ChonkCraft is a playable public beta. Campaigns, skirmishes, combat, economy, construction, fog of war, computer opponents, spells, upgrades, sound, music, save/load and lockstep multiplayer are implemented. All 52 campaign missions load from an authenticated ChonkPack built from the player's original media.
Player orders retain committed native movement before their replacements start. Attack and Move replacements also wait for the native attack callback; explicit Attack keeps the clicked quarry through automatic scans. A replacement received on the firing visit cancels the old attack's unlaunched shot; projectiles already in flight still land. The dragon referees verify the replacement, subsequent movement, first-hit timing and cancellation across save/load against captured BNE cycles. The current controls include the issue-12 stall fixes, consumed and superseded Stop handling, Patrol combat and return timing, Repair approach construction, verified Attack target-retention cases and Patrol save/load state. Follow redirects and repeated clicks retain their native movement and waiting state across save/load. Melee pursuit resumes on the verified attack boundary; wall clicks and mixed-selection Return Goods/Attack Ground enforce the native player command rules. Multiplayer shares the host's speed and reports the world cycle and peer involved in a desync. They are integrated with the current fog, native walking animation, scrolling, menu rendering, graphics-memory limits and multiplayer lobby fixes.
Ships and flyers retain their idle countdown and transport idle state through save/load. Fourteen retail types complete five 800-cycle replay checks with matching movement, both random streams and subsequent commanded siege damage.
The 52-map campaign survey uses the authenticated BNE 2.02b executable and retail pack. Its semantic-v1 comparison covers cycle, synchronized RNG, player banks, unit core and coarse orders:
- All 52 maps match through cycle 400, with no execution failures.
- Through cycle 1,800, 14 maps match and 38 have later divergences, with no execution failures. Every accepted per-map prefix is preserved.
- The shared exact frontier is 403; the sum of per-map exact prefixes, capped at 1,800, is 56,468.
This comparison excludes raw sequence, extended player state, projectiles and mutable terrain. It does not establish complete mission fidelity.
A separate native player-command matrix covers 1,251 commands across all 52 maps, 139 actors and 25 unit types. Both engines accept all commands. Commanded-unit physical state matches through cycle 100 on 43 maps and through cycle 600 on three maps. Physical state means life/on-map state, tile, absolute pixel anchor and HP; it excludes other world state and raw order/sequence. A further 65 Human 1 scenarios match commanded-unit physical state in 61 cases; all observed units also match in 61. Raw order/sequence differences remain in both matrices. The maintained 121-case command gate protects every observed unit's seven physical field prefixes and all 1,374 acceptance decisions, including nine refusals. The current evidence and reproduction workflow are in the parity runbook.
All 18 playability lanes pass with their required authenticated inputs, including the 67-test control gate, 117 movement checks, 65 projectile checks, 42 clean/adverse lockstep cases and real two-process startup. The original 392-sequence freeze reproduction and the 45-type, 11,340-sequence roster complete without a stuck final withdrawal. The 168 roster sequences containing invalid submarine-versus-land attacks retain their explicit rejection classification; their final Move is still checked. Each submarine also passes 252 sequences against legal naval targets. These liveness sweeps are separate from exact native parity.
The integrated pack-plus-Opus suite contains 3,085 tests, with 90 existing specification failures and 318 skips. A release-notes test race found in the latest full run is corrected; the 49-test launcher suite and 200 repetitions pass after the correction. The data-free suite matches its 88 expected failure identities and exactly 1,419 skips, with no unexpected failures. Expected failures still execute; an inventory pass is not an all-tests-pass claim. The canonical authenticated CI job supplies the matching raw media and requires its separate 31-skip classic profile. Coverage rules are documented in the development setup and CI guide.
Explicit attacks now reach Human 14's neutral Dark Portal, including queued commands. Exorcism follows BNE's ordered undead-area scan. Slow and Haste reach the native animation timer, whose attack frames remain authoritative. Healing and Invisibility retain BNE's self-target restriction. The shared sound path applies BNE's 80 ms repeat limit, preserving volume response under coincident effects, and the zero setting emits exact silence.
