fix: drop empty APR timeseries buckets

[matheus1lva]

Summary

Fixes bad zero/empty interior APR datapoints by dropping aggregate buckets that contain no non-zero samples. The existing SQL already treated zero values as missing with NULLIF; this removes the COALESCE fallback that converted fully empty buckets back into chart-visible zeroes.

How to review

• Review the timeseries aggregation changes in packages/web/app/api/gql/resolvers/timeseries.ts and packages/web/app/api/rest/timeseries/db.ts.
• Confirm both GraphQL and REST/cache refresh paths use the same rule: average non-zero samples, then omit buckets where the average is null.
• Confirm this does not add interpolation or smoothing; it only removes buckets made entirely of zero/null samples.
• Note: the unlocked yvUSD wiggle is not fixed here. This change should make bad zero drops disappear, while any remaining oscillation should be investigated in the upstream yvUSD APR webhook/provider data.

Local/staging setup notes

• For API-only validation against existing affected rows, set y our POSTGRES_* env vars to point to the read-only replica.
• have redis running with docker-compose up -d redis
• set redis into the env vars as well

Test plan

• Manual: start local web API from repo root.

bun --filter web dev

• Manual: in a second terminal, set endpoint variables.

export API_ORIGIN="http://localhost:3000"
export GQL_URL="${API_ORIGIN%/}/api/gql"
export LOCKED_YVUSD="0xAaaFEa48472f77563961Cdb53291DEDfB46F9040"
export UNLOCKED_YVUSD="0x696d02Db93291651ED510704c9b286841d506987"

• Manual: run raw locked yvUSD netAPR GraphQL curl.

curl -sS "$GQL_URL" \
  -H 'content-type: application/json' \
  --data-binary @- <<'JSON'
{
  "query": "query LockedYvusdNetApr { timeseries(chainId: 1, address: \"0xAaaFEa48472f77563961Cdb53291DEDfB46F9040\", label: \"yvusd-estimated-apr\", component: \"netAPR\", period: \"1 day\", limit: 1000) { chainId address label component value period time } }"
}
JSON

• Manual: run locked yvUSD netAPR validation. Expected result: output shows zeroRows: 0, nonNumericRows: 0, unordered: 0, and rows greater than 0 if DB has the affected series.

curl -sS "$GQL_URL" \
  -H 'content-type: application/json' \
  --data-binary @- <<'JSON' \
| jq '{
  rows: (.data.timeseries // [] | length),
  zeroRows: (.data.timeseries // [] | map(select(.value == 0)) | length),
  nonNumericRows: (.data.timeseries // [] | map(select((.value | type) != "number")) | length),
  unordered: (.data.timeseries // [] as $rows | [range(1; ($rows | length)) as $i | select(($rows[$i - 1].time | tonumber) > ($rows[$i].time | tonumber))] | length)
}'
{
  "query": "query LockedYvusdNetApr { timeseries(chainId: 1, address: \"0xAaaFEa48472f77563961Cdb53291DEDfB46F9040\", label: \"yvusd-estimated-apr\", component: \"netAPR\", period: \"1 day\", limit: 1000) { chainId address label component value period time } }"
}
JSON

• Manual: run unlocked yvUSD netAPR validation. Expected result: output shows zeroRows: 0 and prints min / max. Any remaining non-zero wiggle is source/provider behavior, not this SQL fallback.

curl -sS "$GQL_URL" \
  -H 'content-type: application/json' \
  --data-binary @- <<'JSON' \
| jq '{
  rows: (.data.timeseries // [] | length),
  zeroRows: (.data.timeseries // [] | map(select(.value == 0)) | length),
  min: (.data.timeseries // [] | if length > 0 then map(.value) | min else null end),
  max: (.data.timeseries // [] | if length > 0 then map(.value) | max else null end)
}'
{
  "query": "query UnlockedYvusdNetApr { timeseries(chainId: 1, address: \"0x696d02Db93291651ED510704c9b286841d506987\", label: \"yvusd-estimated-apr\", component: \"netAPR\", period: \"1 day\", limit: 1000) { chainId address label component value period time } }"
}
JSON

• Manual: compare against production for locked yvUSD netAPR. Expected result: local branch has no zero rows; production may still show zero rows until this PR deploys and cache refresh runs.

for URL in "https://kong.yearn.fi/api/gql" "$GQL_URL"; do
  echo "Checking $URL"
  curl -sS "$URL" \
    -H 'content-type: application/json' \
    --data-binary @- <<'JSON' \
  | jq '{
  rows: (.data.timeseries // [] | length),
  zeroRows: (.data.timeseries // [] | map(select(.value == 0)) | length)
}'
{
  "query": "query LockedYvusdNetApr { timeseries(chainId: 1, address: \"0xAaaFEa48472f77563961Cdb53291DEDfB46F9040\", label: \"yvusd-estimated-apr\", component: \"netAPR\", period: \"1 day\", limit: 1000) { chainId address label component value period time } }"
}
JSON
done

• Manual: run REST cache historical refresh in staging/local if validating cached chart payloads. Expected result: command completes and cached data omits all-zero buckets.

bun packages/web/app/api/rest/timeseries/refresh-historical.ts

• Manual: run REST latest refresh in staging/local if validating latest cache override behavior. Expected result: latest cache does not reintroduce zero rows.

bun packages/web/app/api/rest/timeseries/refresh.ts

• Manual: open affected chart after refresh and confirm locked yvUSD visible zero drop/spike is gone.

open "$API_ORIGIN"

• Automated (when applicable): web tests passed locally before PR update.

bun --filter web test
# Expected: 5 pass, 0 fail

• Automated (when applicable): whitespace check passed locally before PR update.

git diff --check
# Expected: no output, exit code 0

Risk / impact

• Impact is limited to timeseries aggregation responses: all-zero/all-null buckets are omitted instead of emitted as value 0.
• This matches the existing intent of NULLIF(value, 0), which already treated zeros as missing samples.
• Legitimate non-zero samples in a bucket are still averaged and returned unchanged.
  EOF 2>&1

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
kong	Ready	Preview, Comment	May 9, 2026 9:50pm