Where
README.md → line 57, the Cluster bullet in the feature list
docs/src/content/docs/cluster/failure-detector.mdx → "Custom failure detector"
docs/src/content/docs/de/cluster/failure-detector.mdx → same section
What's wrong
README.md:57 advertises φ-accrual failure detection as a shipped Cluster capability:
- Cluster — gossip membership, φ-accrual failure detection, split-brain
The docs contradict this directly. cluster/failure-detector.mdx states:
The cluster's built-in detector is intentionally simple — plain elapsed-time
thresholds, no statistical variance tracking.
and, about PhiAccrualFailureDetector:
It's not yet wired into the cluster as a selectable detector — the cluster
always runs the simple FailureDetector — so for now you use it standalone.
The code agrees with the docs, not the README. PhiAccrualFailureDetector appears
only in its own definition files and the export barrel:
src/cluster/PhiAccrualFailureDetector.ts (definition)
src/cluster/PhiAccrualOptions.ts (definition)
src/cluster/index.ts:77-81 (re-export only)
src/cluster/Constants.ts:24 (a comment mentioning it)
Nothing in src/cluster/ constructs or consults it on the membership path.
So the README overstates a headline capability of a pre-1.0 framework: a reader
evaluating actor-ts for a partition-sensitive deployment is told the adaptive
detector is in the box, and only discovers on page three of the docs that the
cluster always runs fixed elapsed-time thresholds.
Suggested fix
Make the README bullet match reality, e.g.:
- Cluster — gossip membership, failure detection (φ-accrual detector
available standalone), split-brain
Both language mirrors of failure-detector.mdx are already correct and need no
change — this is a README-only fix, plus whatever wording keeps the two in step.
Kind
Factually wrong
Notes
Where
README.md→ line 57, the Cluster bullet in the feature listdocs/src/content/docs/cluster/failure-detector.mdx→ "Custom failure detector"docs/src/content/docs/de/cluster/failure-detector.mdx→ same sectionWhat's wrong
README.md:57advertises φ-accrual failure detection as a shipped Cluster capability:The docs contradict this directly.
cluster/failure-detector.mdxstates:and, about
PhiAccrualFailureDetector:The code agrees with the docs, not the README.
PhiAccrualFailureDetectorappearsonly in its own definition files and the export barrel:
Nothing in
src/cluster/constructs or consults it on the membership path.So the README overstates a headline capability of a pre-1.0 framework: a reader
evaluating actor-ts for a partition-sensitive deployment is told the adaptive
detector is in the box, and only discovers on page three of the docs that the
cluster always runs fixed elapsed-time thresholds.
Suggested fix
Make the README bullet match reality, e.g.:
Both language mirrors of
failure-detector.mdxare already correct and need nochange — this is a README-only fix, plus whatever wording keeps the two in step.
Kind
Factually wrong
Notes
into config), which tracks actually doing the wiring. This issue is only that
the README claims the work is already done. Closing [Feature] Wire the phi-accrual failure detector options into config #840 would make the current
README sentence true; until then it is not.
edge as AMBIGUOUS; verified by hand against
src/.