The simulation-aliveness-3 arc: W0-W10 — a healthy steady state, answerable failures, scenes that read - #102
Merged
Merged
Conversation
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… failures) The aliveness-3 P0 wave (docs/proposals/simulation-aliveness-3-audit.md): - P0-1a: purchase baskets are per-item-optional — 'buy what's there' instead of all-or-nothing (the cream_jar bug starved live towns); a basket that buys NOTHING still fails typed. - P0-1b: purchase-sustainability validator — every food purchase query must be restocked (OAR location output) or placement-seeded; granola_bar/cream_jar joined supermarket_restock, quantities raised to demand scale. - P0-1d: unplannable consequence commits now LOG (failed: inputs_unavailable, rate-limited per person/action/24 ticks) and count toward recency. - P0-1e: the hunger→shopping producer — critical food with nothing edible in hand or at the location proposes the located shopping trip. - P0-1f: went_grocery_shopping fires from bought_groceries, not trip completion. - P1-8: the adjustMoney debit floor + running planned-spend across a commit — balances can no longer drift negative through micro-flows. - P0-4: an empty household draw announces itself in the feed. - P0-5: per-window HUD error boundaries — a crashed inspector closes with a toast instead of unmounting the whole HUD. - Staple generation raised (shelves and pantries seed at grocery scale). Tests: basket/floor/logging suite in marketEndToEnd (revert-dance verified: 3 fail with the fix stashed), the hunger→shop producer in needs.test, the sustainability-validator fixtures, atomicity tests updated to the new logging contract, fed-week recast to bounded-flux, docs + perf baselines re-pinned. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…wn race, coherent aborts) The visuals-debug pass (proposal simulation-aliveness-3, workstream W8): - P0-2.1: mid-flight re-plans never orphan cars — setVehicle despawns any live link (occupant cleared, no phantom drivers); the audit counted 148 stranded vehicles in 32 in-game days. - P0-2.2: the spawn/despawn sprite race closed for vehicles AND people — a removed entity destroys any late-attaching sprite (markRemoved + setAsset self-destroy) instead of ghosting outside the lists. - P0-2.3: coherent travel aborts — WorldAdapter.cancelTransition (optional; live-only), called by the engine whenever a pending handle's intent dies (interrupt/pause/block/finish): the body parks where it stands, a boarded person steps out visible at the car, the car despawns. LiveWorld's pump also aborts travel when a destination vanishes mid-flight. - The false-instant-arrival hole: leaving a building now CLEARS currentBuilding, so a return trip requested mid-street can no longer resolve 'arrived' while the sprite stands on the road. - Belt-and-suspenders: a minute-cadence orphan sweep in City (controlled, unlinked cars despawn; debug wander cars untouched). - W8d: TestHarness.auditSprites() — the standing sprite-vs-state invariant read (orphan/occupied-driverless/visible-indoors counters). - W8e: activity-label collision pass — identical co-located labels merge into one '×N'; distinct labels stagger vertically instead of over-inking. Tests: 4 new commute-suite cases (re-plan leak, cancelTransition, boarded abort, the cleared-currentBuilding false-arrival guard); full suite green. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…n, home seeking, soft snap) Workstream W9 (proposal simulation-aliveness-3): - P0-6 bulldoze truth: the bulldozer resolves any clicked cell to the occupying structure's ANCHOR and stamps soil across its whole footprint — off-anchor clicks used to leave ghost buildings (sprite standing, household/business already destroyed). The bulldoze cursor now shows the grass ghost preview like every other tool (toolAssets 'bulldoze'). - P1-11 physical ejection: demolition (house AND workplace) steps everyone inside — and everyone traveling TOWARD the doomed building — out onto the connected street, visible, travel coherently aborted. The hidden-homeless flow is retired: displaced people stay on the map. - P1-11 home seeking: the new looking_for_a_home ambulatory action (parity with job_hunting — street presence, bubble, asked_about_a_vacancy child), enqueued daily per homeless adult via the agenda (routineId dedup), whose looked_for_housing commit triggers the recovery attempt AT THE DOOR (tryRecoverHousehold extracted from the monthly sweep; attemptRecoveryFor). - W9c much softer snapping: buildings capture the cursor only within 1.5 tiles (was a 4-tile yank); beyond that the preview is honestly invalid. - W9d one click, one building: a successful house/work placement returns the cursor to Select (bus-synced with the toolbar); roads and the bulldozer keep drag-painting. Tests: off-anchor bulldoze full teardown, visible ejection, the seek→recover loop, capture-gate + bulldoze-anchor placement cases; eviction tests updated to the visible-homeless contract; reachability list gains the agenda producer. Full suite green (1904). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ost, migration) Workstream W1 (proposal simulation-aliveness-3 P0-3, the Manager-magnet fix): - Front-line-first hiring: JobMarket's per-workplace pick now prefers the ROLE with the most unfilled slots (fit breaks ties within a role), and Workplace.hire fills the market's CHOSEN position instead of the first-fit walk that made 14 of 24 townspeople 'Manager' in the audit. - The critical-service boost: businesses whose coverage line sits below neutral (services.json facilityBlueprints, derived daily by City) get a decisive cross-workplace score boost — the town staffs its hospital, police, fire, and sanitation before its third gym clerk. - Migration, the labor inflow: when every materialized adult is employed, ≥3 real openings remain, and a fully-vacant home stands, City draws ONE household into it per day through the normal setupHousehold path (hydration + career retcons included) with a feed announcement. Self-limiting: arrivals are unemployed, which blocks the next draw until the market absorbs them. Voluntary job switching stays a documented deferral. Tests: front-line-first + critical-boost cases in the JobMarket suite; the four migration gates (draw / locals-first / not-a-shortage / no-home) in a new execution suite. Full suite green (1910). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…money sink) Workstream W2 (proposal simulation-aliveness-3 P1-1/P1-2/P1-8): - The venue desert closed: venues.json maps restaurant, cafe, bakery, cinema, gym(+sports_complex), library, church, and salon; new located continuous repertoires with real spending (eating_out with ordered_a_meal −22 and a tip, night_at_the_cinema, working_out_at_the_gym, reading_at_the_library, attending_a_service, pampering_at_the_salon — hygiene finally has a venue — and the morning fresh_bread_run buying the bakery's real bread). catching_up_over_coffee moves to the café where it always belonged (the asset audit found it was every life's #1 social outlet AT A SUPERMARKET). All carry the 095 depressed dampener. - Opening hours (P1-2): a venue is open only while an employee is ON SHIFT (derived from authored shifts; no new data) — hasVenue and venue resolution skip dark shops, ending the audit's 2 AM shopping trips. Live-only; bootstrap venues stay abstract (the sanctioned seam). - Routines: gym_habit, weekly_service, dinner_out (money-gated) — the mid-frequency layer that makes venue life habitual, plus the money sink live play lacked. - Fixed in passing: a latent LP-11/121 streamed-asset divergence — the generator's outcome-pass appends (adoptions, counterparts, cohabitation) drained to disk minute-less when a flush boundary landed between; both paths now stamp after the outcome invokes (batch-insensitive by the (person, tick) group rule). Tests: opening-hours cases (unstaffed = closed, off-shift = closed, on shift opens); venue fixtures staff their shops; the fed-week meal filter counts restaurant meals; reactions fixture pins co-location. Docs + perf baselines re-pinned. Full suite green (1912). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…he aftershock) Workstream W3 (proposal simulation-aliveness-3 P1-3): - Natural conclusions: interrupting an instance that already ran its FULL duration records a COMPLETION (onComplete fires) — the audit's every-morning 'sleep interrupted' (26 starts / 27 interrupted / 0 completed) was the wake-up obligation winning arbitration in the very tick sleep would have completed; biographies read as chronic sleep disruption. Mid-duration displacement stays an interruption. - Work→work HANDOVERS: a same-shift task switch (the doctor turning from her charts to a patient) no longer fires stopped_working — the Brain flags work→work replacements and the engine suppresses the lifecycle events while still logging the instance transition. Real exits (shift end, sickness, leisure displacement) fire it as before, so the SkillProgression day credit keeps its once-per-day close. - The officer's side of the arrest: the arrest ceremony now interrupts the arresting officer's NON-work activity first (the audit watched an arrest land mid-coffee-date without the date pausing); on-duty work keeps running — arresting IS the job. - The aftershock (P1-3c): a fresh high-shock event (was_arrested, got_caught, lost_home_to_fire, escaped_a_fire — json/arbitration.json, validated) dampens OUTGOING social/leisure free-time weights ×0.25 for 12 ticks; home/maintenance keep theirs, so the bias reads as lying low, not paralysis. Tests: full-duration-completes + mid-duration-interrupts, the handover suppression vs the real clock-out, and the aftershock cohort comparison. Full suite green (1916). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…tion, the wedding) Workstream W4 (proposal simulation-aliveness-3 P1-4/P1-5/P1-6/P1-10): - Household-unit rehousing (P1-4): displacement offers the WHOLE living group to a relative with room for everyone before any per-member split — the audit's fire rehoused the wife with kin and left her partner homeless at the supermarket. Fallback stays per-member when no host fits them all. - Retirement is real (P1-5): job_hunting (the action AND the routine) gates on retired == false and age < 68 — the audit watched a seriously-ill 75-year-old get hired off a morning application. Retirees gravitate to the park/hobby/walks (retired ×1.6 selection boosts). - Conception rides intimacy (P1-6): a had_sex commit rolls a seeded 10% conception chance and INVOKES the real pregnancy event on the would-be mother — whose own eligibility (recent intimacy, married, wantsMoreChildren, not pregnant) keeps the last word. Shared by City (live) and LogicalWorld (off-map): one salt, identical both sides. The background probabilistic hazard demotes 0.6 → 0.15. - The wedding is a family scene: partnershipFormed fans out a new attended_a_wedding milestone (+2 valence) to both spouses' close kin. - moved_out_of_parents cooldown 60d → 2y (P1-10): the jail-cycle replay that produced 8 move-outs in one asset life reads absurd no more. Tests: the never-split couple rehousing case; the conception suite (from either subject, eligibility last word, pinned chance). Perf baselines re-pinned (conception rolls joined the spine). Full suite green (1920). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…rmations) Workstream W5 (proposal simulation-aliveness-3 P1-7 / Part 5.3): - Curb-bag piles: the town's uncollected curb bags finally RENDER — small brown mounds at the street side of occupied homes, growing with neglect. The sim's curb pool is one shared 'outside' location (the 112 collection contract), so per-house attribution is a documented display approximation (round-robin across occupied homes). Squalor 1.0 looks like squalor instead of a pristine street with a nagbar. - Side-by-side formations: co-located people running the SAME activity — a couple's walk, coworkers, and the chase pair (fleeing↔chasing alias as one scene) — draw with small lateral offsets by stable slot index. Render-layer only: logical positions stay canonical; refreshed per in-game minute, applied in the person redraw closure. - The squalor→cleaning selection loop was verified already bound (LP-8 modifiers present on the cleaning set) — no change needed. Scene-layer changes (excluded from the Jest coverage gate per the repo convention; the Playwright browser suite covers the render glue). Full suite green. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…Event, safe selection) Workstream W7 (proposal simulation-aliveness-3 Part 5.6): the seams every observation/scenario session monkey-patched by hand, promoted to the TestHarness API: - build(tool, row, col, blueprintKey?) — pinned civic placement through the same tileClicked path a construction-menu pick takes (placeholder texture included; the audit session's black-square foot-gun closed). - hireAs(personId, workplaceKey, title) — a REAL hire into a chosen position (WorkLife set, slot consumed) for forced-staffing scenarios. - forceEvent(eventId, personId, params?) — EventEngine.invoke with the live state/clock plumbed; one call to make someone fall ill or adopt. - selectPerson(personId) — the payload-safe PersonSelected dispatch (a hand-built payload once unmounted the whole HUD). - Workplace.avaiableJobs → availableJobs (the Part-5.7 grep-hostile typo). The P1-9 live-map school-attendance keystone stays deferred to W10's speed-invariance suite (arrival-rate assertions belong with the distortion-free time contract they measure under). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…, distortion-free) Workstream W10 (proposal simulation-aliveness-3), the arc closer: - One authoritative frame transform: effectiveFrameDelta(raw, scale) — hitch-capped (MAX_FRAME_DELTA_MS=100: a 5s hang is lost wall time, never a sim leap) and pause-zeroed. GameManager.advanceTime AND Field.update both consume it, so the clock and movement can never diverge, and a stall stalls everything together. - The shipped ladder is pause(0)/1x/4x/8x (was the debug 1x/4x/16x). The setTimeScale/timeScaleChanged bus pair keeps every setter (HUD, debug T key) and every surface (toolbar highlight, debug overlay, fire particle emitters — which now speed up and freeze with the world) in sync. - The time HUD: a floating top-right toolbar — Pause / Play / double- / triple-chevron — active speed highlighted; talks to the game only through the bus (§4.9). No masterSwitch gate: this is a player control. - Tests: the effectiveFrameDelta contract (scale/cap/pause), ladder cycling, schedule-equivalence at the util level, AND the composed contract — equal SIM time at 1x/4x/8x lands a walking commuter at the IDENTICAL position with the IDENTICAL clock; a 5s hitch advances both by exactly the cap; pause moves nothing. The LP-2 field test migrated to the new one-authority stub. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…g, version bump) Workstream W6 (proposal simulation-aliveness-3 Part 4): - The abstract-encounter chase discount (Part 4.2): off-map co-location is a shared abstract venue, so officer-meets-suspect fired constantly and the decoded asset showed 12 catches in 15 petty crimes — 11 jail terms in 27 years, ate_prison_food ×678. LogicalWorld's chase resolution keeps City's catch formula but discounts the final chance ×0.4, restoring live-like CUMULATIVE odds (impunity and recidivism get room to breathe). - AskFirst social pricing (Part 4.4): the social hook now prices intimacy proposals by the SAME standing the target's consent policy reads — strong/kin bonds full weight, strangers ×~0.2, rivals ×0.05. The asset's ~39% lifetime hug-decline rate (1,142 rejected hugs across 30 lifetimes) came from under-priced proposals; the frequency-band test now pins the new regime (strangers barely ask; casual no-consent socials keep the cadence). - Part 4.1/4.3 (cooking:eating, illness cadence) ride the W0/W3/W4 fixes organically — the regeneration re-measures them. - HISTORY_GENERATOR_VERSION 121.0 → 136.0 (the arc's stream shifts). Full suite green (1925); perf op-count baselines re-pinned. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…proposal status v2 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…d-venue choke guard Found by running the arc against the audit's exact civic town in the browser (the fed-town keystone, live): - The basket now spans the FULL grocery restock set (milk, pasta, cheese, butter, cereal joined) — the shelf drained of exactly what the basket asked for while the rest silted (13 butter sticks and 14 cereal boxes nobody could buy). New INVERSE validator rule: every restocked food must be purchasable by some purchase query — which immediately caught the bakery's cake and the restaurant's grilled_steak too (production nobody could buy, the 12k-dough problem's descendants): fresh_bread_run gains bought_a_cake, and ordered_a_meal now consumes the restaurant's REAL steak stock alongside the service charge. - milk/cereal are honestly eatable (ate_breakfast_from_cereal + ate_meal_from_milk OAR alternatives; the Brain meal sets know them) — no granola-bar-deadlock class regression. - The closed-venue guard at the arbitration CHOKE POINT: planner routines (weekly shopping, dinner out) proposed straight into shops that had just closed — a 166-blocked-trip wall at 16:00-19:00, right after the clerk clocks off. One guard in resolveIntents covers every intent source. Live re-verification: orphan cars 0 across the whole run (was ~4.6/day), the town staffed itself (1 Manager vs the audit's 14; nurses, officers, a firefighter, collectors), bags collected, purchases flowing. Full suite green; perf baselines re-pinned. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…the menu config The textures are generated per menu ENTRY id (civic_landfill), not per blueprint key (sanitation_depot) — harness placements of the landfill and prison rendered as missing-texture black squares. Verified live: the time toolbar's full bus loop (click → setTimeScale → timeScaleChanged → highlight), the ×2 label merge, and empty midnight streets all confirmed in the browser. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… holds WHILE it runs Aliveness-3 follow-up (maintainer observation): the JSON location field was only enforced at start — a running location-gated instance whose person got physically displaced (the W9 ejection class, off-map detention relocations) kept 'running' wherever they stood. The advance loop now re-validates the concrete location kinds (home / building:<key> / person-resolved) every tick: a displaced instance reverts to PENDING and re-materializes (walks back and continues) or gets replaced at the flip. Structurally, 'Sleeping' can never run on the street again. (The observed street labels were the '→ Sleeping' TRAVEL labels — heading somewhere to sleep — but the gate makes the violation class impossible rather than merely unobserved; a live probe over a running town found zero standing violations.) venue:* stays ungated here (it resolves to a concrete host at materialization; the handle is gone by running time), as does 'outside'. Test: displacement mid-nap bounces the instance out of running and it resumes honestly after returning home — revert-dance verified (fails with the gate stashed). Full suite green (1926). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Two maintainer reads from the live session: - '→ Sleeping' over a street walker read as street-sleeping — travel labels are now destination-first: '→ home: Sleeping', '→ Oliveira-Saraiva: Treating patients', '→ the supermarket: Going on a shopping trip'. The destination resolves from the instance's own location requirement (home / venue kind / hosting business name / a visit). - Pedestrians all walk the EXACT curb polyline, so co-located walkers rendered perfectly stacked regardless of activity. A stable per-person render offset (±4px, hashed from the id — people using different parts of the sidewalk) separates everyone; the W5 formation offsets stack on top for same-activity groups. Render-layer only; logical positions and every stream untouched (no regeneration impact). Verified on screenshot: three hospital staff converging by name with staggered labels, the collectors' '×2' merge, visibly-separated sprites. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…lding Aliveness-3 follow-up (maintainer read): the per-frame redraw closure hid the sprite on entry but returned early WITHOUT touching the bubble (or the pet dot) — the label stood at the door until the next per-minute refresh, a ghost over every entrance during arrival waves. Entering now hides sprite + bubble + pet dot frame-accurately. Live probe across an arrival wave (8 samples, up to 6 visible bubbles): zero visible-bubble-while-indoors ghosts. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Maintainer spec: line one says where ('Going home', 'Going to Moraes
S.A.'), line two what for ('Plan: Spending time at home'), centered, no
arrow. Bubble labels also strip unresolved {param} template segments (the
LP-14 resolver lives in the inspector; the destination line carries the
entity here) — 'Plan: Applying for a job', never '…at {employer}'.
Verified live at the evening rush: 'Going home / Plan: Spending time at
home ×2' (the merge suffix rides line two), job seekers converging on
named businesses.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…amping, wander stall Live-found (fresh-game observation, real browser): people at home popped visible "on top of their house" when their activity changed, and outdoor actions left sprites stuck at houses. Four root causes, all in the physical layer: 1. Placement never set currentBuilding — materialization, newborns, recovery movers and loads left everyone "indoors with a null link", so the first located action ghost-commuted to the house the person was already inside (sprite pops out, walks to a car, drives home), and LiveWorld's identity-only arrival check could deadlock located actions forever (a pending 'home' handle 12 sim-hours old on a man standing in his own living room, sleep waiting all night). Placement sites now set the link; LiveWorld's arrival check falls back to physical containment (indoors + the tile under the pixel references the destination) and heals the link; SaveManager re-derives it on load. 2. walk() steps were unclamped — speed × timeDelta overshoots the <1px arrival window at large frame deltas (4×/8× throttle, harness stepping, hitch cap) and the walker ping-pongs across the target forever (observed: frozen 1px from the commute car for 21 sim-hours). Steps now clamp to the remaining distance. 3. updateDestination committed the destination before the reachability check — one unreachable (or own-building) pick froze an ambulatory walker for the rest of the session. The pick only commits with a real path; failures retry. 4. Outside transitions parked the body at the entrance pixel (inside the footprint — read as "standing on the house sprite"). stepOutside() now lands on the curb of the connected street, entrance fallback in road-less worlds. Verified live: max pending handle age 22 -> 4 ticks, frozen walkers 269 -> 0, stuck sleeps 4 -> 0 (8 running at 06:00), sprite invariants all zero. Generator untouched (live/physical layer only — the running regeneration stays valid). Revert-dance: stashing the fixes fails exactly the 5 new tests. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…stat (v137.0) Mid-regeneration the living count compounded past target (335 at year 30 → 490 at ~year 90 against a 250 target): the W4 conception channel (had_sex commit → 10% pregnancy invoke) is the dominant conception path and manual invokes bypass setProbabilityScale — the thermostat was braking only the demoted background hazard. maybeConceive now multiplies its chance by the engine's scale for 'pregnancy' (a new getProbabilityScaleFor read); live play never installs a scale, so live behavior is untouched. Generator stream changes → HISTORY_GENERATOR_VERSION 137.0; the in-flight regeneration is restarted. Also folds the doc addendum lint fix (MD004: a continuation line starting with '+' parsed as a list marker). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ew node's default The v137.0 regeneration OOM'd at recording year 96 (six years into the 10-year hourly hot band) at node's ~4GB default old-space. The hot band holds up to flushIntervalYears (5) of HOURLY action logs for ~250-270 people between drains, and the aliveness-2/3 arcs raised per-tick log volume (work wiring, food chain, venue repertoires, minute stamps) past the default's headroom — no aliveness-3 regeneration had ever completed. cross-env pins NODE_OPTIONS in the script so the requirement is documented and repeatable (machine needs ~24GB+ RAM; the dev box has 32GB). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ines (v137.0) The arc's closing evidence run: 100 recorded years (90 daily + 10 hourly), 2h22m, 704 retained people, 1.59 GB. Population holds 207-294 around the 250 target across all 120 simulated years — the thermostat brake works through the W4 conception channel now (the pre-fix run compounded to 490). Decode re-measures vs the audit Appendix baselines: - cooking:eating ~14:1 → ~2:1 (p600: 4,125 cookings / 1,782 meals), with typed ate_a_meal:failed entries where scarcity bites instead of pantomime cooking - sleep is real in the hot band (8h ×2,552 vs 24h ×277) and COMPLETES (2,829 completed vs 1,219 interrupted; baseline: zero completions ever) - justice: the median profile is 9 shopliftings → 2 caught → 1 jail term over 27.5y (baseline: 15 → 12 caught → 11 terms) — the abstract-encounter discount works; a habitual-criminal tail case remains (p650: 23 terms/19.3y, the crime-habit escalation loop) — flagged as a post-arc balance item - hug-ask volume collapsed ~12× (2,946 → ~240 attempts/lifetime — the standing pricing works); decline share among remaining asks is largely unchanged - the W2 venue repertoires are lived off-map: library/salon/cinema/gym/church all appear in every decoded life; illness ~2.5/yr (was 3-4) Boot-verified live: asset draws hydrate (158k-181k pre-game entries/person) and drawn people arrive with the placement link set. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The aliveness-3 player-experience arc (W0–W10), one PR per the maintainer's bundling rule. Proposal + findings + re-measure evidence:
docs/proposals/simulation-aliveness-3-audit.md; engine narrative:CLAUDE.md§4.17. Theme: a healthy steady state, answerable failures, and scenes that read — the month-long observed town was a chronic-famine, garbage-saturated, mood-collapsed place nobody could fix; this arc closes that loop.The workstreams (sequential commits)
inputs_unavailable), the hunger→shopping producer, the adjustMoney debit floor with running planned-spend, empty-draw announcements, per-window HUD error boundaries.cancelTransition,abortTravel, thesetVehicleguard — the audit counted 148 orphaned cars in 32 days), the spawn/despawn sprite race closed, leaving a building clearscurrentBuilding, label merge ×N + stagger,auditSprites()as the standing invariant read.looking_for_a_homeambulatory wired to recovery-at-the-door, 1.5-tile snap capture, tool reset to Select after building placement.had_sex(background hazard demoted), wedding kin fan-out, move-out cooldown 2y.build(blueprintKey),hireAs,forceEvent,selectPerson;avaiableJobs→availableJobs.timeConsistency.test.ts.generate-history(the hot band OOM'd node's 4GB default — no aliveness-3 regeneration had ever completed).The post-arc live-found wave (fresh-game verification)
A maintainer fresh-game session surfaced sprites popping visible "on the house" and sprites stuck at houses during outdoor actions. Root-caused live and fixed (audit doc Appendix B.1): placement never set
currentBuilding(ghost commutes to the house you're already inside + located-action deadlocks — healed via a physical-containment arrival fallback, set at all placement sites, re-derived on load), unclampedwalk()steps (overshoot-oscillation at 4×/8× frame deltas — a walker frozen 1px from his car for 21 sim-hours), the wander commit-before-path-check freeze, and doorstep parking (stepOutside()lands on the curb now).Evidence
Flagged follow-ups (documented in the audit doc): the location-less-actions data pass (people idle wherever the body is stranded — needs a
home-location pass, deferred because it perturbs generator streams), and the habitual-criminal tail (one decoded life with 23 jail terms — the crime-habit escalation loop wants a cooling tune).🤖 Generated with Claude Code