary for the Sentinel support Telegram or a GitHub issue, so a node's problems actually reach the people who can help

💬 NodeAdvisor — community feedback

The data tells you how a node behaves; NodeAdvisor tells you how it feels to use. On any node, anyone can leave structured feedback:

  • Quick tags — positive (fast, stable, good uptime, good value, connects first try) and negative (drops often, high latency, slow, hard to reach)
  • Optional comment (up to 500 chars)
  • Aggregated view — feedback is shown as tag counts ("12 · fast connection · 3 · high latency"), not a single falsifiable star rating
  • Light anti-abuse — one vote per node per browser, plus server-side de-duplication. Not bulletproof (nothing browser-side is), but it keeps casual self-voting down, and comments can be moderated.

Feedback is stored in a free Supabase table read directly from the static page — still no server of your own to run.

🌐 ASN Intelligence — "Where to host"

A dedicated view for deciding where to put a new node: it groups active nodes by hosting provider (ASN), counts how many run on each, and flags the saturated ones — and now tells you, honestly, which providers are even worth considering.

  • Per-ASN node counts & saturation flags — ASNs over the PlanWizard whitelist limit are marked SATURATED; nodes there rarely get leased
  • Least crowded vs most crowded — every country below the free-slot threshold, ranked, so you can spot genuine opportunities
  • Lease % per ASN — not just how many nodes, but how many are actually earning a lease there
  • Provider honesty badges — recommended hosts (verified on their official site) get a direct link; major clouds (AWS, Oracle, Alibaba…) carry a may-not-be-whitelisted warning per the Sentinel docs; national ISPs / telecoms are flagged as not a hosting provider; everything else is marked unverified — check before use. "Free slots" no longer misleads you toward a phone company or a reseller
  • Lease uptime (historical) — for each node, the % of the last 7 days it actually held an on-chain lease — the real "am I earning?" signal, distinct from a one-off snapshot
  • Hosting column + diagnostic — every node shows its ASN in the table, and a node on a saturated ASN gets a clear warning in its diagnostic ("may not get leased/earn")
  • Pick a clean ASN — low-count, non-saturated providers are the smart choice for a new node

This turns a painful, learned-the-hard-way lesson (datacenter ASNs like IONOS, Oracle and OVH are saturated; residential ASNs are not) into something you can see before spending a cent. The same view doubles as a node placement advisor: search an ASN or a country and get an instant verdict (⛔ saturated / ⚠️ filling up / ✅ room available) with capacity-used %, plus the least- and most-crowded countries.

🏆 Top Reliable

The 25 most reliable active nodes ranked by historical uptime, stability and track record — not by who happens to be online right now. A node only ranks here by being consistently up across many samples.

📈 Trends — the network over time

The Scorecard records a snapshot of the network every 2 hours and shows how it moves over time:

  • Network Pulse — active-node count over time
  • ASNs gaining / losing nodes — watch providers fill up or empty out (IONOS went 285 → 188 active nodes in two days; this view catches that)
  • Node movers — nodes whose 7-day uptime improved or declined the most vs their own history

🧊 Time Capsule — on Walrus (Sui)

The Scorecard's historical snapshot is archived on Walrus, the decentralized storage protocol on Sui. Each capsule is an immutable, content-addressed blob with a verifiable hash and a public link shown in the Trends tab. The history of a decentralized VPN, stored on decentralized storage.

The Scorecard itself is also published as a Walrus Site — the static page lives on Walrus, fully decentralized, with no central server hosting it. A planned next step is integrating Walrus Memory to give an AI agent persistent, verifiable memory of node history — an assistant that remembers how the network and individual nodes have behaved over time.


The score

A 0–100 score built to have real variance (no more "everyone gets 90"):

Component Weight Source
Uptime 32% measured % active over collected history
Longevity 25% history samples / network max (proven track record)
Stability 15% penalizes frequent active↔inactive flips
Freshness 13% recency of the on-chain heartbeat
API health 10% node API reachable + active on-chain
Payment tokens 5% number of accepted denoms

New nodes start lower on longevity and climb as they build a track record — exactly how trust works.


Architecture

Everything runs on a single always-on device (e.g. a Raspberry Pi). No server, no database, no cost.

collector.py     -> samples every node from the Sentinel LCD (status, price, protocol,
                    peers, bandwidth), appends to history.jsonl. Retries each node's
                    API twice to minimise blank fields from transient timeouts.
                    Enriches active nodes with their ASN (IP->ASN lookup, cached on
                    disk so each IP is resolved only once).
make-summary.py  -> aggregates history.jsonl into history-summary.json (uptime %,
                    stability, longevity + a compact per-node timeline). Soft-purges
                    nodes not seen for 30+ days to keep the files lean.
make-trends.py   -> appends a network snapshot per run (active count, per-ASN and
                    per-country totals) and computes node movers (7d uptime vs prior)
                    from raw history. Writes + publishes trends.json.
walrus-capsule.py-> once a day, archives trends.json on Walrus and publishes
                    capsule.json with the blob ID, hash and verification link.
push-latest.sh   -> publishes latest.json + history-summary.json to this repo
index.html       -> static UI, fetches the two JSON files and renders everything
                    client-side. Community feedback (NodeAdvisor) is read/written
                    directly to a free Supabase table from the browser.

The UI is a single static HTML file — host it on GitHub Pages or any static host. It reads the published JSON directly. The only external dependency is a free Supabase project for NodeAdvisor feedback (optional — the rest works without it).


API / Data access (for developers & white-label apps)

All Scorecard data is served as plain JSON over HTTPS — no key, no auth, no rate limit. White-label apps, dashboards, or bots can consume it directly. These are read-only endpoints.

Per-node current snapshot — latest.json

The most recent sample of every node on the network.

