sumatra_java_S20.md

world/regions/sumatra_java/page/sumatra_java_S20.md

Section 20 — Session Planning and Source Guide

CANON SUBORDINATION — this document is PROPOSAL-TIER: it serves canon and never outranks it.

The canon: the CVD · the T1 foundation docs · the T0 registries (registries/) · the spine

(docs/spine/CH_*.md) · the region pages (_source/02_Tier_2_Region_Pages/). Authority order: docs/DOC_MAP.md § 0.

Canon served (scanned from this document's own citations — widen it by hand where it is thin):

T0_Source_Material_Registry

READ THAT CANON FIRST — open it and derive from it before you build anything from this document.

If this document disagrees with canon, CANON WINS and this document is the defect — fix the

document, never the canon. Nothing here is applied until it is ratified into canon.

This section is the build-gating record for the region: which processed source feeds which authoring

session, what convention each source's page anchors follow, what state the retrieval layer is in, and

what is honestly still closed. Pass-2 changed the arithmetic and it changed the routing. The page was

cut against four processed sources; the shelf now carries SIX, all six deep-read and all six audited,

and the two new ones (the Van der Tuuk collection article and Kozok's script history) land squarely in

the literacy sections rather than adding breadth. The Region Identity header was refreshed to the

full six-source roster at this assemble pass, so the header and the roster below now agree. Three of

Section 16's routing claims are corrected here on the audits' own evidence.

The processed source roster

Chunk counts are the T0_Source_Material_Registry declared values, 660 across the six. All six rows

carry ocr_method native_text, so nothing on this shelf rides an OCR-recovery path and no chunk is

quarantine-flagged. Across the six records 134 typed rows exist outside the civilizational-mark class:

97 advance, 26 are HELD by critic MAJORs, 11 were cut at the audit. By class that is 50 story seeds

(38 advance, 11 held, 1 cut), 30 weave links (14 advance, 7 held, 9 cut), 34 entity candidates (26

advance, 8 held, none cut) and 20 L3 connections (19 advance, 1 cut). A further 25 civilizational

marks carry no status field at all, and a mark cut by the audit lives only in that record's

care_findings — so any session that would consume a mark reads the findings first.

Three corrections pass-2 makes to Section 16's routing

• SRC_00152 is no longer the Section-7 sculptural anchor, and it was never a Toba lake-country

source. The audit fixed the site precisely at Pasaguan on the high bank of the Aek Tarum in Bandar

Pulau sub-district, Asahan regency, on an old coast-to-highland trail [SRC: SRC_00152, p. 1] — a

different place and a different community from Samosir, and a grep for "Samosir" across the whole

source returns nothing. Seven of its twelve rows are held, including both entity rows and every

image-bearing seed, so the record's Section-7 value is deferred until those clear. What advances is

three rows and they are Section-1 and Section-9 material: the caravan route with its documented

cargo [SRC: SRC_00152, p. 1, p. 12], the cadas welded tuff the images are cut from

[SRC: SRC_00152, p. 1], and the indigo dye [SRC: SRC_00152, p. 12].

• SRC_00226 does not carry the gondang ensembles. The audit's own search returns zero hits for

"gondang" across all 234 pages, and it struck the ambience clause from the one seed that carried it.

Section 16 credits this source with "the gondang/gordang ensembles" and that credit does not hold.

The single attested instrument row on the shelf is the kulcapi lute cognate set

[SRC: SRC_00446, PDF p. 65 = print 64], so Section 10 draws that and takes the ensemble line to

deep-research or to a real music source.

• The Kozok MUST-ACQUIRE entry is narrowed, not struck. SRC_01758 is Kozok's script history,

processed, audited and query-ready, and it carries the diacritic system, the regional letterform

variation, the script's genealogy and the dating-by-letterform method, so the substance of the

acquisition rationale is served. Section 16 already records it that way, as "Kozok, the fuller

published apparatus — LARGELY RESOLVED in substance by SRC_01758", and what still wants buying is

a securely paginated edition with the letterform plates intact, for any claim that needs a citable

plate or a stable page reference. The two pass-2 citation rows are in place as well: SRC_00228 and

SRC_01758 each carry their own Section 16 row, and both are tagged SUBSTRATE.

