last-hearth/docs/m4-verification.md

M4 implementation and verification — The Last Hearth

Updated: 2026-11-02. Scope: renewable ecology, food preservation, pottery, small-tier metalworking and bounded offline accounting. This record separates automated evidence from the required group playtest. It does not claim public service readiness or human acceptance.

Offline contract

System	Rule
Player needs	Pause while the player is logged out.
Hearth fuel	Uses absolute UTC epoch deadlines from os.time(), not preview/solar time. Sticks add 45 s, logs 180 s; stored reserve is capped at 600 s. Fuel may expire during absence.
Hearth jobs	Cooking 6 s / 20 reserved fuel seconds; boiling 4 s / 10; smelting 60 s / 30; charcoal production 60 s / 30; pottery firing 120 s / 30; drying 300 s / 30. Inputs are consumed only after a valid job and timer are established. A completed result stays queued until an authorized player collects it. Timer and load reconciliation are idempotent.
Charcoal	One log plus clay forms a charcoal charge. The process consumes 60 s plus its 30 s reserved cost and yields charcoal worth 90 s; remaining heat plus output cannot exceed the source log's 180 s.
Crops	Planting time is stored as a decimal epoch string. Growth replays local historical solar temperature in six-hour samples, pauses below 3°C, runs at half rate below 10°C and caps replay at 30 days. Stages are derived from planting time, not timer count.
Berries	Harvested bushes use one absolute 24-hour regrowth deadline; no repeated yield while bare.
Saplings	Planted trees wait three real days, then grow only if a bounded 5×6×5 volume is loaded, clear and allowed by claim/protection rules. No placement beyond map bounds.
Wildlife	Hares are passive, non-dropping, nonpersistent, have a 900-second loaded lifetime, and are limited to two near a player and 24 globally.
Food age	General spoilage is disabled. Dried rations are shelf-stable. No hidden or irreversible item-age simulation exists.

Automated verification

• Verified technical contract: Luanti 5.18.0, game 0.5.0-m4, save schema 1. The automated technical gates listed below have passed; human M4 acceptance remains open.
• Engine: run scripts/run.mjs test using the pinned Luanti 5.18.0 runtime. It creates fresh disposable worlds; it must never access the persistent play world.
• Map seeds: run scripts/run.mjs test:seeds after ecology/mapgen changes.
• Node tooling/assets: run node --test tests/tooling.test.mjs tests/assets.test.mjs.
• Coverage: crop historic temperature, catch-up cap, stage derivation, invalid timestamps; renewal and inventory conservation; charcoal/smelting/drying/pottery jobs; output collection, offline completion and repeated reconciliation; wildlife persistence and population bounds; registration and real-engine craft smoke.
• Results: Luanti 5.18.0, game 0.5.0-m4: final scripts/run.mjs test passed check, native smoke and all 21 isolated suites, including six ecology tests, fourteen process tests and three wildlife tests. Eight fixed-seed runs then passed safe arrival and required resource checks after tree density was raised to address a no-log seed. The final full engine run passed on the updated tree. Node tooling/assets tests passed. A server warning reported maximum lag peak around 1.23 seconds during the bounded smoke workload; this is not a client-FPS or performance certification. No human acceptance is inferred from automated tests.

M4 group acceptance remains required

Have a small group play several sessions across at least one real week in a disposable/staging world. Record returns after absences, crop and regrowth outcomes, resource availability near arrival, pantry usefulness, process completion and whether the group chooses new goals without daily chores. Compare loaded and unloaded outcomes over equal elapsed time. Check harvesting under claim sharing, full inventories and ordinary latency. Preserve test world identity, exact release, actions, and reproducible defects. Do not enable general spoilage as part of this test.

M1 voluntary repeat-play/performance, M2 unfamiliar-first-night, M3 two-player cabin/permissions, and operator approval of shared climate coordinates remain separate open gates. M4 technical completion permits M5 engineering to proceed by explicit user direction; it does not mark M4 human-accepted or waive earlier gates.

Botanical journal and ecosystem survey — pending main validation

2026-11-02 implementation addendum. The preceding evidence is retained unchanged and describes its earlier tested tree. This newer ecosystem/art/UI increment does not inherit that full-engine certification and does not bump the game version or save schema. See living ecosystems and art direction.

