MANGGARAI_ARCHITECTURE_REFERENCE.md

world/research/MANGGARAI_ARCHITECTURE_REFERENCE.md

Manggarai architecture — the build-usable reference pack

What this is. The reference substrate the architecture lane asked for. The game repo's

QA/look/architecture_reference_plan.md names five structures it must model, splits each into

ATTESTED and NOT ATTESTED, and lists the reference each one needs before anything is built. This

pack fills the open-source half of that list: structure specs per building class, the geometry and

materials tables, the eye-level look register, and the §17.1 care layer — every fact carried with

its agreement tag, every disagreement declared rather than averaged, and a named worklist of what

only the Erb anchor can deepen.

Tier. Reference substrate, proposal-tier. It ratifies nothing, mints no registry row, and flips

no status on any board or bible. Canon is docs/spine/CH_03.md, the region page

_source/02_Tier_2_Region_Pages/flores_island.md Section 7, and the CVD; where this pack and canon

disagree, canon wins and this pack is the defect.

Method. Two independent deep-research passes ran against the open literature and returned

separately. This pack is the cross-check: a fact that both passes reached independently is tagged

BOTH; a fact only one pass reached is tagged and named as single-source rather than quietly promoted;

a fact this repo's own processed corpus already carries is tagged REPO and is the strongest class in

the pack, because it is inside the citation apparatus rather than beside it. Arithmetic done here is

tagged DERIVED and is never presented as a source claim. Two arithmetic errors and one canon

conflict in the incoming research were caught in this pass and are recorded in section 13 rather

than silently corrected.

Compliance. All facts paraphrased with per-fact attribution. No bulk text reproduced from any

source. No images acquired, downloaded, or placed in the pipeline by this pass.

---

0. How to read the tags

• BOTH — both research passes reached it independently. The pack's working-truth class.
• A-ONLY / B-ONLY — reached by one pass. Declared single-source; usable, but it carries the

weight of one chain and it is a named Erb verification target if it is load-bearing.

• REPO — already attested inside this repo's processed corpus (the region page, the spine node,

or a SRC_ row). Strongest class: it is already citable by the normal apparatus.

• CONFLICT — the sources disagree. The disagreement, the resolution, and the reason for the

resolution are all stated. Nothing is averaged.

• DERIVED — arithmetic performed in this pack from figures cited above it. Not a source claim.
• UNSOURCED — named as a hole. Never filled with a plausible number.

Source authority, highest first: the measured structural analysis (Pradipto and Tristanto, *Jurnal

Arsitektur Pendapa* 4(1), 2021, UGM — a wind-resistance study of the Wae Rebo mbaru niang with plans

and sections, the only measured engineering source either pass reached) · the elder-informant

nomenclature study (Jem and others, *Symbols in Mbaru Gendang*, Migration Letters 20 S11, 2023) ·

the heritage-body project record (UNESCO Asia-Pacific 2012 winners profile; the Aga Khan

Development Network project page) · institutional secondary (BME Budapest *habitatio*) ·

Indonesian architecture-media and eyewitness build accounts, used only where several agree.

---

1. The site frame

The seven houses are widely held to answer the seven peaks that ring the village, honoured as

protectors (A-ONLY). That is a good reading and it is not needed to justify the count, which is

independently attested and already canon.

---

2. Mbaru niang — the five-level program

The interior program is the part the game repo's plan lists as NOT ATTESTED and blocks on. It is

attested in the open literature at the LEVEL grain — names, order, and function — by both passes and

by the measured study. What remains genuinely Erb-only is the arrangement *within* a level: partition

sizes, furnishing, per-level floor heights. Section 15 keeps that split explicit.

Bottom to top:

CONFLICT, resolved. The English Wikipedia article inverts levels 3 and 4 (making *lentar* the

crop-failure store and *lempe rae* the seed store) and counts six levels by folding the under-croft

in as a level named *ngaung*. Both research passes independently found the peer-reviewed study and

the majority of Indonesian-language sources against it. **Build L3 = seed, L4 = drought reserve, and

five named levels above the under-croft.** The six-level count is the same building described with a

different boundary, not a different building.

The vertical cosmology (B-ONLY, from the elder-informant study): three bodies — *kaki* / *badan*

/ *kepala*, foot / body / head — mapping onto under-croft / living floor / roof loft, and read as the

world of the dead, the world of the living, and the seat of the ancestors and God. The under-croft is

the darkest register in that scheme and was historically also the bond-servant's register. The five

named levels sit *inside* that three-body frame; they do not replace it.

The level-1 plan rule (BOTH, and the strongest single staging fact in the interior):

• The hearth sits at the centre of the floor, at the foot of the sacred post, and it is what

divides public from private. There is no partition doing that job.

• The men's / guest zone is at the front, by the door; the women's activity is at the hearth and

kitchen side; the family chambers (*lo'ang*) are partitioned around the rear.

• Sleepers lie with their bodies oriented toward the centre post.
• At a gathering the community sits in a ring and **the elder sits in front of the post, facing the

door.**

Doors (B-ONLY): *para olo* is the front door — guests, men, the wife-giver line; *para musi* is

the back, for the household and for kitchen access during ceremony. **At Wae Rebo every front door

faces the compang**, so all seven face inward to the ring's centre. That is the orientation rule the

layout lane needs, and it is single-source: treat it as the working rule and put it on the Erb list.

Windows and doors as a period marker (B-ONLY, CONFLICT): the original interior is described as

dark, one door, no windows; the rebuilt houses are described with two doors and four small windows.

Both are attested, and the reconciling reading is that this is a **pre-rebuild / post-rebuild period

marker** rather than a contradiction. If the build wants the older register, it is one door and a

near-black interior.

---

3. Mbaru niang — the structural system

This is the section the plan calls for by name ("structural section — the ring-beam/post arrangement

that carries the cone… what makes the silhouette correct rather than approximately conical"). Every

row here is BOTH unless marked, because the measured study is the one source both passes found and

neither pass found anything contradicting it.

• Nine main posts (*sembilan tiang utama*): one centre, eight around it. They are driven directly

into the ground and they rise only to the level-2 floor beams, where they stop.

• The two volumes are separate structures. The middle (living) volume and the upper (roof)

volume are mechanically divorced, and the living-floor deck is framed independently of the nine

posts. Wind load on the roof does not transfer into the dwelling floor. This is the design's

signature and it is legible in play: damage, collapse, and repair states can be authored

per-system, because the building genuinely has two systems.

• The centre post is not one continuous timber. Two members, pin-jointed:
• *hiri bongkok* (also *tiang bongkok*) — from the ground and under-croft up to level 1;
• *tiang ngando* — from level 2 to the apex;

joined by a pin/hinge connection (*sistem sendi*) at the upper-room floor, reinforced by

clamping between two bamboo poles whose rungs double as the climb to the upper levels.

CONFLICT, resolved: popular sources describe a single intact ~15 m *worok* trunk running ground

to peak. The measured structural survey contradicts this and both passes flagged it independently.

Build the two-piece pin-jointed post.

• Four inclined struts (*pengga ngando*), set to the four directions, bear on the upper-room

floor beams and hold the *ngando* plumb. The consequence is the interesting one: **the ngando is a

stay-post, not a bearing post** — the struts take the force path.

• The engineering intent, stated by the measured study: the pin joint neutralises downward wind

force at the splice so it is never passed into the *bongkok*, keeping the nine posts unloaded by

wind, and it shortens the buckling length against one continuous member.

• Foundation: *ceblokan*, a fixed-embedded (*jepit*) system — posts buried directly, not seated

on *umpak* stone pads. The study reads this as an explicit anti-uplift, anti-slide wind choice,

consciously traded against the earthquake and insect performance a pinned stone base would give.

Embedment 1.5-2.0 m (CONFLICT: one outlier gives ±80 cm for the lower columns; use the band).

Post feet are wrapped in *ijuk* against rot (B-ONLY); contemporary practice adds plastic under the

*ijuk*, which is a present-day period marker (B-ONLY).

• Ladder: the *rede*, bamboo, lashed to the centre post from level 1 upward — distinct from the

bamboo clamp poles that also serve as rungs.

• Form and wind logic: a cone, wide at the base and narrowing to the apex, stable from any wind

direction; the roof skirt descends to within about a metre of the ground, which blocks

horizontal wind entry under the building and resists roof uplift.

• Rebuild-era innovation (heritage-body record): thicker bundled-reed ring beams stiffening

the bamboo roof frame against cyclonic wind. This corroborates the pin-joint finding from the

project side as well as the survey side.

3.1 Joinery and lashing — the whole fastening vocabulary

Timber is never mortised or notched. Members stay whole so the full section strength is retained,

and everything is tied. Rattan outperforms pegs and bolts here because rigid fasteners split and tear

the timber, and a lashed joint **flexes and slips slightly under wind without displacing or

fracturing** — that is the stated design rationale, not a folk explanation. Three named types:

**Zero nails is a stated design law. A visible metal fastener anywhere on this building is a canon

break.**

---

4. Geometry

4.1 The measured and corroborated figures

CONFLICT, resolved. A "15 m diameter" figure circulates widely. It is not the floor. Read it as

the roof's ground spread, which is necessarily wider than the raised floor — see 4.2. Separately, the

measured study prints "diameter 14 cm" at one point, an obvious typesetting error for metres; do not

cite that figure in either unit.