Game startup now retires the previous simulation, audio and pending result. Real-window campaign checks protect repeated launch input, load-error recovery and replacement while a defeat is queued. Soundtrack routing preserves the race and screen across source changes; unidentified numbered recordings fall back to the matching archive score. Nineteen recorded music roles are checked against the BNE scene table and decoded PCM. Recorded battles advance through their race's tracks with two-second gaps; the original campaigns omit the first battle row, matching BNE's scene selection. Playback checks exercise three successive recordings for both races in the standard pack and the German reporter's supplied pack. Arbitrary unidentified numbered-disc imports remain outside that verification.
Both demolition squads accept targeted commands, approach the selected point or unit, and apply the BNE blast. Native captures cover interrupted approaches, rapid retargeting, blast damage and forest clearing. The control gate also checks Move/Stop for 52 mobile types, Attack/retarget for the 43 types with native Attack buttons, and effects for all 50 unit/ability combinations. These checks establish command liveness; complete per-unit timing remains precision work.
Workers keep their completed cargo when redirected between mines and trees. Unfinished chopping is cleared by a new harvesting job, and the gold loop retains the native depot-exit pause. Stop now ends an outbound mining trip after its committed stride; the final boarding approach retains BNE's different behavior. Fourteen native captures cover both worker races, loaded returns and blocked approaches, with continuation checks across save/load.
Rescuing a hall transfers its former owner's surviving village, including workers inside mines or depots. Farms still transfer individually. Human 8's building-first native capture and controlled retail type-flag checks establish the exception. The regression checks cover the hidden workers' ownership, sight, worker counts, later commands and saving around the handoff. Loading a campaign also clears its starting units from owner rosters, preventing a later rescue from transferring absent units or counting their workers again.
Production game updates are published from master. The
signed update catalog is
the authority for the current public version. Publication requires successful
authenticated and data-free Tests jobs on the exact revision being released.
It also verifies installation
through the production launcher against both the built catalog and the public
endpoint. The source checkpoint is retained in Git for comparison.
- Stop and point-order startup still need the wider native timing corrections. That candidate remains withheld because it regresses command-campaign comparisons; the released Attack/Move fix does not establish complete command parity.
- Saving during active combat can still change later projectile behavior and damage rolls. The lost naval/flying idle countdown is fixed; the command and cancellation referees preserve their observed handoffs and first-hit timing; they do not establish complete combat-state continuation.
- Exact everyday controls need broader coverage: group Follow redirects, queued orders, modifiers, congestion and visible-target attacks. Eight new authenticated handler pairs cover Follow and combat, but the 532-cell transaction matrix remains incomplete. Verified capture provenance is not content equality.
- Explicit Attack retains its quarry but later route refills still differ. Worker coverage must follow containment through resource return and credit; approach agreement alone does not certify the economic loop. Return Goods eligibility and the player gold-loop depot exit now agree with BNE. Six fresh 1,500-cycle cases cover repeated credit and standing or moving allies obstructing the route. Wood, oil, depleted resources and denser congestion still need the same complete-loop coverage. Loaded resource clicks and interrupted chopping now have native witnesses; complete wood-loop timing and blocked forest destinations still differ.
- Native handler probes do not establish operating-system hit-testing or actual sound playback. A distant siege-target probe requires visibility validation before it can represent an ordinary player click.
- Multiplayer flight records retain map, cycle-zero save, accepted commands, controller/race state and build identities. Legacy recordings lack some of those seals and remain diagnostics, not complete fidelity certificates.
- The reported Asahi/Wayland rendering corruption still needs target-device comparison. The multiplayer desync report needs both peers' recordings if it recurs; the speed and diagnostic fixes do not establish its root cause.
- The shared runtime's fake-clock retry test can advance before the next deadline is scheduled. Its correction must land in Seven Days to Tomorrow and be synchronized here before the release gate can rely on stable CI.
A passing playability gate and preserved campaign prefixes do not mean the whole game is an exact recreation. Further changes must preserve those gates while closing independently measured native command and lifecycle differences.