dgb(phase-b): AutoRatchet 60% tail-guard SSOT + conformance KAT

[frstrtr]

Phase-B pool/share conformance pillar — work-weighted 60% tail guard vs the frstrtr/p2pool-dgb-scrypt oracle.

What

Lift the work-weighted 60% tail-guard predicate that gates AutoRatchet VOTING -> ACTIVATED into a fenced SSOT (auto_ratchet_tail_guard.hpp) and pin it against the oracle switch rule (data.py:1399):

SWITCHED iff target_weight >= floor(total_weight * 60 / 100) over uint288 work weights.

This is the mint<->accept coupling (#288-class no-miss property): without it a 95%-by-COUNT activation can outrun the 60%-by-WORK accept gate and wedge the crossing.

Conformance finding (flagged for review)

The SSOT implements the exact canonical floor. The live inline in auto_ratchet.hpp uses the exact-rational form target*100 < total*60, which diverges from the oracle floor in exactly one measure-zero case — target == floor(total*60/100) AND total*60 % 100 != 0 — where the inline waits one extra work-quantum (negligibly STRICTER; never observed at real uint288 scale). The KAT pins both the canonical floor and that boundary divergence non-circularly via a verbatim inline replica.

Fenced

• Additive header + test only; auto_ratchet.hpp consensus path NOT rewired (delegation = byte-identity follow-on).
• Both build.yml arms + CMake allowlist updated (ci: build test_dgb_subsidy in both Linux test jobs (fix master-red NOT_BUILT) #143 NOT_BUILT trap).
• dgb-tree-local; no shared layer / other-coin tree touched.

Proof

dgb_auto_ratchet_tail_guard_test 9/9 green locally (-DCOIN_DGB=ON). GPG-signed.

No self-merge — surfacing for tap/auto-merge per fenced band.