4.2 DERIVED geometry — the numbers the dressing and greybox lanes actually need

All rows below are arithmetic from 4.1, not source claims.

The look-law this settles: it is not a fat tepee, it is a needle — 65 to 70 degrees. Any

greybox at 45-55 degrees reads instantly wrong. And the silhouette hierarchy in the ring is by

fatness, not by height: all seven stand at the same 15 m, and the chief's house is the wider,

blunter, more shouldered one.

4.3 The blockout's proportions, answered

The plan's section 1 table invites correction and forbids silent change. Three of its rows move.

4.4 The ring diameter — UNSOURCED, and the discipline around it

Neither pass could source the centre-to-centre ring diameter or the house spacing. It is the largest

open geometric hole in the pack and it must be measured, not invented.

The correct chord formula for seven houses evenly spaced on a ring of centre-to-centre diameter

D is:

• chord (centre-to-centre between neighbours) = D · sin(π/7) = 0.434 · D

*(One incoming pass wrote this factor as 0.868·D. That is the radius form — 2·sin(π/7)·R — mislabelled

onto the diameter, and it doubles every spacing derived from it. Corrected here; recorded in

section 13.)*

What that constrains, DERIVED:

• Skirts must not overlap: chord ≥ ~14 m (the two ground spreads averaged) → D ≥ ~32 m.
• The ring plus its plaza, the field fan and the approach must sit inside a ~6,500 m² site → **D

comfortably under ~60 m** (a 60 m ring already encloses 2,827 m² before anything outside it).

• So the defensible band is D ≈ 32-60 m, i.e. a ring RADIUS of 16-30 m — against the blockout's

52 m radius.

The measurement route: satellite imagery of Wae Rebo, Satar Lenda, Satar Mese Barat, at

~1,100-1,200 m. Measure the ring, record it as a datum with its capture date, and write it into the

blockout. Until that is done, the band above is the sanity check and no point value is canon.

---

5. Materials

CONFLICT, resolved. The English Wikipedia article gives a partly different species list (majegau,

mahogany, Bolly beech). Prefer the worok / ajang / uwu set: it recurs across independent

Indonesian architecture sources, it matches the elder-informant study's *haju worok*, and it is what

both passes converged on.

Thatch preparation (BOTH): *alang-alang* is usable once it reaches about a metre in length. Panels

are woven — reported at ~9 m long (A-ONLY) — and then cured by smoking over the kitchen hearth

before installation. The drier the *ijuk* and *alang-alang*, the more durable the roof. In service the

permanent hearth smoke continues the job: it deters moths and borers and retards rot.

Palette discipline for the dressing lane: the whole village is four materials — dark

hardwood, bamboo (pale when new, silver-grey weathered), grass (gold to silver), black palm fibre —

all tied with rattan. No metal anywhere.

---

6. The skin — tiered thatch bands

This is the highest-value dressing law in the pack, and it is also the pack's weakest citation. Both

passes returned it and both traced it to the same vernacular chain, so it is **reported twice but

sourced once**. It is Erb verification target number one.

The layering gradient, bottom to top:

• from the skirt upward, **2 courses *alang-alang* to 1 course *ijuk***, repeating up to level 4;
• at level 4 the ratio changes to 1:1;
• at level 5, the top cone, it is ijuk only, closed at the point with the *wolet ngando* wrap.

The look-law that follows: the cone is a value gradient, not a uniform texture — pale

straw-gold at the skirt, progressively darker in discrete bands, and a **solid black-brown cap on the

top fifth** with a visibly bound collar at the point. Band boundaries are horizontal circles, so the

gradient is stepped, not smooth. A uniformly brown cone is wrong.

Horizontal ribbing is structural, not decorative (BOTH): concentric rattan and reed ring bundles

girdle the cone — the longest at level 1, each shorter going up — plus the bamboo *reng* laths binding

alternating thatch bundles. This is what produces the signature tight concentric ribbing, and it means

the texture direction is circumferential. It is what sells the silhouette in profile and at

distance.

Thatching direction: bottom-up. One aggregator claims a top-down build; it is contradicted by an

eyewitness build account and by the gradient itself, which is defined from the skirt upward. Closed.

