Problem
The registry file and its git history contain the full parameter record (emission_share, maintainer_cut, issue_discovery_share, label multipliers, scoring overrides) but in a form only developers can read.
Scope
- Ingestion lane: fetch the registry file from the upstream repo on a schedule; diff against the last stored version; persist versioned records keyed by repo + parameter with effective timestamps and source commit SHAs
- Backfill: walk the file's git history once to reconstruct the complete parameter timeline from first appearance
- Handle repo additions/removals (registration/deregistration events) as first-class records
- Provenance on every record: source commit SHA + upstream file path, so every value is checkable against the upstream repo
Forward source note
Upstream is moving repo hyperparameters toward an API (entrius/gittensor#1511, open). The registry file + git history remain canonical for the timeline (commit provenance is the point); when the upstream API lands, add it as the live-values source and reconcile against the latest file commit — a divergence between the two is itself a signal worth surfacing.
Implementation pointers (v1)
- Script:
scripts/sync-sn74-registry.ts (new), modeled on scripts/sync-subnets.ts / scripts/snapshot-adapters.ts
- History: GitHub commits API filtered by file path on entrius/gittensor
master_repositories.json; parse each version, diff params per repo, emit dated change records with source SHAs — one backfill pass, then incremental on schedule
- Artifact: adapter lane, slug
sn74-registry, validated against schemas/components/09-schemas-adapters-r2.schema.json conventions; registry/adapters + R2 snapshot path
- Workflow: copy
.github/workflows/gittensor-impact.yml skeleton (scheduled SN74 job precedent) — daily cron + workflow_dispatch, scoped write like that file
Acceptance
Every parameter value ever present in the registry file exists as a dated, commit-linked record; a spot-check of any record against upstream git history matches.
Problem
The registry file and its git history contain the full parameter record (emission_share, maintainer_cut, issue_discovery_share, label multipliers, scoring overrides) but in a form only developers can read.
Scope
Forward source note
Upstream is moving repo hyperparameters toward an API (entrius/gittensor#1511, open). The registry file + git history remain canonical for the timeline (commit provenance is the point); when the upstream API lands, add it as the live-values source and reconcile against the latest file commit — a divergence between the two is itself a signal worth surfacing.
Implementation pointers (v1)
scripts/sync-sn74-registry.ts(new), modeled onscripts/sync-subnets.ts/scripts/snapshot-adapters.tsmaster_repositories.json; parse each version, diff params per repo, emit dated change records with source SHAs — one backfill pass, then incremental on schedulesn74-registry, validated againstschemas/components/09-schemas-adapters-r2.schema.jsonconventions; registry/adapters + R2 snapshot path.github/workflows/gittensor-impact.ymlskeleton (scheduled SN74 job precedent) — daily cron + workflow_dispatch, scoped write like that fileAcceptance
Every parameter value ever present in the registry file exists as a dated, commit-linked record; a spot-check of any record against upstream git history matches.