Validator Performance

Read installed bundle and configuration before running commands. Resolve paths from this loaded SKILL.md, not the session working directory.

Run from the resolved bundle root:

bun src/skills/validator-performance/scripts/performance.ts \
  --vote-account <VOTE_ACCOUNT> \
  --epochs 30

Use --validator <VOTE_OR_IDENTITY> to resolve a current identity through the configured RPC. Add --include-current only when in-progress data is requested. Output formats: markdown, csv, or json.

The helper uses the operator-selected mainnet RPC for current epoch/vote state and JPool/SVT for historical epoch metrics. Apply the shared RPC selection policy in the runtime reference. No credential is bundled.

Report current state, completed epoch range, per-epoch metrics, totals/averages, and any unavailable epochs. Use validator-revenue when the question is about SOL earned.

When JPool/SVT returns no rows for the first requested epochs, the helper shortens the window to the available epochs and states which requested epochs are unavailable (requestedFirstEpoch and firstEpoch in JSON; standard error for CSV). Repeat that scope whenever quoting totals. The helper cannot tell a validator that started later from rows missing upstream, so do not describe the first available epoch as the validator's first epoch without another source. A gap after the first available epoch, or a missing latest epoch, still fails rather than producing a partial window.

With --include-current, the in-progress epoch is listed and marked, but upstream publishes partial or placeholder values for it, so it is excluded from window totals and averages. It is marked in every format: inProgressEpoch and a per-row inProgress in JSON, and an in_progress column in CSV. Its zero vote credits, leader slots and MEV commission are upstream placeholders and are reported as unavailable, not as zero. Do not read its row as a completed result. A window with no completed epoch has no totals.

Alpenglow credits (from epoch 1064)

Mainnet switched from TowerBFT to Alpenglow in epoch 1064 (migration slot 459,653,000). Every row carries creditScale (credit_scale in CSV, Credit Scale in Markdown): towerbft before 1064, alpenglow from 1064.

• TowerBFT epochs: TVC % is credits / (slots × 16) and TVC Rank is JPool/SVT tvcRank, unchanged.
• Alpenglow epochs: vote credits are lamports of nominal vote reward, not vote counts. Agave credits a voter half of epochInflation × stake / (slotsPerEpoch × totalStake) for each slot whose reward certificate includes it, and credits the leader that includes the certificate the other half for every included voter. Credits therefore scale with stake, epoch inflation and leader slots, so raw credits rank stake, not performance. TVC % is JPool/SVT creditsRatio (credits / that epoch's maximum) and TVC Rank is JPool/SVT tvcRank, which ranks creditsRatio across the cluster. Both are unavailable (— / null) while the ratio is unpublished, including the in-progress epoch.
• Epoch 1064 is a migration epoch: its credits also include about 5,000 slots of TowerBFT credits and its ratio is approximate.

Never sum, average or compare credits or TVC % across the two scales. The window summary reports credit totals and TVC % averages separately for each scale, with its epoch range; block production, skip rate and commission averages still cover the whole completed window. Quote the scale and range with every credit figure.

Skip rate and block production come only from leader-slot counts and are unavailable for an epoch without leader slots. JPool/SVT's skippedSlots field is the vote-credit shortfall, not a block skip rate, and is never used as one.

Local profiles and first use

Read chain profile selection and first use. Use --profile NAME and optional --config PATH to reuse setup; resolve missing configuration through onboarding, then resume this check.