The apex finial — OPEN. The elder-informant study documents, on the *mbaru gendang* type recorded

at Ruteng, a mangka (a wooden plaque carrying a painted human face) flanked by rangga kaba

(buffalo horns). Whether the Wae Rebo niang carry this is unconfirmed — the Wae Rebo sources name

only the *ngando* apex and its *wolet ngando* binding. B-ONLY, and a genuine open question. Working

disposition: if used at all, use it on the drum house only, and resolve it on Erb or on high-resolution

apex reference.

---

7. The compang

CONFLICT, declared and NOT averaged. Two compang forms are attested. The wider Manggarai

description carries a platform that can rise to about 2 m, with a large *beringin* standing in or

beside it and an ancestor grave at the centre. Wae Rebo's is consistently described as **low and

round**, with a square stone and a bamboo offering rack. Read the tall stone-and-tree version as the

inland or older variant and build Wae Rebo's as the low platform. Do not blend them into a

compromise object that exists nowhere.

**The drum house — *niang gendang* / *mbaru gendang*.** The principal house, typically the largest.

It keeps the heirloom drums and gongs, it houses the adat head, ceremonies are held there, and **all

arrivals are received there (BOTH). The governing philosophy is "gendang oné, lingko péang"** —

drum-house inside, garden outside — one indivisible social-territorial unit binding dwelling,

communal labour, and cosmic order (BOTH; orthography varies as *gendang one, lingko pe'ang*).

This is a settlement-design law, not decoration. House-ring and field-web are one unit. A region

page or a level that draws only the houses has drawn half the thing.

**The five parts of a *beo*** (B-ONLY, and worth treating as a completeness checklist): *mbaru

gendang* the drum house · *natas* the plaza where all social and ceremonial life happens · *compang*

the altar standing inside the natas · *wae teku* the water source · *boa* the cemetery — with *lingko*

outside. A Manggarai settlement is not complete without all five, and a village dressed with houses

and no water source and no cemetery reads wrong to anyone who knows. Note that the water source is

already canon here as a staged site (wae_water_spring); the cemetery is not, and that is a real gap

worth naming rather than a licence to invent one.

Orientation, and a tension worth holding rather than averaging (B-ONLY): Manggarai has no

cardinal-direction system. The four terms are *awo* (sunrise), *sale* (sunset), *le* (upstream,

toward the mountain and the water source, the source of life) and *lau* (downstream). Elsewhere in

Manggarai the *mbaru gendang* faces *le* with the compang opposite. At Wae Rebo the ring geometry

supersedes that and every front door faces the compang. Flag the tension, do not average it: a

ring village orients to its own centre; a single-house village orients upstream. The four *pengga

ngando* struts set to the four directions are an orientation datum baked into the roof itself.

---

8. Lodok / lingko — the field geometry

The repo's own processed corpus already attests the core: *"the named lingko with its ritual

center-stone (lodok), the tu'a teno who reads, plants, and ritually manages it, and the uma randang

festival at which new fields are opened"* — that is the region page's Section-17 mark citing

SRC_00203, and it is the strongest citation in this section because it is already inside the

apparatus. Note that the repo's phrasing is a ritual centre-stone and research pass A's is a

*teno*-wood pole. Both may be true of different lingko or of the same site at different moments;

neither pass resolves it. Named as an Erb item, not silently merged.

8.1 The division mechanic

The household head holds a number of fingers against the teno pole; that finger-span is marked on

the pole, and **two lines are struck from those two marks out to the *cicing***, defining the wedge.

The span scales to household size and adat standing:

• *moso rembo* — five fingers, the largest share
• *lide* — three fingers
• *koret* — one to two fingers, typically newcomers

The geometric consequence, and it is a good one: every household's plot is **narrow at the sacred

centre and widens outward** — equal proximity to the sacred origin, unequal area at the rim.

Tag: A-ONLY. This is the single most build-usable institutional fact in the pack and it rests on

one chain. The game repo's plan already blocks the lodok allotment geometry on Erb and says plainly

that blocking is correct and guessing is not. That block stands. This pack does not lift it; it

gives the block a specific shape, so that when Erb lands the question is "does Erb corroborate the

finger-span system, and with what spans" rather than "what is the geometry".

8.2 Authority and sequence

• Tu'a Teno — Lord of the Land: directs siting and measurement, holds customary right over the

adat territory. (A-ONLY for the title's scope; REPO for the office itself, which the region page

carries as *tua teno*.)

• Tu'a Golo — the village head. (A-ONLY / REPO, same split.)
• Tu'a Kilo / Tu'a Panga — household and lineage heads, who determine beneficiaries. (A-ONLY.)

