Parent: #11066. Producer: the enrich container class in metagraphed-infra.
Why the current signals are thin
src/github-signals-sync.ts runs 20 6 * * * capped at 230 repos per run by the
1,000-subrequest ceiling, and reads only what the GitHub API exposes. Several subnets' signals read
as stale because there are no releases to read — which says something about the release process, not
the project.
The signal that matters most for this registry
Does the declared API surface match the code? We publish "how to call this subnet." A declared
OpenAPI that no handler implements is the highest-value gap we can find, and list_gaps has nowhere
to learn it today.
Alongside it: does it build, do the tests pass, real commit cadence, bus factor, dependency
freshness, licence as declared in the tree.
Scope
- New signal fields on the subnet profile and the curation/gap surfaces
- Feeds
list_gaps, list_enrichment_targets and the coverage/curation levels
- Every signal carries its observation date; a build result from three weeks ago is not a current
claim
Fairness discipline
These signals rank other people's projects. Publish the method, make each signal explain itself
(what "does not build" meant — which command, which output), and never let a signal we cannot
compute read as a negative one. absent is null, never zero applies with force here.
Acceptance
Parent: #11066. Producer: the enrich container class in metagraphed-infra.
Why the current signals are thin
src/github-signals-sync.tsruns20 6 * * *capped at 230 repos per run by the1,000-subrequest ceiling, and reads only what the GitHub API exposes. Several subnets' signals read
as stale because there are no releases to read — which says something about the release process, not
the project.
The signal that matters most for this registry
Does the declared API surface match the code? We publish "how to call this subnet." A declared
OpenAPI that no handler implements is the highest-value gap we can find, and
list_gapshas nowhereto learn it today.
Alongside it: does it build, do the tests pass, real commit cadence, bus factor, dependency
freshness, licence as declared in the tree.
Scope
list_gaps,list_enrichment_targetsand the coverage/curation levelsclaim
Fairness discipline
These signals rank other people's projects. Publish the method, make each signal explain itself
(what "does not build" meant — which command, which output), and never let a signal we cannot
compute read as a negative one.
absent is null, never zeroapplies with force here.Acceptance
null, never a failing score