feat(statistics): assurances before rollover + 3-tuple accumulation stats + C(13) prefixes (GP v0.8.0) (#1021)

[HanaYukii]

⚠️ Behavioural + state-root change

Not a pure notation refactor. Assurances are now credited before the epoch rollover — this changes validator-statistics values on epoch-boundary blocks — and the per-service accumulation stat becomes a 3-tuple, which changes the statistics state bytes (state root). Please review the ordering and the wire shape, not just the length prefix.

Part of #1012 · Closes #1021

Stacked on #1033 (← #1032 ← #1031 ← #1027 ← #1026). Base is feat/accumulation-1019-accseq-v080 (this PR consumes its processed-transfers output); review only this commit. Retarget as the stack merges.

What — three v0.8.0 changes

#	change	GP ref
1	Assurance ordering (behavioural): assurances credit π_V† before the epoch rollover — on an epoch-boundary block they roll into π_L′, not the fresh accumulator. Other five counters stay post-rollover	statistics.tex activity-spec restructure
2	Accumulation stats become 3-tuples (N(s), T(s), G(s)): T(s) counts processed transfers per destination service (from #1033's eq:accseq 5th output). Transfer-only services enter the key domain; (0,0,0) entries drop. Wire: ServiceActivityRecord gains accumulate_transfers_count between count and gas	eq:accumulationstatisticsdef/spec
3	C(13): π_V / π_L become var length-prefixed sequences (decode still validates count = |κ|)	merklization.tex C(13)

Already aligned: vector sizing to |κ|/|λ| (hard-validated to ValidatorsCount; offenders zeroed in place), update_statistics.go STF wiring.

Tested

• ✅ TestValidatorStats_AssurancesBeforeEpochRollover (new) — epoch-boundary: assurance lands in π_L′, fresh π_V′ stays 0; same-epoch: lands in π_V′
• ✅ TestCalculateAccumulationStatistics_V080Transfers (new) — T counting, transfer-only service inclusion, zero-tuple drop
• ✅ TestEncodeValidatorsStatistics_V080LengthPrefix (new) — prefix + round-trip
• ✅ statistics, accumulation, types, extrinsic, jam_types green; build/vet/gofmt clean
• ⏭️ statistics + preimages STF modes skipped (preimages vectors carry π_S in state, hit by the 3-tuple change); matching unit vector tests skipped. validate_fuzz.sh statistics expectation moves 1/2 → 0/0 while the mode is skipped — restore with official v0.8.0 vectors (Update to v0.8.0 #1012)

[yu2C]

Summary

Solid v0.8.0 statistics alignment. The three behavioural/wire changes are implemented consistently: assurances credit π_V† before epoch rollover (with a focused boundary test), accumulation stats become (N, T, G) 3-tuples wired through accumulation → statistics → codec, and π_V/π_L gain C(13) var length prefixes. Vector skips and validate_fuzz.sh expectation updates match the rest of the v0.8.0 stack.

One housekeeping follow-up: fold the new behavioural test into statistics_test.go instead of a separate statistics_v080_test.go.

---

Issue Verification

Test	Status	Evidence
go test ./internal/statistics/... -count=1	PASS	exit 0; TestValidatorStats_AssurancesBeforeEpochRollover passes; vector tests skipped
TestCalculateAccumulationStatistics_V080Transfers	PASS	go test ./internal/accumulation/... -run V080 exit 0
TestEncodeValidatorsStatistics_V080LengthPrefix	PASS	go test ./internal/types/... -run ValidatorsStatistics exit 0
Statistics / preimages STF vectors	SKIPPED	pending official v0.8.0 vectors (#1012)

Linked issues: #1021 (Closes), #1012 (Part of)

---

internal/statistics/statistics_v080_test.go

[SUGGESTION] Fold v0.8.0 behavioural test into statistics_test.go

TestValidatorStats_AssurancesBeforeEpochRollover is the only test in this new file. statistics_test.go already owns vector skips and TestMain; please move this test there and delete statistics_v080_test.go to match the pattern used elsewhere in the v0.8.0 stack (codec round-trips stay in internal/types/encode_test.go; package behavioural tests live in the main *_test.go).

---

internal/accumulation/accseq_v080_test.go

[NIT] Consider moving TestCalculateAccumulationStatistics_V080Transfers to accumulation_test.go

This test exercises calculateAccumulationStatistics (3-tuple / T(s) counting), not accseq prefix length or gas budget. It would read more clearly alongside the other accumulation integration tests. Low priority — behaviour is correct and the test passes.

[HanaYukii]

Thanks @yu2C — both applied (63dabfd0):

• SUGGESTION: TestValidatorStats_AssurancesBeforeEpochRollover folded into statistics_test.go; statistics_v080_test.go deleted.
• NIT: TestCalculateAccumulationStatistics_V080Transfers relocated to accumulation_test.go next to the other accumulation integration tests — accseq_v080_test.go now holds only the eq:accseq prefix/recursion tests.

Also resolved the conflict against the base (#1033's review follow-up touched the same test file); the branch is green (types / statistics / accumulation / extrinsic) and the PR shows CLEAN again.

[HanaYukii]

Rebased onto the integrated base per the plan on #1042: the PR is now a single commit (a472da8c) directly against 1012-update-to-v080 — no more stacked-base indirection. Increment verified identical to the previously reviewed content (same method as #1042: old-vs-new diff match), and re-tested against the new base including the post-#1042 work (disputes fault fix, guarantee BadErasureShards check): types / statistics / accumulation / extrinsic / work_package / safrole / recent_history / erasure_coding all green, build/vet/gofmt clean. The review follow-up (test relocations) is folded into the commit.