Per-section population plan

The plan is per-session, not per-source: a session opens one section, pulls the named rows, and stops

at the section boundary rather than harvesting a whole record. The centre of gravity on this shelf is

writing and divination, and the plan is built around treating literacy as a playable system rather

than as decoration.

• Sections 4 and 5 (people and culture, spiritual traditions and occult) — the largest block, and it

runs as several passes rather than one. SRC_00226 is the keystone: the datu's oracle ladder in its

own order, pormesa na sampulu dua, panggorda na ualu, pehu na pitu, pormamis na lima

[SRC: SRC_00226, p. 70-71]; the porhalanan divination table, which never had a computational

function [SRC: SRC_00226, print p. 77-79]; the porsili substitute effigy that draws the

sickness-spirit onto itself [SRC: SRC_00226, p. 68-69]; the manuk di ampang and the wider omen

inventory [SRC: SRC_00226, p. 73-74]; the pangulubalang bound war-champion

[SRC: SRC_00226, p. 69-72]; the bius village-cluster with its parbaringin and paniaran and no

centralised state [SRC: SRC_00226, p. 17-18]. SRC_00158 joins for the datu's working objects and

the calendar itself — the pustaha's bark-and-fold construction [SRC: SRC_00158, p. 165], the

porhalaan and pormesa zodiac table [SRC: SRC_00158, p. 170] — and its content is attributed to the

BATAK throughout regardless of its shelf label (see the provenance items below). SRC_00228 supplies

the transmission structure, the hadatuon chain of teachers [SRC: SRC_00228, p. 11], and SRC_01758

the script itself [SRC: SRC_01758 — § Ina ni Surat; § Anak ni Surat]. Two standing accuracy rules

bind this block: the four diacritic functions ship as four with their regional restrictions, since a

Karo or Pakpak puzzle is unsolvable without the schwa and the fricative, and the letter inventory

ships as a range rather than standardized to nineteen.

• The literacy system specifically (Sections 4, 5 and the Section-13 minigame surfaces) — four

language-script rows and three inscription rows are queued across this shelf and they are the

region's strongest rewards-intelligence surface, so they are planned as one system and not scattered.

The spine of it is that Batak literacy is a three-medium ecology with a real content-to-material

logic [SRC: SRC_00228, p. 11, p. 13, p. 15], that lay literacy is attested outside the ritual

specialists [SRC: SRC_00228, p. 13], that a manuscript's home can be read off the shape of a single

letter [SRC: SRC_01758 — § Silsilah Aksara; § Analisis Ina ni Surat], and that the power of an

inscribed object lay in the act of writing rather than in the content, to the point that amulets

carry erased or meaningless characters [SRC: SRC_00226, print p. 82]. That last row is what keeps a

script minigame from becoming a spelling test.

• Sections 6 and 9 (weapons, minerals and trade) — SRC_00446 for the blade and the chemistry: the

piso sanalenggam [SRC: SRC_00446, PDF p. 67 = print 66] and the gunpowder recipe

[SRC: SRC_00446, PDF p. 69-70 = print 68-69], with two corrections that must land before either is

specified. The blades are made by welding together thin strips of steel, which is the source's own

wording and is not the same claim as pattern-welding, and it is documented for late-nineteenth and

early-twentieth-century examples only, so a 1200 CE build does not inherit it as ancient. The

gunpowder additive is citrus, not calcium — a science-grounding row implying quicklime would teach

false chemistry. SRC_00226 adds the protective-inscription and amulet layer

[SRC: SRC_00226, print p. 82-83], and SRC_00152's caravan route is the Section-9 trade spine.

• Section 9 and the caravan seed carry their own documented history. The paragraph that gives the

route its economic substrate — the 600-perahu port, the rattans, salt-fish, horses and silk cloths —

also records that enslaved people were brought down from the interior in considerable numbers

[SRC: SRC_00152, p. 12]. That ships. The audit ruled it explicitly and in the anti-conservative

direction: the caravan the player runs moved people as well as salt, and the honest telling is the

correct one. It is routed through the institution and its real causes — debt-bondage, captive-taking

and elite brokerage at both ends of the route under real economic pressure — with the story

antagonist positioned INSIDE that apparatus and never substituted for it. Concentrating a

documented regional economy into one corrupt broker is the scrub, not the safeguard, and highland

communities stay agents inside the network rather than only its cargo. The player's participation or

subversion is a playable integrity test.