Ritual sequence around a division (A-ONLY): *barong boa* (summoning the ancestors) → *teing hang*

(feeding and offering) → *reke lodok* (the planning assembly) → *wuat wa'i* (the final blessing) →

*lilik compang* (circling the altar). Planting, harvest and new-lingko rites may include water-buffalo

sacrifice.

Live context, not a historical curiosity (A-ONLY): post-colonial statute law progressively eroded

adat authority over allocation, and the grievance is current — the Poco Leok geothermal land

litigation is the contemporary corroboration. The Ch-3 spine's own conflict is a land-grab against

lingko-adjacent forest, so this is the real-world register the chapter is already written in, and

knowing it is what keeps the beat from playing as invented melodrama.

8.3 The spider-web at three scales

The single strongest visual-structural idea in this culture, and it is free to use (B-ONLY): the

radial roof frame (*siri leles*) is explicitly built in the same figure as the *lodok* field

division — a centre point with wedges radiating out, one wedge per family. **Roof, village, and

farmland are the same diagram at three scales.** The spine already carries the first and third of

those as Thread 13's sacred geometry; the roof frame is the missing middle term.

---

9. Eye-level look register

What a dressing lane needs at standing height, in priority order.

1. Needle cones at 65-70°, one fatter and blunter than the other six. Skirt stopping about a

metre off the ground. No visible wall plane anywhere — the cone is the whole building, roof

and wall are one surface.

2. Stepped tonal banding up the cone: straw-gold skirt, mid bands alternating, a **black ijuk cap

on the top fifth**, bound at the point. Stepped, never a smooth gradient.

3. Horizontal ribs every course — ring bundles and bamboo laths reading as tight concentric

lines. Texture direction is circumferential.

4. The under-floor gap of ~1.2 m on every house: dark, cluttered, chickens, firewood, tools,

someone weaving at the light edge. **This shadow band under a huge pale cone is the strongest

eye-level read in the whole village**, and it is the read a massing blockout cannot produce.

5. Smoke. No chimney, so smoke leaks through the whole roof mass — a permanent haze bleeding out

of the thatch, thickest at dawn and dusk cook times. At altitude, in a bowl, sitting in cloud.

6. Per-house weathering variance keyed to occupancy. This is documented, not invented: an

architect's on-site account records that houses with a daily fire hold their thatch measurably

better than the guest house where nobody cooks. So in a village of otherwise identical structures,

the lived houses are darker and tighter and the unused one is greyer and shaggier. It is a

free, canon-grounded reason for variance that costs nothing and defeats the copy-paste read.

(B-ONLY, but it is an eyewitness observation and it is mechanistically consistent with the

smoke-curing fact both passes carry.)

7. The ground plane: a swept open *natas* with the stone compang at its centre, coffee and grain

drying on mats, the ring of doors all facing in.

8. Four materials, no metal. See section 5.

---

10. Build sequence, labour, and state authoring

For build-in-progress, damage, and repair states.

• ~1 year gathering materials from the surrounding forest; ~3 months from foundation to

finish, by cooperative male labour in evening sessions where elders transfer the knowledge, each

session opened and closed by drum. (B-ONLY.)

• Roof completion: 40 people, 2 days. (BOTH.) Thatch is pre-dried in the smoke of an existing

house first.

• The main post is guyed by three or four rattan lines to stakes while it is set. (B-ONLY.)
• Ritual sequence for a drum house (B-ONLY, elder-informant study): *tombo pande mbaru* (the

address) → *nempung weki bantang cama* (planning) → *racang cola* (sharpening the tools) → *paki

haju* (felling the main-post tree) → roko molas poco — parading the tree in as a "forest

maiden", a young girl riding it, rituals at every river crossing, the tree welcomed exactly as a

bride from another village → *hese siri bongkok* (raising the main post) → *we'e mbaru* (entering)

→ *congko lokap* (closing and cleaning). Research pass A independently attests ceremonies at the

felling of the central post, at the start and completion of construction, and at family occupation,

with animal offerings — so the felling rite is BOTH and the named sequence is B-ONLY.

• Durability (CONFLICT): one line of sources gives thatch 7-10 years and bamboo 10-15; Indonesian

media claim selected-*ijuk* roofs last "tens of years". The reconciling variable is almost certainly

the daily fire. Working assumption: **~10 years to re-thatch on a fired house, considerably shorter

without.**

10.1 The conservation record

Useful as the living-heritage register, and all of it BOTH or heritage-body:

• By August 2008 only four mbaru niang survived, two of them needing renovation.
• Architect Yori Antar and Rumah Asuh initiated a community-led revival of the traditional

technique. Design 2008; two houses restored 2009-2010 and three new-built, complete 2011; the

full ring of seven reinstated. Architecture students documented the work annually.

