Public API contractVerified 2026-08-13Not yet executable

Three mining labs are embedded for development acceptance but deliberately excluded from the customer catalog.

Lua v4 Development Test Workloads

Development workloads are diagnostic scripts, not customer macros. Each one tests one engine boundary with a deliberately small request. The embedded mining labs live outside the official catalog, so customers and marketplace listings cannot discover or start them through the normal script UI.

Mining labs now present

Embedded lab	Public action exercised	Reduced acceptance scope
Mining Exact Lab	hexis.mining.mine_block	Observe the exact block under the crosshair, require the current world generation and exact registry ID, then mine only that coordinate.
Mining Connected Lab	hexis.mining.mine_connected	Observe one crosshair seed and mine its face-connected component, intended for a controlled two-to-eight-block fixture.
Mining Area Lab	hexis.mining.mine_area	Observe one block ID and mine matching blocks inside a radius-4 resident scan, including disconnected matches.

All three use the held selector Pickaxe, Drill, or Axe. They request the mining, navigation, pathfinding, and world.crosshair.read capabilities. They use default confirmed handoff and default no-Ether-Transmission navigation. The bundles have no customer configuration because their only job is controlled acceptance.

These labs do not prove a production workload. In particular, they do not cover server-specific block priority maps, distant route hints, selling, inventory-full recovery, hazards, strip-mining geometry, or long-running scheduling. See Lua v4 Mining Actions for the exact public requests and result truth.

Live acceptance checklist

For each lab, record the bundle identity, loader and module versions, server, fixture coordinates, expected exact block IDs, tool display name, and terminal result. Verify:

1. a current crosshair snapshot supplies the same world_generation used by the action;
2. a wrong or changed exact ID is never mined as a substitute;
3. a reachable target uses a stable visible aim point;
4. a distant target selects and reaches one of several engine-generated vantages;
5. success appears only after the exact position receives server packet-backed block-change evidence;
6. cancellation, user stop, world change, unexpected screen, and tool change release movement, camera, attack, and hotbar ownership;
7. nonresident terrain is reported honestly instead of triggering a chunk load;
8. connected and disconnected-area behavior remain distinct.

Predictive comparison

The request dialect accepts both handoff = "confirmed" and handoff = "predictive". Today both execute confirmed behavior. A predictive request must report handoff_effective = "confirmed" and prediction_fallback = true; it must never imply that Tick Glide ran.

The eventual controlled comparison should log predicted break tick, observed server confirmation, next-target aim start, false predictions, and recovery. An early target switch must not claim the old block was mined or abandon it without typed evidence. Until those measurements exist, predictive handoff is an accepted request with an honest confirmed fallback, not an implemented performance optimization.

Other workload status

Workload	Current reduced acceptance	Important gap
Navigation	The typed navigation action exists.	A separate local Navigation Lab is not part of the embedded mining-lab slice.
Combat	Official reduced Zealot and Sven bundles exercise combat.fire and combat.engage.	They do not provide full Zealot Killer or Sven automation parity.
Farming	No v4 farming action exists.	Repeated routes, exact look policy, crop evidence, inventory recovery, and stop semantics need a bounded task action.
Fishing	Galatea exercises the specific bounded Tadpole transaction.	General bobber ownership, bite evidence, cast/reel, and recovery are not designed.
Strip mining	No action or lab exists.	Direction, tunnel dimensions, liquids, falling blocks, supports, unknown chunks, and stop policy need a controlled fixture first.

Do not use the development bundles as a hidden substitute for missing public features. When a repeated behavior needs continuous movement, camera, attack, or recovery, define one bounded task action rather than rebuilding a per-tick controller in Lua.

Clean-room mining reference

V5's nearby miner behavior and recorded distant-vein behavior remain useful comparison points, but not source for Hexis. The exact pinned GPL-3.0 references and the behavioral comparison are recorded on Lua v4 Mining Actions. The three Hexis labs contain independently designed requests and do not ship V5 code, data, hardness tables, or coordinate sets.