• Section 16 and the colonial record — the same source's page 12 carries Anderson's 1823 slur, and

the audit ruled it attributed-and-contradicted rather than authorial-only. A player who reads the

colonial record meets the slur AND meets its refutation on that same page: Batak cotton grown in the

country, woven at home, dyed with home-grown indigo, coloured with forest woods, and traded through

a 600-perahu port [SRC: SRC_00152, p. 12]. It is never repeated as fact and never presented as a

player-facing truth about the community. Deleting it would protect nobody and would hide how the

defamation was manufactured.

• Sections 3 and 7 (fauna, architecture) — the thinnest sections against this shelf. SRC_00446 is the

only architecture source that advances anything, and its house-facade material is one seed

[SRC: SRC_00446, PDF p. 40, 46, 24 = print 39, 45, 23]; SRC_00152's image material is held. Fauna

has no processed anchor at all beyond one cosmogonic entity, so the Khan-bond tiger loop in the Zone

Catalog draws on nothing in this corpus and is a deep-research gap, exactly as the Section-3

tracking loop was on the neighbouring page.

• Sections 13 and 14 (quests, trades) — the seed layer, and it is split across TWO queue files rather

than one (below). Trade-4 draws the pustaha and pagar craft loops from SRC_00158 and the bark-book

craft from SRC_00226 [SRC: SRC_00226, p. 65-67]; the divination minigames draw SRC_00226 and

SRC_00158; the script-literacy acquisition minigame draws SRC_00228 and SRC_01758.

• Section 15 (cross-regional) — the weave lane is the most heavily cut class on this shelf: of 30

links, 14 advance, 7 are held and 9 were cut, most for failing the requirement that both ends be

named AND cited. Section 15 draws the fourteen advanced links and nothing else. Two of the held

rows are held on named corrections and neither enters the section until it re-audits: SRC_00158/W02

holds until its Indic parent is restated as Pallava (South India), which is not the Tamil script,

and SRC_01758/W07 holds until its highland-Sumatran end is re-originated onto the scripts the

source actually attests rather than onto unattested casualties. The Pallava parentage itself is

available to the section without either of them, on an advanced row

[SRC: SRC_01758 — § Aksara batak dan sejarahnya; § Silsilah Aksara].

• Sections 18 and 19 (fluctuation strata, modern revisit) — the colonial-provenance and

vanishing-knowledge material lands here rather than in the 1200 CE present. Van der Tuuk was

employed by the Dutch Bible Society to translate scripture and to describe the language for the

instruction of Protestant missionaries [SRC: SRC_00228, p. 10], and that project carried 152

pustaha [SRC: SRC_00228, p. 11] and the 25 volumes he assembled at Barus and had bound in the

Netherlands [SRC: SRC_00228, p. 14] into Leiden University Library, while the same collection saved

an important part of the Dairi heritage from oblivion [SRC: SRC_00228, p. 19]. Both are true and the

build carries both. The causes of the diminishment are plural and named — Christianity, Islam,

modernization and globalisation, plus a dwindling number of scholars who can read the manuscripts

[SRC: SRC_00228, p. 10, p. 19] — and single-cause routing through a colonial apparatus erases the

others and, worse, erases Batak agency. Named Batak actors carry the strand: the datu who taught Van

der Tuuk to interpret pustaha and whom he employed to copy them [SRC: SRC_00228, p. 11], the Kelasen

storytellers whose repertoire was written down for the first time, and Ompu ni Manungkol Langit of

Aek na Uli, who would not part with his pustaha until he was paid one European dog as large as a

calf, a dozen flasks of jenever, ten Spanish mats and three padang rusak [SRC: SRC_00228, p. 19]. A

Batak owner driving a hard bargain with a Dutch collector is the beat, not a community things were

done to.

Anchor conventions, per source

Anchor conventions differ per source and are not interchangeable. The neighbouring page shipped anchor

drift in two of four records; this shelf's audits re-derived every offset from the printed folios

rather than assuming, and the conventions below are the ones this page cites.

