Skip to content

[G4 P0] Decode exact native Anvil block consequence independently (CPU source gate) - #471

Merged
mark-e-deyoung merged 5 commits into
experiment/gha-kvm-system-rdtefrom
candidate/g4-anvil-server-owned-source-20261015
Oct 15, 2026
Merged

mark-e-deyoung merged 5 commits into
experiment/gha-kvm-system-rdtefrom
candidate/g4-anvil-server-owned-source-20261015

Conversation

@mark-e-deyoung

Copy link
Copy Markdown
Contributor

Authority

Agent Dispatch private #627. Accepted OP4H-015 real same-native-App offline Mineflayer client visibility remains client-only; backend_independent_world_oracle_pass=false until live independent server-owned region readback.

Change

Add dependency-free, read-only, bounded MCA/NBT decoder for fixed source-controlled r.0.0.mca chunk (0,0) y70 blocks at (0,70,0) redstone, (2,70,0) iron-bars→air, (1,70,1) black-concrete→sea-lantern candidate vs rollback. Strict sector/compressed length/cross-chunk/section/palette index checks, 64 MiB region + 16 MiB chunk bounds, no raw region/world contents in published receipts. Synthetic adversarial candidate/rollback, stale core, absent effect, foreign chunk, corrupt/external compressed format, bad stage/replay negatives. Add to already existing public CPU-only native Mineflayer qualification job.

Precise limitation / next gate

This is source parser only, and server_origin_authenticated=false, world_quiescence_observed=false, backend_independent_world_oracle_pass=false. Caller-supplied region bytes do NOT become native evidence. Next bounded patch must consume actual guest middleware TLS-authenticated core.download/filesystem.get from exact owned WORLD/region/r.0.0.mca only at two app.stop-verified, flush/quiesced points after client action candidate and after client action postrollback, plus a pre-action clean backend fixture baseline, tie to same request/App/owner/source and cleanup. Keep all bytes locally, publish only fixed triples/digests, verify exact protocol/client causality separately. Then independently source-sign and admit a new one-time 016+ public KVM only with empty queue; no native run in this PR, no old receipt promotion, no operations auth change, no RCON, no production scope, physical APPLY HOLD.

Do not merge overall experiment PR #39 or replay requests 011–015.

@mark-e-deyoung
mark-e-deyoung marked this pull request as ready for review October 15, 2026 04:39
@mark-e-deyoung
mark-e-deyoung merged commit c84ab5c into experiment/gha-kvm-system-rdte Oct 15, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant