You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Can I get a report listing repeaters ranked by neighbor count? #2100
First, thank you very much for providing the CoreScope mesh analyzer. I am in Colorado and I administer about twenty repeaters, and your analyzer is a great help to me in keeping on top of the health of my repeaters.
Now to my "ask". If I drill down on any single node to see its "reach", the analyzer will say the rank of that node compared with the other nodes, ranked by neighbor count.
But suppose I'd like to find out which node is ranked number one? Is there a way to do that? Is there a way to list, say, the top twenty nodes ranked by neighbor count?
Thank you so much again and thank you in advance for your help on this.
Hi Carl, we ran into the same question on our deployment ("my node is Rank #X — who is #1?") and built a Reach leaderboard for it in our fork. In case it is useful for you or for upstream:
Lists every ranked node with its placement, name (or pubkey if it has none), neighbour count, and a link to the node's existing Reach page.
Searches by name or pubkey, and pages through results 50 rows at a time. A search never renumbers the list: a node keeps its global placement.
On each node's Reach page, the Rank card links to the leaderboard ("View leaderboard").
Includes a JSON endpoint: GET /api/reach-rank?q=&offset=&limit=.
How the rank is defined
The leaderboard uses the same rule and the same snapshot as the Rank card, so the two always agree.
Neighbour count: all-time distinct neighbours from neighbor_edges, within the ingestor's retention. Only valid pubkey pairs count, and duplicate or reversed pairs count once.
Ranked nodes: only nodes that have a Reach page. Blacklisted nodes and observers, and names that match the hidden-name prefixes, are left out.
Ties: competition ranking, so equal counts share a placement (1, 1, 3).
What it is not: the page labels it as a historical neighbour count. It is not a measure of radio quality or range.
Code:dborup/CoreScope#69. It adds a new page and endpoint, makes no schema changes, adds no dependencies, and comes with unit and E2E tests. Upstream is welcome to port it as is or adapt it. If a PR against upstream would help, we're happy to prepare one.
On 10/2/2026 10:49 AM, dborup wrote:
*dborup* left a comment (Kpa-clawbot/CoreScope#2100)
<#2100 (comment)>
Hi Carl, we ran into the same question on our deployment ("my node is
Rank #X — who is #1
<#1>?") and built a
*Reach leaderboard* for it in our fork. In case it is useful for you
or for upstream:
*Live:* https://meshview.dk/#/reach-rank (our deployment, Danish mesh)
*What it does*
* Lists every ranked node with its placement, name (or pubkey if it
has none), neighbour count, and a link to the node's existing
Reach page.
* Searches by name or pubkey, and pages through results 50 rows at a
time. A search never renumbers the list: a node keeps its global
placement.
* On each node's Reach page, the Rank card links to the leaderboard
("View leaderboard").
* Includes a JSON endpoint: |GET /api/reach-rank?q=&offset=&limit=|.
*How the rank is defined*
The leaderboard uses the same rule and the same snapshot as the Rank
card, so the two always agree.
* *Neighbour count:* all-time distinct neighbours from
|neighbor_edges|, within the ingestor's retention. Only valid
pubkey pairs count, and duplicate or reversed pairs count once.
* *Ranked nodes:* only nodes that have a Reach page. Blacklisted
nodes and observers, and names that match the hidden-name
prefixes, are left out.
* *Ties:* competition ranking, so equal counts share a placement (1,
1, 3).
* *What it is not:* the page labels it as a historical neighbour
count. It is not a measure of radio quality or range.
*Code:* dborup/CoreScope#69
<dborup#69>. It adds a new page and
endpoint, makes no schema changes, adds no dependencies, and comes
with unit and E2E tests. Upstream is welcome to port it as is or adapt
it. If a PR against upstream would help, we're happy to prepare one.
—
Reply to this email directly, view it on GitHub
<#2100?email_source=notifications&email_token=B4BCAZWBBAZJHFGJHXFZ5ED5R7MDNA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTKOJVG4YDIOJUGQ42M4TFMFZW63VGMF2XI2DPOKSWK5TFNZ2KYZTPN52GK4S7MNWGSY3L#issuecomment-5957049449>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/B4BCAZVKV7KGB3LYY5Y7BGD5R7MDNAVCNFSNUABGKJSXA33TNF2G64TZHMYTCOBVGYZDKMZUGI5US43TOVSTWNJWG4YDGMJWHA4TNILWAI>.
Triage notifications, keep track of coding agent tasks and review pull
requests on the go with GitHub Mobile for iOS
<https://github.com/notifications/mobile/ios/B4BCAZQE6JAGFPYSAXNR6K35R7MDNA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTKOJVG4YDIOJUGQ42M4TFMFZW63VGMF2XI2DPOKSWK5TFNZ2KUZTPN52GK4S7NFXXG>
and Android
<https://github.com/notifications/mobile/android/B4BCAZUQZG7QAE4KIJ4NE6D5R7MDNA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTKOJVG4YDIOJUGQ42M4TFMFZW63VGMF2XI2DPOKSWK5TFNZ2K4ZTPN52GK4S7MFXGI4TPNFSA>.
Download it today!
You are receiving this because you authored the thread.Message ID:
***@***.***>
Activity
dborup commentedon Oct 2, 2026
Hi Carl, we ran into the same question on our deployment ("my node is Rank #X — who is #1?") and built a Reach leaderboard for it in our fork. In case it is useful for you or for upstream:
Live: https://meshview.dk/#/reach-rank (our deployment, Danish mesh)
What it does
GET /api/reach-rank?q=&offset=&limit=.How the rank is defined
The leaderboard uses the same rule and the same snapshot as the Rank card, so the two always agree.
neighbor_edges, within the ingestor's retention. Only valid pubkey pairs count, and duplicate or reversed pairs count once.Code: dborup/CoreScope#69. It adds a new page and endpoint, makes no schema changes, adds no dependencies, and comes with unit and E2E tests. Upstream is welcome to port it as is or adapt it. If a PR against upstream would help, we're happy to prepare one.
carl-mempool commentedon Oct 2, 2026