aboutsummaryrefslogtreecommitdiffstats
path: root/docs/federation.md
diff options
context:
space:
mode:
Diffstat (limited to 'docs/federation.md')
-rw-r--r--docs/federation.md14
1 files changed, 8 insertions, 6 deletions
diff --git a/docs/federation.md b/docs/federation.md
index 53da9da..ad2f507 100644
--- a/docs/federation.md
+++ b/docs/federation.md
@@ -572,10 +572,12 @@ operator pins.
* **Changes it shouldn't apply.** A change is first of all one Automerge itself
writes, or it is refused before Automerge sees it, in every swarm *(checked:
`document::changes::Written`)*.
-* **Writers a node starts.** A node starts at most so many writers in an epoch, a bound
- judged at a change's history as any rule is: every writer costs every node memory
- (below), and one node starting one with every change would run the others out of
- memory, which is more than slow.
+* **Writers a party's nodes start.** A party's nodes start at most so many writers in
+ an epoch together, the party each node's entry names counting, a bound judged at a
+ change's history as any rule is: every writer costs every node memory (below), and a
+ party starting one with every change, on as many nodes as it lists, would run the
+ others out of memory, which is more than slow. The figure leaves room for its honest
+ nodes' starts, rejoins and drops, each a writer, until a remake.
The rest are recommendations: none breaks a promise, only how fast a node is, or how
long nodes stay together.
@@ -711,7 +713,7 @@ What an attacker does, and what happens. *Unit*: the judging, with crafted chang
| 75 | A version, set, list, path or value larger than the federation allows | Refused; none can be too large for a sync message | unit |
| 76 | Two thirds back a proposal that is no version | It can't be written, nor die; the federation is bound to its version. A node shows that it is none, and clients don't back it | unit |
| 77 | A party's vote committed again at a history before the nays that freed it, its clock set between, then a remake | The replay refuses it at first, and takes it once the rest is in | unit |
-| 78 | A listed node starting a fresh writer with every change | Refused past the bound on the writers a node starts in an epoch | unit |
+| 78 | A party's nodes starting a fresh writer with every change, or the party listing new nodes to start more | Refused past the bound on the writers a party's nodes start in an epoch | unit |
| 79 | Keys voting twice, or a writer forked, part honest nodes | An operator has its node drop what it took and take the other side's; no fork | sim |
| 80 | An honest party backs a second proposal once its first is dead; another party later signs `true` on the first too | It isn't named: its first proposal is dead by all the votes; the party with `true` and `false` on one is | unit |
| 81 | A node whose operator had it drop a version is sent it again, or holds the votes that passed it | It refuses those changes and writes the version no more, in that epoch, until its operator says otherwise | unit |
@@ -827,7 +829,7 @@ On a federated swarm:
touching what the federation reads (versions, proposals, votes, key sets, node
lists), as `access::screen` moves it today for what continuum's rules read; the byte
formats (a version's JCS bytes, hashes in paths and how two compare, the meta's
- version); the bound on the writers a node starts in an epoch; and the sizes a
+ version); the bound on the writers a party's nodes start in an epoch; and the sizes a
version, a set, a list, a path and a value may have, each within what a sync message
carries. A node that doesn't know the policy stops taking changes. Until stage 5, a
federated swarm takes no remake: one that grows too large starts afresh by a fork