GET https://superpios.github.io/node-scorecard/latest.json

Returns an array of node objects:

Field Type Meaning
addr string Node address (sentnode1…)
moniker string|null Node name (null if its API didn't answer at sample time)
status string active / inactive (on-chain)
ts string ISO timestamp of this sample
inactive_at string On-chain heartbeat-expiry timestamp
remote string|null Announced host:port
protocol string|null v2ray / wireguard / openvpn
peers int|null Connected peers at sample time
price_hr / price_gb number|null Price in P2P per hour / per GB
dl_mbps / ul_mbps number|null Measured bandwidth
version string|null Node software version
country / city string|null Geolocation
asn string|null Hosting ASN + provider (e.g. AS8560 IONOS SE); active nodes only
gb_tokens int Number of accepted payment denoms
api_ok bool Whether the node's API answered the collector

Per-node historical summary — history-summary.json

Aggregated history + a compact uptime timeline per node.

GET https://superpios.github.io/node-scorecard/history-summary.json
Field Type Meaning
a string Node address
mon string Moniker (or address if unnamed)
n int Number of samples collected
uptime number % active across all samples
stab number Stability score (0–100)
trans int Count of active↔inactive transitions
ul number|null Average upload Mbps
country string|null Country
sc int Rough composite score
tl string Compact timeline: 1=active, 0=inactive, oldest→newest
tl_from / tl_to string Timeline start / end timestamps
stale bool Last sample older than 24h
age_h number|null Hours since last sample

The on-page score uses a richer weighting than the sc field above; sc is a lightweight pre-computed approximation. To reproduce the full score, see the weights table under The score.

Network time series — trends.json

One snapshot every ~3 hours: {ts,total,active,stale,api_ok,asn:{...},country:{...}} per point, plus movers_up/movers_down (nodes whose 7-day uptime changed most vs their prior history).

GET https://superpios.github.io/node-scorecard/trends.json

Walrus archive index — capsule.json

The latest and last 30 daily Walrus capsules: {date,blobId,sha256,bytes,url,network}. Fetch the url to retrieve the archived snapshot from a Walrus aggregator and verify its sha256.

GET https://superpios.github.io/node-scorecard/capsule.json

Community feedback (NodeAdvisor)

Feedback is stored in Supabase and is readable via its auto-generated REST API (read policy only). Example — fetch visible feedback for one node:

GET https://<project>.supabase.co/rest/v1/node_feedback?node_addr=eq.<addr>&hidden=eq.false&select=tags,comment,created_at
Header: apikey: <publishable-key>

Feedback writing goes through the web UI, which applies anti-abuse measures (one vote per device, de-duplication). Writing directly via API bypasses those, so it isn't documented here by design.

Notes for integrators

  • Data refreshes roughly every 2 hours; cache accordingly.
  • Fields can be null for nodes whose API was unreachable at sample time (~half the network at any moment) — handle nulls gracefully.
  • This is a community project; endpoints may evolve. Open an issue if you build on it and need stability guarantees.

Data sources & honest limits

  • Prices, status, heartbeat come from the Sentinel LCD and are available for every node.
  • Protocol, peers, bandwidth, version come from each node's local API — available only for nodes that expose it publicly, so these fields are blank for unreachable nodes (roughly half the network at any moment). This is a limit of public data, not a bug.
  • Historical metrics improve the longer the collector runs. A fresh install needs a few days of samples before uptime/timeline are meaningful.
  • The diagnostic's optional live peer fetch hits a node's API directly from your browser; nodes use self-signed certs, so most browsers block it and the value falls back to the last collector reading. Expected behaviour, not an error.

Setup

# on the always-on device
git clone https://github.com/superpios/node-scorecard   # or download the scripts
cd node-scorecard

# first manual run (takes ~10 min for the full network)
python3 collector.py --all
python3 make-summary.py

# schedule collection (crontab -e) - every 2 hours
0 */2 * * * /usr/bin/python3 /path/to/collector.py --all >> collector.log 2>&1 && \
            /usr/bin/python3 /path/to/make-summary.py >> collector.log 2>&1 && \
            /usr/bin/python3 /path/to/make-trends.py >> collector.log 2>&1 && \
            /bin/bash /path/to/push-latest.sh >> collector.log 2>&1

# daily Walrus capsule (optional)
50 23 * * * /usr/bin/python3 /path/to/walrus-capsule.py >> capsule.log 2>&1

Then publish latest.json and history-summary.json to the repo (see push-latest.sh) and open index.html.


Roadmap

  • ✅ Public read-only JSON API (done — see API section)
  • ✅ ASN intelligence — per-provider saturation + placement advisor (done)
  • ✅ Reliability grade + global rank, Top Reliable leaderboard (done)
  • ✅ Trends — network time series, ASN movers, node movers (done — accumulating history)
  • ✅ Walrus Time Capsule on Sui (done)
  • ✅ Published as a fully decentralized Walrus Site (done)
  • ✅ "Where to host" — provider honesty badges, lease %, per-country opportunity (done)
  • On-chain earnings trend per node → ROI calculator (opt-in, by wallet)
  • Objective ASN classification via PeeringDB (hosting / ISP / enterprise)
  • Walrus Memory integration — persistent memory for an AI node-advisor agent
  • Clean versioned API endpoint with developer-friendly field names
  • Multi-network support (agnostic core - Sentinel first, others later)
  • Alerting hooks for operators

Ecosystem

Friends & complementary tools: BlueFrens Hub · SentNodes · SuchNode · p2pscan · stats.sentinel.co


Built by a node operator, for node operators. Not affiliated with the Sentinel Foundation.

About

Network intelligence & historical diagnostics for Sentinel dVPN nodes. Reliability scores, lease tracking, "where to host" advisor — no backend, fully static, archived on Walrus.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors