PR4: lgj-abi consumes the shared masking floor instead of being a second evaluator - #78

Merged

AdaWorldAPI merged 20 commits into

mainfrom

claude/clone-repositories-71a5sw

Sep 16, 2026

Commits on Sep 14, 2026

ABI minor 11: the ndarray masking facade reaches the ABI in three symbols

AdaWorldAPI
and
claude
committed
42f4a49
Java side of ABI minor 11 — UNVERIFIED: JDK 26 not available in this sandbox

AdaWorldAPI
and
claude
committed
ec5092e
Verify minor 11 Java side under JDK 26 (409/409); narrow ApiSurfaceTest's fence to project-declared members

AdaWorldAPI
and
claude
committed
5e1296f
plan: PR4 falsifier set and the two membrane rulings that gate it
claude
committed
6ec158e
board: verify the ternlog ruling's root cause against ndarray's actual exports
claude
committed
dfbdfcf
plan: OQ-5 resolved — the all-set accumulator is a prefix skip, not a missing op
claude
committed
e9466c8
plan: OQ-1 — per-pattern scratch is disqualified; `acc` can be deleted outright
claude
committed
5e94989
CLAUDE.md: the isomorphism's middle row is wrong — Valhalla and Panama are orthogonal
claude
committed
b78c4f3
lgj-abi: PR4 C1 — the frozen pre-PR4 oracle, and why mask-risc's own is not enough
claude
committed
e0338f5
plan: close OQ-1, OQ-2 and OQ-4 for PR4
claude
committed
ee6ee32
lgj-abi: PR4 C1 — the seed and matrix falsifiers, and a bound that was too weak
claude
committed
e306c77
lgj-abi: PR4 C2 — plan_eval lowers to the one evaluator instead of being a second one
claude
committed
93da288
plan: the C2 disable run, and the two arms that came back vacuous on cue
claude
committed
456cf56
lgj-abi: PR4 C3 — the allocation gate, and the warm-up that was hiding its own property
claude
committed
9058f5a
lgj-abi: PR4 C1's last falsifiers — dst reuse, and a disable that was aimed at an unreachable line
claude
committed
585f284
lgj-abi: pin plan_lower equal to quack's lowering, and two dead fixtures
claude
committed
8ad1a1b
lgj-abi: the operand is what makes the differential a test, measured
claude
committed
e9bf3aa

Navigation Menu

PR4: lgj-abi consumes the shared masking floor instead of being a second evaluator - #78

PR4: lgj-abi consumes the shared masking floor instead of being a second evaluator#78
AdaWorldAPI merged 20 commits into
mainfrom
claude/clone-repositories-71a5sw

Commits on Sep 14, 2026

ABI minor 11: the ndarray masking facade reaches the ABI in three symbols

Java side of ABI minor 11 — UNVERIFIED: JDK 26 not available in this sandbox

Verify minor 11 Java side under JDK 26 (409/409); narrow ApiSurfaceTest's fence to project-declared members

plan: PR4 falsifier set and the two membrane rulings that gate it

board: verify the ternlog ruling's root cause against ndarray's actual exports

plan: OQ-5 resolved — the all-set accumulator is a prefix skip, not a missing op

plan: OQ-1 — per-pattern scratch is disqualified; `acc` can be deleted outright

CLAUDE.md: the isomorphism's middle row is wrong — Valhalla and Panama are orthogonal

lgj-abi: PR4 C1 — the frozen pre-PR4 oracle, and why mask-risc's own is not enough

plan: close OQ-1, OQ-2 and OQ-4 for PR4

lgj-abi: PR4 C1 — the seed and matrix falsifiers, and a bound that was too weak

lgj-abi: PR4 C2 — plan_eval lowers to the one evaluator instead of being a second one

plan: the C2 disable run, and the two arms that came back vacuous on cue

lgj-abi: PR4 C3 — the allocation gate, and the warm-up that was hiding its own property

lgj-abi: PR4 C1's last falsifiers — dst reuse, and a disable that was aimed at an unreachable line

lgj-abi: pin plan_lower equal to quack's lowering, and two dead fixtures

lgj-abi: the operand is what makes the differential a test, measured

Commits on Sep 16, 2026

CLAUDE.md: the JDK is one apt source away — never skip the Java half again

knowledge + agent: the JDK acquisition ladder, and a correction to my own fix

Java surface is sql(), not the mask algebra; leak-on-throw fixes