feat(horizon): expose Provision.delegatedTokensActive as the APR/APY denominator

[cargopete]

feat(horizon): expose Provision.delegatedTokensActive as the APR/APY denominator

Summary

Adds an explicit, maintained delegatedTokensActive field to the Provision entity:

delegatedTokensActive = delegatedTokens - delegatedThawingTokens

This is the correct denominator for delegation APR/APY. It already exists implicitly
(the subgraph computes the same subtraction for delegationExchangeRate and indexer
capacity), but consumers computing returns tend to divide by delegatedTokens, which
includes both in-period thawing and completed-but-not-yet-withdrawn delegation —
systematically depressing reported APR/APY.

Why this is the right layer

Thawed-but-unwithdrawn delegation remains counted in delegatedThawingTokens (mirroring
the on-chain tokensThawing, which is only decremented when withdrawDelegated fulfils a
request — not when the thaw clock merely expires). So subtracting delegatedThawingTokens
already excludes it. No protocol change is required; this is purely a reporting fix, which
is the conclusion reached in the forum discussion
(thread).

For consumers who additionally want the in-period vs completed split of thawing
delegation, the existing ThawRequest entity (thawingUntil, fulfilled, type) is
already sufficient to compute it client-side.

Changes

• schema.graphql — new Provision.delegatedTokensActive: BigInt! with documentation.
• helpers.ts
  • createOrLoadProvision: initialise the field.
  • updateAdvancedProvisionMetrics: recompute it (covers delegate, add-to-pool,
    undelegate and slash paths, which already call this helper).
• horizonStaking.ts — handleDelegatedTokensWithdrawn: set the field directly (this
  handler doesn't go through updateAdvancedProvisionMetrics).

Validation

• yarn prepare:arbitrum (codegen) ✅ — schema parses, types regenerate.
• yarn build (AssemblyScript compile) ✅.

Note: yarn test currently fails to compile on master due to a pre-existing,
unrelated issue in tests/helpers.test.ts (calls createOrLoadIndexer with 2 args; it
takes 3). Not touched by this PR.

Notes / scope

• Field added to Provision (the Horizon delegation unit). The legacy Indexer-level
  equivalent is derivable the same way (Indexer.delegatedTokens - Indexer.delegatedThawingTokens) and is already used internally for capacity; happy to
  add an Indexer.delegatedTokensActive mirror if wanted.
• Subgraphs re-index from genesis on deploy, so the new non-null field is populated for
  all provisions via createOrLoadProvision; no migration concern.

[cargopete]

Context: off-chain fix for the delegation APR/APY distortion

This PR is the off-chain half of the fix discussed in forum: Separate withdrawable bucket for fully-thawed delegation tokens and Accurately Representing Delegated GRT in APR/APY Calculations.

The key point: thawed-but-unwithdrawn delegation stays counted in delegatedThawingTokens (mirroring on-chain tokensThawing, which is only decremented on withdrawDelegated, not when the thaw clock expires). So delegatedTokens - delegatedThawingTokens is already the correct actively-earning base — the distortion comes purely from consumers dividing by delegatedTokens. This PR surfaces that base explicitly as Provision.delegatedTokensActive, requiring no protocol change.

For completeness, a hardened protocol-level variant (exposing the same split on-chain, with the storage/slashing issues from the original attempt fixed) is available at graphprotocol/contracts#1340 — but the recommendation is to fix this off-chain via this subgraph change.

[tmigone]

Looks good but please check tests that are failing.

[cargopete]

Thanks @tmigone. I dug into the failing checks — none of the failures are caused by this PR. Here's what I found and what I've done:

1. Compile error (the blocker that aborts the whole suite) — tests/helpers.test.ts called createOrLoadIndexer with 2 args after it gained a third (graphNetwork) parameter. This makes graph test fail to compile before any test runs. Fixed in this PR (passes the loaded graph network through).

2. Four runtime failures in staking.test → StakeDelegated (DelegatedStake with id …006-…003 not found). These are pre-existing on master and unrelated to this change. I verified by reverting this PR's mapping/schema changes to master and re-running: identical result, 4 failed, 84 passed. The mismatch is in the legacy DelegatedStake id scheme (the test expects delegator-indexer, the migrated handler now builds a Horizon-style id) — it looks like it belongs with the in-flight legacy-test work rather than here, so I've left it untouched to avoid conflicting with that. Happy to take it on in a separate PR if useful.

So with the compile fix, the suite runs and this PR's own paths are green; the remaining 4 are a pre-existing legacy-migration item. I've also re-signed the commits so the signature verifies.