One more anchor-adjacent hazard sits upstream of SRC_00158 in particular. The RAG chunking of that

source splices footnote text into body text at page boundaries, most consequentially at its print

166/167 boundary, where footnote 6's continuation runs into a body sentence and produces a confidently

wrong claim about a real people's institutions — a hereditary-profession sentence that finishes with

another footnote's tribe list. This shelf's records did not fall into it; any regeneration or

downstream extraction over those chunks can.

Processing and query-readiness

• All six sources are processed and query-ready in the book_substrate collection with region_tags

sumatra_batak, and the query layer expands the page's own slug at query time: python

harness/rag/query.py "<question>" --collection book_substrate --region sumatra_java. Region

filtering is not optional here. Semantic-only retrieval over a multi-region corpus ranks by topic

and not by place, which is why an unfiltered Borobudur query once ranked a Himalayan pilgrimage

source.

• The expansion is one-to-MANY, unlike the neighbouring pages' one-to-one maps: sumatra_java resolves

to sumatra_batak AND malay_world. The malay_world shelf carries one source, a construction-sacrifice

study at 50 chunks, which is deep-read but NOT audited and was not part of pass-2. A region-filtered

query on this page can therefore return unaudited material, and a session that treats every hit as

audited will be wrong wherever a hit comes off that shelf. Check the SRC id against the roster above

before consuming a retrieval result.

• The registry lags the audit on every row. All six read extraction_status deep_read_extracted while

all six EXR records read deep_read_audited, so any consumer gating on the registry field will

under-read this shelf's actual state. The registry titles are filename stubs rather than

bibliographic titles and source_author and source_year are null on all six, so a title lookup will

not identify any of them — the roster table above carries the correct citations. Three records also

carry an empty chapter_scope_hints in the EXR file, SRC_00152, SRC_00446 and SRC_01758, and one of

those three, SRC_00446, is empty in the registry as well; all three are demonstrably CH_05 material.

• A duplicate sits on the shelf. SRC_01914 is a second conversion of the same Kozok work, 27 chunks,

with no extraction_status at all. It is not in the roster above and nothing was harvested from it,

but it is in the store, so a script query retrieves the same passage twice under two ids and doubles

its apparent corpus weight. De-duplication is owed to the shelf owner before the next index pass.

Reading the harvest without consuming what is held

The two consumption manifests at docs/proposals/region_pass2/sumatra-batak_CONSUMPTION_MANIFEST.md and

docs/proposals/region_pass2/sumatra_java_CONSUMPTION_MANIFEST.md are script-emitted from the records'

typed status fields and together are the only work-list. Both are opened, because this page's harvest

is split across two shelf keys.

• The derived views are convenient and none of them are filtered against holds. The seed queue is

likewise split in two, docs/proposals/quest_seed_queue/sumatra-batak.md with 41 seeds and

docs/proposals/quest_seed_queue/sumatra_java.md with 8, and between them they list 49 of the shelf's

50 seeds. The view drops the one rejected seed and shows all eleven HELD seeds inline with no marker

distinguishing them. This is not hypothetical: the first entry in the larger queue is a held seed

and it still carries verbatim the buffalo-horn detail the audit proved absent from its source, where

the source says Malay-style headdress three separate times. A Section-13 session that opens the

queue first and the manifest second will consume it.

• Regenerate, never hand-edit, either file.
• The quest_type position has changed and the queue view has not caught up. The §6.5 quest_type enum

IS ratified ACTIVE T1 canon with eighteen members and is ruled by spine role, and the audit note on

all six records now says so. The queue view's TO-FILL line still advertises five legal values, so a

§7.5 typing session that trusts the view will under-read the vocabulary by thirteen members. The

extension to twenty-one is a live ruling in the parallel lane; per-seed typing belongs to the §7.5

step and nothing on this shelf invents a value.