• UNESCO Asia-Pacific Awards for Cultural Heritage Conservation 2012 — Award of Excellence. The

official citation credits the valorization of traditional knowledge in continuing architectural

form and construction practices, the re-established sustainability of the local built environment,

and the safeguarding of both tangible and intangible heritage values. (REPO as well — the

region page Section 7 and docs/spine/CH_03.md both carry the 2012 award.)

• Aga Khan Award for Architecture 2013 — shortlisted, one of twenty.

Two consequences for the build. First, construction-sequence photography exists and is the best

future reference set — it is third-party rights-reserved and is reference-for-human-eyes only, never

a source any asset is derived from. Second, the building standing today is a **2011 reconstruction of

a documented technique by the community that owns it**, which is precisely why the technique is so

well recorded and why the *ceblokan* foundation and the pin joint are described rather than inferred.

---

11. The §17.1 care layer

The frame, stated first, because it decides everything below. The Care Doctrine (CVD §17.1) is

affirmative: care means weight and authenticity, never omission. Everything in this section is

representation guidance — it tells the build how the place is inhabited so that it reads as

inhabited rather than exhibited. None of it is a reason to leave anything out.

The floor. The CVD §17 hard lines bind and nothing in this pack approaches them. §17.1's

represent-with-care duty is collective: it protects the community from being cast as villain,

defamed, or exterminated. The Ch-3 spine already routes the chapter's antagonist through the outside

broker's crew — the oppressor apparatus — and never through the Manggarai, and it classes the

community as a protected host with standing set by the player's conduct. So the architecture

lane's care duty here is accuracy and weight, not restriction. The rated-M bar is untouched:

combat at the lingko-edge clearing line against the broker's crew is core and unrestricted.

Capture rights: RULED, no gating. The thirty-first sitting ruled it directly — do not gate,

block, or be over-conservative; this is a rated-M action-adventure video game and the artist's

creative rights are retained. There is no capture-rights process doc and no clearance gate. The

game repo's plan carries an unanswered site-rights flag in its section 3; that flag is **already

resolved by standing ruling** and is recorded as resolved in docs/FACTORY_CONTINUATION.md. It is

not re-surfaced, and this pack does not re-open it.

What Lane R actually is, since it is easy to confuse with a rights gate. "Real cultural

architecture is reference-built, never raw-generated" is an art-lane routing rule — it says where

the pixels come from, because a generative model hallucinating a mbaru niang produces a

tepee-with-thatch and that is an authenticity defect. It is a quality rule, not a permission gate,

and it does not make the building conditional on anything. This pack is itself the reference the rule

asks for, at the structural and dimensional grain.

The living-community facts, and what they become in the build. These are the strongest staging

material in the pack, because each is a behaviour rather than a restriction:

• Arrival is gated by ritual. Every visitor is received in the drum house for the **Waelu /

Wae Lu'u** welcome; an elder speaks a Manggarai formula asking the ancestors to accept the newcomer

as a guest and extend protection. You are not present in the village until this is done. → **The

village opens to the player through the drum house, not through a gate or a quest marker.** That is

a better first beat than any of the alternatives, and it is real.

• The compang is not climbed or sat on. → NPCs use it, the player character does not climb it,

and the ground around it is visibly swept. The board already rules upkeep as the whole signal,

and upkeep is exactly what this fact produces.

• Alcohol is ceremonial — taken at the elders' invitation, not carried in. → a hospitality beat,

not a prohibition.

• The hearth is never out, and the smoke is the building's preservation system. → permanent

ambient, permanent haze, and the occupancy-weathering variance in section 9.6.

• *hekang kode* is the sacred summit. **No source either pass reached states an explicit

who-may-enter rule. Treat access as elder-mediated and do not invent a prohibition beyond

that** — inventing a taboo is as much a fabrication as ignoring one.

• *Penti*, the obligatory annual thanksgiving and agrarian new year: 15-16 November, one full day

and night from 06:00, anchored at three sites — the springs (the origin of life), the **village

entry (*pa'ang*), where prayers are said for women whose husbands live away, and the village

rear (*mandong*)** for health and prosperity — with percussion, chanting, the *caci* whip-duel, and

*Sanda* circle-chants running until dawn. (A-ONLY.) The *caci* is already canon in Ch 3; **the

three-site anchoring is the part the build does not yet have**, and it is a free structural gift for

any festival staging.

Three delegated calls, named so they can be vetoed rather than buried.

1. The seven real house-names are not lettered onto the seven houses. They are current,

family-held names of living households; canon needs none of them; the build gains nothing from

them. This is a director-delegated call under the care doctrine, not a prohibition, and it costs

the build nothing.

2. No rite, no offering, no clan-specific marking is staged at the compang in this pass — the

board already rules this and the reason is accuracy: a specific rite staged wrong is worse than a

platform standing quiet. The bound-twig brush on the swept-approach plate remains a **declared

proposal-tier inference** and does not enter the build as fact.

3. The interior is authored only to the level grain this pack attests — the five-level program,

the central hearth, the post-oriented sleeping, the front-door public zone — and not to the

furnishing grain, which stays on the Erb list. Any render made from the hearth-level brief carries

its inferred-interior declaration on its face.

One incoming recommendation is ADJUDICATED AGAINST. Research pass B recommended building a

renamed cognate rather than Wae Rebo itself — no real house-names, no real compang as a

lootable object. The rename half conflicts with locked canon: docs/spine/CH_03.md names Wae

Rebo, the compang, the *mbaru niang*, the lingko and lodok, Todo and Ruteng directly, and the region

page Section 7 does the same. Canon outranks a proposal-tier suggestion, so **the settlement is built

under its own name.** It would also invert the Care Doctrine's own logic: care is weight and

authenticity, and erasing a real name to be safe is omission — the exact failure mode the doctrine

was written against. The non-rename half of that recommendation stands and is already covered: no

asset is derived from any single photograph, the compang is not a lootable or destructible, and the

community is never the antagonist.

---

12. What this pack changes for the architecture lane

The game repo's plan splits every surface into ATTESTED (buildable now) and NOT ATTESTED (blocked).

This pack moves items across that line at a declared confidence. **The disposition is the architecture

lane's and the director's, not this pack's** — what follows is the proposal.

---

13. Verification findings — what the cross-check caught

Recorded because a synthesis that reports only agreement has not verified anything.

1. A chord-formula error. One pass derived house spacing as 0.868 · D_ring. That factor is

2·sin(π/7), which applies to the radius; against the diameter the factor is sin(π/7) = 0.434.

Every spacing derived from the wrong factor is double. Corrected in section 4.4, and the pass's

downstream conclusion (a ring diameter around 36 m) happens to survive because it was cross-checked

against a plaza width rather than against the chord.

2. A blockout ring that exceeds its own site. The 52 m ring radius encloses ~8,494 m², against a

6,500 m² awarded site area that both passes carry. The two figures were in different sections of

two different documents and neither pass compared them. Section 4.3.

3. A canon conflict inside a research recommendation. The renamed-cognate recommendation

contradicts the Ch-3 spine node, which names Wae Rebo and every institution in it. Adjudicated in

section 11.

4. Two "conflicts" that are the same building described twice. The five-versus-six level count

(the under-croft folded in or not) and the *kolong*-versus-*ngaung* naming are boundary and

vocabulary differences, not contradictions. Both are resolved in section 2 without discarding

either source.

5. A false independence. The thatch-layering gradient was returned by both passes and reads like

two-source agreement. It is not — both traced to the same vernacular chain. Downgraded to

single-source-class and promoted to Erb target number one. Section 6.

6. A typesetting artefact in the highest-authority source. The measured study prints "diameter 14

cm". Do not cite it, and do not convert it.

---

14. Declared disagreements, in one place

---

15. The Erb polish worklist — what only *The Manggaraians* can deepen

Erb, *The Manggaraians*, is the MUST-ACQUIRE anchor. The game repo's plan calls it the single

highest-value item on its acquisition list by a wide margin, three of five structures block on it, and

docs/FACTORY_CONTINUATION.md records it as Josh's purchase and the single highest-value acquisition

open. This is the worklist to run against it the day it arrives, in the order it unblocks work.

Each item names what specifically to look for, so the polish pass is a lookup rather than a re-read.

Tier 1 — items where Erb converts a single chain into a citation, and work is blocked meanwhile.

1. The lodok allotment geometry. Does Erb corroborate the finger-span system at the teno pole

— *moso rembo* five fingers, *lide* three, *koret* one to two — and the two-lines-to-the-*cicing*

construction? What spans, and set by whom? This is the plan's declared substrate gap and the reason

the blockout's nine even spokes across a 46° fan are labelled geometry rather than adat.

2. The lodok centre marker. Stone, pole, or both — reconciling disagreement 12.

3. The mbaru niang interior at the furnishing grain. *lo'ang* partition sizes and count, per-level

floor heights, what is actually stored and how on levels 2 to 4, and the arrangement around the

hearth. The plan blocks all furnishing on this and the block is correct.

4. The hearth count — one central *sapo* used in turn, or one per family (disagreement 9).

5. The thatch layering gradient — the 2:1 → 1:1 → ijuk-only progression and the *wolet ngando*

apex wrap. The pack's weakest load-bearing citation and its number-one verification target.

6. The two size classes — the drum house at Ø 14 m against the six at Ø 11 m. Single-source, and

it is what sets the silhouette hierarchy of the whole ring.

Tier 2 — items where Erb settles a real open question that is currently a coin-flip.

7. The apex finial — do the Wae Rebo niang carry *mangka* and *rangga kaba*, or is that a

*gendang*-type marker recorded at Ruteng only (disagreement 11)?

8. Compang construction — course, stone size, how the stones are set, and the low-versus-tall

variant boundary (disagreement 8).

9. The door and window register — whether the one-door, no-window interior is genuinely the older

state (disagreement 10), and what *para olo* / *para musi* usage looks like in practice.

10. Post embedment — the 1.5-2.0 m band against the ±80 cm outlier (disagreement 5).

11. The timber species roster — which member takes which species, and whether the seven-species

count holds (disagreement 6).

12. **The *hekang kode* access rule** — no source either pass reached states one. If Erb states it,

the build has a real rule; if Erb is silent too, the silence is itself the finding and elder

mediation remains the working read.

Tier 3 — items where Erb would enrich material the build can already stage.

13. The village plan and the ring diameter — if Erb carries a measured or surveyed village plan it

closes section 4.4 outright. Otherwise the satellite measurement stands as the route.

14. **The five parts of a *beo*** — corroboration of the *mbaru gendang* / *natas* / *compang* / *wae

teku* / *boa* completeness set, and in particular where the *boa* sits relative to the ring, which

is the one canon-absent member.

15. The Penti three-site anchoring — the springs, the *pa'ang*, and the *mandong* — and how the

*caci* and *Sanda* sit inside the day, since the *caci* is already a Ch-3 staged beat.

16. The building rites — corroboration of the *roko molas poco* forest-maiden procession and the

full *tombo pande mbaru* → *congko lokap* sequence, which is the best material in the pack for a

build-in-progress or repair state.

17. The orientation tension — *le*-facing single-house villages against ring villages facing their

own centre, and whether Erb states which rule governs when.

18. The lingko authority ladder — Tu'a Teno, Tu'a Golo, Tu'a Kilo / Tu'a Panga: their actual

division of decision rights, which is quest material as much as architecture material.

Non-Erb closers, named so the worklist is not the only route. Three sources would close items in

tiers 1 and 3 independently and neither pass could reach them: a 2025 terrestrial-laser-scan and

photogrammetry study of Wae Rebo, which states it found dimensional variation *between* houses and is

the only source likely to carry exact per-house measured geometry (it would close items 6 and 13 and

the embedment question outright); Panjaitan and Ellisa, *Interior Architecture of Vernacular Mbaru

Niang of Wae Rebo* (item 3); and the UKDW *Studi Elemen Struktur* paper on member sections and rafter

spacing. All three returned access-denied to open fetch and want an institutional route or a purchase.

---

16. Consumers

• C:\dev\Humanity\Humanity\QA\look\architecture_reference_plan.md — the game repo's

architecture lane. Sections 3, 4, 5, 6, 7 and 8 of this pack answer its per-structure reference

requests; section 4.3 answers its section-1 blockout table; section 12 proposes the

attested/not-attested dispositions; section 11 answers its section-3 rights question with the

standing ruling.

• docs/art_bibles/flores_island.md — the pointer lives at the end of that file. The bible's

fields 3 and 4 (material vocabulary; silhouette and architectural grammar) are what this pack

supplies evidence for. This pack ratifies nothing there and flips no status.

• docs/art_boards/flores_island/BRD_0001/ and the Ch-3 leaf boards — the mbaru niang, compang

and lingko rows are the plates this pack is reference for.

• docs/BOOK_ACQUISITION.md — the Erb entry. Section 15 is the worklist that runs on arrival.
• docs/spine/CH_03.md and _source/02_Tier_2_Region_Pages/flores_island.md Section 7 —

the canon this pack derives from and never overrides.

---

*Authored 2026-08-04 by the architecture-reference synthesis pass, from two independent deep-research

returns, cross-checked fact by fact. Proposal-tier reference substrate. Every load-bearing fact

carries a two-source agreement tag or a declared single-source tag; every disagreement is in section

14; every arithmetic step is marked DERIVED; the six verification findings are in section 13.*

Generated by harness/site/structure_site.py — the URL path is the repo path. review root