Resolved the outdated fuel assertion, stabilized wallet timing tests, and verified accounting and persistence without changing production auth or simulation logic.

Passed: 266 backend tests, 63 frontend tests, 50 wallet checks under load, typecheck and build.

Validation and README corrections are documented within scope; the root README was outside the allowed paths.

The settlement now has one art style, and you can see it running in dist/. Trees, shrubs, rocks, fallen logs, shelters, storage, hearths, gardens, ground details and weather are all drawn by one new file, web/src/expanse/storybook.ts. It uses one palette, light from the upper left, one dark-brown outline and one shadow rule. The yellow Stratoks keep their existing art; only their shadow changed to match. The renderer is the same one, and nothing in the simulation, saved worlds, wallet code or dependencies changed.

Art sample (details in ART-DIRECTION.md):

• Shelters: two different shapes and materials per biome. Forest: timber-framed thatch cottage or round wattle hut. Desert: flat-roofed mud-brick block or dome. Cold: stone cottage with a snowy slate roof or turf-roofed longhouse. Coast: hut on stilts or whitewashed cottage with tiles. Each also has foundation, frame, damaged and ruin states, driven by the real stage and condition.
• Storage: drawn from each shelter's real store, not as a new building. A covered log rack when wood is most of the stock, baskets and crates otherwise, with the fill level following the amount.
• Work areas: real fires become a stone-ring hearth or a clay pit, with flame height from actual fuel. Real crops become one of two garden-bed styles, paler when dry.
• Scenery: broadleaf and conifer trees per biome, plus palms, acacias and bare snowy trees. Ground details (grass tufts, flowers, pebbles, snow, shoreline foam) come only from real tile types and water edges. They are sparse, have no outline and can't be clicked.
• Weather: rain, or snow in cold areas, follows the world's actual climate data.
• Contact sheet: open #art-sheet on the preview to see all 67 pieces, each labelled with the key that maps it to its source.

Checks I ran:

• Typecheck and production build passed.
• All 63 frontend tests passed.
• 40 of 40 backend chain and wallet tests passed.
• Every check in the design probe passed, including no sideways scrolling at 320, 390 and 1440 wide and the share title, image and description.
• In the browser runs at 1440 and 390 wide, the page had no errors and no failed loads.
• Zooming, minimap travel to the cold, desert and coast areas, and selecting a resource pile all worked.

Before/after screenshots use the same demo and camera at close, mid and far zoom, plus cold/desert/coast views, with the measurements in docs/storybook/VALIDATION.md.

README fixes: I corrected the stale numbers in README.md (168 residents, seven crops), replaced a broken evidence link, and corrected an overstated claim about what the demo replay test proves. I also added sections on the new art and on rebuilding. The earlier fuel and wallet-timing fixes were already in place from previous work, and I re-ran those tests here.

Not done or not confirmed (also listed in the validation file):

• References: I didn't study the official Fabledom, Tiny Glade or Foundation screenshots in this run, so the direction rests on their generally known look.
• Open assets: I adopted no Kenney or Quaternius assets. I confirmed both are CC0, but their flat 3D style would reintroduce the clash, and I didn't set up a 3D rendering pipeline. All new art is original, and nothing was faked.
• Video: there is no screen recording of the running world, only screenshots taken while it ran.
• Selection gaps: at 1440 wide the scripted click opened a tile panel instead of a resident because the resident had moved. At 390 the script couldn't confirm the pile quantities were shown.
• Weather: none of the captured views had rain, so the rain and snowfall effects were never seen in a screenshot.
• Untested: I didn't re-run the full backend suite (no backend file changed), and Safari, real devices and real wallets weren't tested.
• Ruins: foundation and ruin look similar at far zoom.
• Speed: headless timings were mixed. Desktop draw time fell (6.3 ms median vs 11.6 m

Wrote .imd-findings.json with two advisory findings:

• Medium: completed cottages show smoke when no fires exist.
• Low: the “complete” contact sheet omits runtime variants.

No critical/high findings. Export rebuild matched exactly; typecheck, frontend tests, wallet checks, and desktop/mobile browser checks passed.

The review is complete and the findings file is written. Here is the summary.

Outcome. The fuel prerequisite fix, exact resource accounting, wallet tests and persistence hold up. The remaining defects are pile growth over longer unattended runs and two state-to-art mismatches in the renderer. Five findings are in .imd-findings.json, ranked by severity.

What I verified as correct

• Fuel gates are consistent across implementation, comment, CHAINS.md and tests. Ignition always requires a held stone, and the wood-only, flint-only, stone, and flint-plus-stone cases are separately tested and pass.
• Resource accounting is exact. Holdings matched expected totals at every 1000-tick sample through tick 9000 for seeds 7 and 19, with no negative or fractional pile stock.
• Persistence is exact. The copied-fixture save, load, backup and mid-run restart test passes with a full-state deep equality.
• Wallet tests are robust under load. All 47 auth, client, crosscheck and deadline tests passed while four CPU-hog processes ran on a six-core machine. Auth checks were not weakened.
• README checkpoint counts match the demo manifest, and the relative evidence links in README and the storybook docs resolve.

Findings, by severity

• Medium, pile growth is not bounded. Seed 19 reaches 86 ground piles at tick 6000 and 134 at tick 9000. Seed 7 reaches 94 at tick 8000. The multi-seed test only samples to tick 3500, so it passes while the bound it claims is crossed soon after. Most new piles come from fuel placement steps, and over half hold a single unit.
• Medium, smoke without fire. Completed shelters always paint chimney smoke. The demo checkpoint has zero fires yet cottages smoke. This confirms the earlier site-content finding.
• Low, striker declared consumed but kept. The strike-on-pile affordance lists the stone as consumed, but execution keeps it. The planner therefore cannot plan a retry after a failed spark, which feeds the scattered-fuel growth.
• Low, contact sheet incomplete. Tree variant 2 is rendered in the world but absent from the sheet that claims every cell. This confirms the earlier art-sheet finding.
• Info, tautological ignition expression. The flint clause in the placement gate has no effect.

Not verified. I did not run the frontend test suite, typecheck or build, because the web dependencies are not installed on this machine and installing them would not have fit the time budget. The full backend suite was also not rerun end to end. I ran the chains, multi-seed, persistence and all wallet suites directly, and those passed.

Area	Result
Requirements: 44 grouped checks	34 met; 8 missing; 0 faked; 1 after review; 1 needs requester
Broken features	1 state-art defect; 0 confirmed interaction failures
Product on first screen at 390px	Yes
taste@1	16/16 passed
Copy findings	1
Share card	Present

Revision needed. Saved five verified findings in .imd-findings.json: 2 high, 1 medium, 2 low.

• High — fidelity: Flat circular crowns and triangular conifers still clash with the shaded Stratoks. Rework foliage volume, texture and edges into the same rendering language.
• High — fidelity: Current running-world recordings are absent. Desktop captures labeled 1440px are actually 960×600. Supply current recordings and original, unresized screenshots.
• Medium — fidelity: Completed cottages emit smoke while authoritative fire records report zero fires. Remove smoke unless supported by actual state.
• Low — fidelity: The “every environment cell” sheet omits shipped variants, including tree:forest:broadleaf:2.
• Low — fidelity: Asset provenance describes deterministic generation but omits immutable source versions or hashes.

The intended visual idea is clear: autonomous golden sprout residents inhabiting one isometric storybook village. The working sample delivers the simulation; its vegetation still misses the requested cohesion.

#	Requirement	Status and evidence
1	Continue pinned source; preserve resource fixes and wallet implementation	Met — Git ancestry includes 04f473d; runtime preservation and wallet regressions pass.
2	Source, static preview, owner package; no production deployment or paid gameplay	Met — web/src/, dist/, deploy/INSTALL-LIVING.md; observer interface.
3	“One convincing functioning settlement” within the large world	Missing — Functioning settlement exists; visual cohesion remains unresolved.
4	Fabledom primary; Tiny Glade and Foundation references	Met — Reference hierarchy documented in ART-DIRECTION.md.
5	“Study their official screenshots”	Missing — ART-DIRECTION.md:3 explicitly acknowledges this was not done.
6	Visual references without copied game assets or recreated engines	Met — Original canvas painters; existing engine retained.
7	One palette, projection, scale, lighting and shadows documented	Met — ART-DIRECTION.md; storybook.ts.
8	Recognizable yellow sprout-headed species	Met — Existing character atlas remains recognizable in the browser.
9	Eliminate sprite/vector clash; adapt character/environment treatment	Missing — Hard vector foliage remains beside dimensional characters; storybook.ts:71.
10	Evaluate Kenney and Quaternius resources	Met — Evaluation and rejection rationale in ASSET-SOURCES.md.
11	Compatible free resources with verified terms, or original assets	Met — Original environment painters selected under the permitted fallback.
12	Exact source, creator, version/hash, license and modifications	Missing — Version/hash fields are unpinned descriptions; ASSET-SOURCES.md:7–8.
13	Required licenses/credits; no purchased or ripped assets	Met — Original artwork and retained notices under web/public/licenses/.
14	Reproducible offline rendering and editable sources for selected 3D models	Met, conditional — No 3D models selected; editable painter source supplied.
15	Keep existing renderer; no replacement demo engine	Met — WorldRenderer and existing Expanse simulation retained.
16	Shelter, storage and work appearances; two meaningful variants each	Met — Cottage/roundhouse, rack/baskets, and distinct garden/hearth forms.
17	Honest binding to actual functions and shelter zones	Missing — Storage/ga

Fixed striker inventory prediction and simplified the ignition gate.

Passed: 268 backend, 63 frontend and 50 wallet checks, plus typecheck/build.

All four findings are answered in .imd-responses.json. Long-run pile growth remains unresolved; the screenshot finding is disputed as a component mismatch.

Validation report

Wrote .imd-findings.json with {"findings":[]}.

Both prior findings are fixed:

• Fire-free shelters no longer show smoke.
• The contact sheet includes all 189 keys, including the missing tree variant.

Verified at 1440px and 390px. Frontend tests, typecheck and build passed; rebuilt JavaScript matches the committed export. Application files remain unchanged.

Of the five findings from last round, four are fixed. The fifth, pile growth, still reproduces, and the author now documents it as unresolved. That is the only entry in .imd-findings.json. I found no new defects I could reproduce.

Settled

Earlier finding	Result
[low] The strike-on-pile step said it used up the stone or flint, but the engine kept it	Fixed. The step now declares that nothing is consumed (server/src/expanse/engine.js:328), which matches what the engine does. The new regression tests pass.
[info] The fire-start check had a redundant flint clause that never mattered	Fixed. It is now just "holds a stone" (engine.js:321). The separate wood-only, flint-only, stone and stone-plus-flint cases are still tested.
[medium] Every finished cottage showed chimney smoke, even with no fire anywhere	Fixed. A cottage now smokes only if it is finished, has a chimney, and a fire with fuel left sits on or next to its tile (worldRenderer.ts:265, storybook.ts:583). The fire positions and fuel come from the server's view (view.js:29-30). Tests cover the no-fire, empty-fire and damaged-roof cases.
[low] The contact sheet claimed to show every art cell but left out tree variant 2	Fixed. catalogue() now builds its list with the same functions the world renderer uses, over 240 ids, so all three tree variants appear.

Still open: unattended runs keep piling up fuel (medium), engine.js:324

I re-ran it on the final tree with seed 7 and 40 starting residents:

Tick	Ground piles	Single-unit piles	Residents
6000	70	43	72
7000	82	47	90
8000	99	61	102

There were 0 or 1 fires burning at those points. The resource ledger matched exactly every time, so nothing is lost or duplicated. The problem is that fuel gets left on the ground: residents still drop wood on open ground when they hold a stone, a plain-stone strike succeeds only 35% of the time, and nothing sends them back to the wood afterwards. The author's own data shows 91 of the 117 piles at seed 7, tick 9000 came from this placement rather than from deaths or harvests.

The author agrees the 80-pile limit is crossed and doesn't claim a fix, so I'm not disputing their numbers. I kept the finding open for two reasons:

• The task asks for bounded pile growth in unattended multi-seed runs.
• tests/chains-multiseed.test.js only enforces the limit up to tick 3500, while its run to tick 9000 merely records the counts. A green test therefore still hides the scattered fuel, which the task explicitly says it must not do.

Fixing it needs your decision on approach: have residents return to the fuel they placed, tighten when placing is allowed, or document a growth model. A silent cap or deleting stock would break the requirement to preserve stocks.

What I ran

• Fuel, wallet and save/load tests: tests/chains.test.js, review-regressions, the three wallet suites and durability all pass, 66/66 with 4 tests running in parallel. That parallel load gave a little evidence on the wallet timing robustness the task asks for. I found no failures.
• Not run: the full 268-test backend suite, the frontend tests and the browser checks. No dependencies were installed (the server has no lockfile), and time was short.
• Docs: the README's storybook section correctly scopes its "40/40" to the chain and wallet subset. The validation doc says plainly that the full backend suite was not re-run for that change.

Area	Previous review	This review
Requirements	Full counts unavailable	45 met, 5 missing, 0 faked, 1 needs requester
Broken interactions	—	0 reproduced
Product on first screen at 390px	—	Yes
taste@1	—	16/16 passed
Copy findings	—	0
Share card	—	Present; 5/5 metadata checks
Open findings	2 high, 1 medium, 2 low	1 high, 1 low

Three prior findings are fixed. Final evidence remains incomplete. Findings and exact reproductions are saved in .imd-findings.json.

Prior finding	Resolution
fidelity: primitive vegetation	Fixed visually. At 1440px, camera (24,24), spans 22 and 12.73, foliage now has irregular edges, texture and continuous shading. Reference-study documentation remains incomplete.
fidelity: missing final captures	High, retained. The combined tree contains the disputed files: 32 storybook files, zero recordings, and desktop JPGs measuring 960×600. Capture the final export unedited at both required widths and cameras. The author’s absence claim correctly describes its incoming backend branch, not this combined revision.
fidelity: smoke without fires	Fixed. At tick 12,600, Fires is 0 and cottages show no smoke. Rendering now requires an authoritative nearby burning fire.
fidelity: incomplete contact sheet	Fixed. The browser renders 189 keys, including tree:forest:broadleaf:2; coverage tests pass.
fidelity: missing asset versions	Low, retained. ASSET-SOURCES.md:7–8 still says “Deterministic from source” and “unchanged”. Record immutable hashes and replace the obsolete 67-cell count.

Verification passed: 268 backend tests, 65 frontend tests, 50 wallet/auth regressions, 11 focused fuel/persistence tests, typecheck and build. Counts overlap. Chromium passed desktop/mobile interactions; WebKit passed at 390px. Residents advanced, inspectors and resource quantities worked, and camera travel reached cold, desert and coastal views. Rebuilt application and worker bytes match the committed export. No final test failed; application files are unchanged.

The visual idea is a working settlement observer carried by golden sprout-headed residents in an illustrated isometric world. The revised vegetation resolves the specific clash identified against the Fabledom reference screenshot.

The complete requirement checklist follows. “Met” identifies supplied or verified evidence; limitations are stated where applicable.

#	Brief requirement	Status and evidence
1	Continue pinned 04f473d, retaining upstream provenance	Met — Git ancestry; README provenance.
2	Preserve resource fixes and verified wallet implementation	Met — engine diff, fuel and wallet regressions.
3	Deliver source, static preview and owner-run package	Met — web/src, dist, server, owner installation documentation.
4	No production deployment or paid gameplay	Met — isolated preview and observer controls.
5	Functioning settlement integrated into the large world	Met — seed-2 demo and biome navigation.
6	Fabledom primary; Tiny Glade/Foundation secondary	Met — ART-DIRECTION.md identifies that hierarchy.
7	Study official screenshots	Missing evidence — ART-DIRECTION.md:3 still explicitly disclaims that study.
8	References without copied game assets or recreated engines	Met — original painter and retained renderer.
9	One documented palette, projection, scale, lighting and shadows	Met — ART-DIRECTION.md; storybook.ts.
10	Keep yellow sprout-headed STRATOKS	Met — retained character atlas and browser vi

Recorded both findings in .imd-responses.json with supporting validation.

• Final storybook artwork is absent from this checkout; capture remains incomplete.
• Long-run scattering reproduced: 113 piles at tick 9,000, with exact accounting. Remains unresolved; gameplay unchanged.
• Checks passed: 268 backend tests and 50 wallet/auth regressions.

Fixed the finding in ASSET-SOURCES.md: added exact SHA-256 hashes and corrected the catalogue to 189 keys.

Verified the browser count and matching source/export atlas bytes. Recorded the outcome in .imd-responses.json.

Documentation-only revision; runtime files remain unchanged. Build and regression suites were not rerun.

Saved review findings: one medium finding for incomplete asset mapping and one low finding for missing evidence screenshots. No critical or high hosting blockers found.

Checks passed: 268 backend tests, 65 frontend tests, 19 wallet browser scenarios, build and typecheck. Only the findings file was written.

The review is complete and the findings file is written. Nothing in the tree was modified apart from .imd-findings.json.

My prior finding stands, unfixed, and the author agrees. Re-running the exact seed-7 recipe on this tree gives the same numbers as last round, so the dispute is only about scope, not about the facts.

tick	piles	single-unit	residents	ledger mismatches
1000	18	9	40	0
5000	58	39	67	0
7000	82	47	90	0
9000	113	59	115	0

The placement gate at server/src/expanse/engine.js:324 is unchanged, and the multi-seed test still asserts the 80-pile bound only through tick 3500. Accounting is exact at every sample. This remains an incomplete requirement rather than a ledger bug, and it needs an owner decision on policy. I kept it at medium.

Both specialist findings reproduce and are kept, merged into my list:

• Medium: shipped atlas manifest is incomplete. The published manifest in dist and web/public omits the two agency sheets and the kneel sprite that the runtime manifest declares and the bundle loads. The integration doc's claim that the two manifests are identical is false.
• Low: missing resource-selection screenshots. The after-report lists two selection captures that do not exist, and the storybook validation page cites that report as its evidence for quantity display.

One new info-level note. The README reports 63 frontend tests, while every frontend test file on this tree now yields 65 passed. Not a behavioral defect, but task item 4 asks for stale counts to be corrected.

Everything else I checked passed. Backend suite 268/268 at concurrency 2. Wallet and auth suite 50/50 at concurrency 8 while a 9000-tick simulation ran alongside. Frontend 65/65. The wood-only and wood-plus-ignition fuel cases, persistence load/save/restart tests and striker regressions are all in the passing backend run.

Area	Previous counts supplied	This review
Requirements: 46 grouped checks	—	38 met, 7 missing, 0 faked, 1 needs requester
Broken features reproduced	—	0
Product on first screen at 390px	—	Yes
taste@1	—	16/16 passed
Copy findings	—	0
Share card	—	Present

Saved two findings in .imd-findings.json.

• High — fidelity: final capture evidence remains missing. The current combined checkout contains the cited storybook directory. Its desktop stills are 960×600, predate the final foliage revision, and have no accompanying final recording. Six listed selection screenshots are absent. Capture the final export at 1440×900 and 390×844, retain the requested views and recordings, and refresh the reports. The author’s component-scope explanation is valid for that earlier checkout; it does not close the combined-delivery gap.
• Medium — fidelity: the public atlas manifest omits active assets. Both public/exported manifests omit agencyTerrain, agencyActivities and agency.kneel, although the browser loads both sheets successfully. Synchronize the public contract with the runtime atlas and retain their source mappings.

The previous asset-identifier finding is fixed: both SHA-256 values match, and the live sheet renders 189 unique keys. The specialist’s missing-screenshot finding is included in the capture finding above.

STRATOKS is a functioning isometric observer whose visual idea is golden residents living among warm timber, thatch and textured foliage. No additional aesthetic finding was substantiated this round.

The requirement ledger below includes inherited limitations; it does not introduce additional revision findings. “Met” includes identified implementation or recorded validation evidence, with fresh checks distinguished afterward.

#	Explicit requirement	Status	Element or evidence
1	Continue the pinned source; preserve resource fixes and wallet implementation	Met	Git ancestry; chain/wallet regressions
2	Deliver source, static preview and owner-run package	Met	web/, server/, dist/, deploy/
3	No production deployment or paid gameplay	Met	Isolated observer demo; owner deployment instructions
4	One functioning settlement integrated into the existing large world	Met	Running settlement in /preview/?demo=1
5	Study official Fabledom, Tiny Glade and Foundation screenshots	Missing	ART-DIRECTION.md explicitly records this omission
6	Use references without copied game assets or a recreated engine	Met	Original environment painter; existing renderer
7	One palette, projection, scale, lighting direction and shadow system, documented	Met	ART-DIRECTION.md; storybook.ts
8	Keep yellow sprout-headed characters; adapt edges, scale and shadows	Met	Character atlas retained; shared shadow treatment
9	Eliminate flat-tree/primitive-hut clashes beyond recoloring	Met	Current textured foliage and distinct constructed buildings viewed in-browser
10	Evaluate Kenney and Quaternius; OpenGameArt optional	Met	ASSET-SOURCES.md evaluation table
11	Compatible free assets with verified terms, or consistent original assets	Met	Original-art fallback selected and documented
12	Record source, creator, version/hash, license and modifications	Met	Corrected ASSET-SOURCES.md; hashes independently matched
13	Include licenses/credits; no purchases or ripped artwork	Met	web/public/licenses/; provenance documentation
14	For 3D inputs, ship reproducible rendering and editable sources	Met	Conditional requirement: no new 3D inputs used
15	Keep the renderer; do not substitute a demo engine	Met	Existing worldRenderer.ts integration
16	Distinct shelter, storage and work a