Implemented in the owned interface and focused recipe/journal harness:

• Shared eom.dev Mocha base/mantle/raised/border and text/muted anchors; restrained mauve, teal and yellow accents, pressed-sprig dividers, consistent buttons, hover/pressed states, fields, inventory lists and tooltips across guide, recipe, inventory, location and survey forms. Seven guide pages stay inside 14×10; Living ecosystems replaces the optional Riverward goal without removing fishing recipes/process discoverability or changing fishing nodes/mechanics.
• Read-only /ecosystem, guide button and refresh: one call to the actual hearth.succession.habitat(player:get_pos()) Lua API plus one hearth.wildlife.status() snapshot. Plant counts are a local 5×5×5 natural-origin sample, with fern/tree cover combined; incomplete/unloaded cells are disclosed. Fauna counts are global active hare/deer/fox totals, explicitly not a local census. No scheduler, forage, deposit, metadata write, entity scan/spawn, emergence or force-load is invoked by the survey.
• Current wildlife status reports global cap 32, local cap 6 and per-player cap 6; this supersedes the older hare-only contract above for the working ecosystem implementation, not for that historic verification run. Wild succession is loaded-only without offline replay; crop/berry/planted-sapling clocks remain separate. Dynamic Life is inspiration only; implementation and art are original.
• HUD panel with labeled percentages, color-coded reserve tracks and low-reserve warnings; established title, clock, needs, context, help session keys remain. The authored hotbar slot is repeated using [combine:320x32 at ten 32-pixel offsets (no leading parenthesis), with hearth_selected.png. Image setters are optional for older mocks. Standard crosshair texture override is part of the main generated-asset integration, outside the owned files here.
• Inventory craft/result/main actions and listrings are unchanged. Location writes remain singleplayer-only with finite-coordinate validation and 16-character input bounds. Recipe searches remain literal, escaped and 48-character bounded. No credentials, cryptography, certificates or network access are introduced; CodeGuard secret avoidance and bounded/escaped input handling were applied.

Focused validation actually run: the existing Fengari runner under Node 24.22.0 completed all 13 recipe/journal tests, including survey read-only wiring, escaped/bounded fields, guide bounds, inventory actions, graphical needs and exact ten-slot combine syntax. It also completed 27 succession tests, including the real habitat API's read-only/loaded contract. Both emitted their expected completion markers, exit 0 and empty stderr. Earlier focused attempts exposed two harness issues (a missing heat fixture value and an escaped-comma expectation); both were corrected before this successful journal run.

A related standalone wildlife-suite invocation was attempted, but the terminal did not return a confirmed completion result; it is not counted as passing here. Editor diagnostics reported no errors in the five owned files. No main integrated/full-engine tests or native rendering checks were run by this UI task. Generated-asset/native client rendering, UI scaling, contrast and main validation remain pending. M1–M4 human gates, long-session ecological balance and operator acceptance remain open.

Integrated ecosystem, art and seasons verification

2026-11-02 final integration addendum (supersedes the pending automated status above, not the historical evidence or human gates):

• Pinned Luanti 5.18.0: final disposable-world check/smoke passed; 52 Lua files compiled and all 23 isolated suites / 377 checks completed. Includes nine seasonal, 36 succession, 18 wildlife, 31 content and 13 journal/recipe checks.
• Final eight fixed-seed run completed all eight individual success markers and the aggregate seed PASS marker for starting resources and safe arrival.
• All 48 Node tests passed, covering tooling, original art, models, seasonal atlases, deterministic regeneration and package integrity. Optional full audio codec regeneration was not configured; framing/hash/CRC checks still passed.
• Earlier validation attempts encountered a seed timeout and socket-bind failure; these are not counted as passes. Subsequent complete runs exited zero.
• Generated pixel-art contact sheet inspected. In-game meshes, materials, alpha, seasonal transitions and HUD scaling have not received native visual or human acceptance. Headless checks cannot certify appearance, enjoyment or FPS.
• Seasonal leaf loss also reduces simulated canopy shade; dormant plants preserve ecological age, and seasonal food patches preserve their harvest cooldown.
• Game contract remains 0.5.0-m4 / save schema 1; user-undone release changes were not reinstated. No persistent play world was opened or regenerated for tests.

Native visual review, long-session population balance, multiplayer protection and the existing human/operator gates remain open.