• Twenty-two mint flags stand across the six records and their reliability is uneven in a specific,

checkable way. The reason most of them were raised — that no ratified quest_type vocabulary exists —

is retracted, and the records record that retraction at two different depths. Three carry the repair

script's typed premise_corrected stamp, the three whose own prose its fixed-string marker matched:

SRC_00152/S02, SRC_00226/S05 and SRC_01758/S03. Seventeen in total, those three included, carry a

typed mint_reevaluation_2026_07_22b field recording the premise as FALSE and the blocker as CLEARED

— seven of them hand-read and resolved against the live enum, ten cleared with the typing itself

deferred to §7.5 at class level. Five carry no correction at all, and only two of those five are

quest_type flags: SRC_00228/S03 and S04, whose stated reason still reads that the quest_type mint is

HELD with nothing attached to qualify it. The other three, SRC_00152/EC_01, EC_02 and W01, were

never quest_type gaps — an entity_kind for an uninscribed carved memorial marker, an entity_kind for

a community as distinct from a persona, and a note recording that no mint is required at all. So a

§7.5 session reads the resolution field on the seventeen rather than the flag prose above it, and

re-evaluates SRC_00228/S03 and S04 against the live enum before treating either as evidence for a

category mint.

Open provenance items

• The SRC_00158 shelving split. That record is authored primary_region sumatra_java and region_tags

sumatra_java while its shelf reads sumatra-batak and its registry region_tags read sumatra_batak,

and it is the sole reason this page's harvest carries two manifests and two seed queues. A pustaha

is a Batak divination manuscript and the source is an article about Batak manuscript culture, so the

CONTENT is Batak and is attributed to the Batak everywhere it lands, whatever the shelf label says.

The split key is the open item: it is a taxonomy question for the shelf owner under the

substrate-region ruling, which expands at query time and never bakes the slug, and until it is

resolved every session reads both manifests. Nothing about the attribution waits on that resolution.

• The colonial-provenance strand is the shelf's largest owed ADDITION, and no row carries it. Two

independent audits found the same gap from opposite ends. SRC_00158 is an article about seven

pustaha held in four Polish institutions, framing the manuscripts as objects that function in

parallel in Western and Eastern societies and recording that from colonial times they were often

brought to Europe as souvenirs, with roughly 2,000 now in worldwide collections

[SRC: SRC_00158, p. 164-165] and van der Tuuk named at [SRC: SRC_00158, p. 169] — yet not one of its

twenty-three rows touches the fact that the datu's sacral books are in European museums. SRC_00226

tells the same story about its own collection with unusual candour: it became the goal of the

mission civilisatrice to annul the datu's role in society by taking objects out of circulation

[SRC: SRC_00226, print p. 17], the mission doctor who studied the language also collected the

artefacts [SRC: SRC_00226, print p. 11], and one object was acquired around 1910 and sold to the

Hamburg ethnological museum in 1912 [SRC: SRC_00226, print p. 81]. Suppression, extraction, sale,

display — a complete documented chain, and the honest answer to why the player can read these

manuscripts at all. It is queued as an addition on the next extraction pass, with the antagonist

being the collecting and mission apparatus and Batak agency SHOWN rather than routed around.

• Care tiers read five elevated and one standard, and one of the five is stale. The SRC_01758 audit

ruled the tier down to standard on the merits — the source is a public outreach essay by the leading

philologist of Batak manuscripts, written to enable students to read original manuscripts, with no

restricted ritual content — but care_tier is not an audit-owned key, so the field still reads

elevated and the ruling is advisory until a separate record edit lands. The narrow carve-out that

survives it is real and binds every session on this shelf: pustaha form, materials, script and

provenance forensics ship at full specificity, while actual divination and medicine formulae are

referenced and not reproduced.

• Two entity rows cannot become valid registry entries as written and are held for it: one bundles two

distinct deities into a single row, and one bundles six peoples' scripts into a single script entity

that the source itself distinguishes. Both clear by splitting into per-being and per-people rows at

an extraction pass, never by deletion.

Build-ready gating

The machine layer is build-ready now against the v3.1 spine: the Zone Catalog rows copy the CH_05 site

rows verbatim and add only page-side geometry, and the Derived Indexes resolve as live joins into the

typed layer. Nothing in the six processed records touches the antagonist apparatus — they are a

manuscript catalogue, an art catalogue, a paleography, two manuscript-culture articles and a field note

— so the chapter's institutional-strand is spine-authored throughout and needs no source session.

The headline gate did not move, and pass-2 made it starker rather than smaller. All six sources are

Batak and North Sumatran. The Java half of this page has ZERO processed substrate: Borobudur,

Prambanan, Merapi and the Palembang/Srivijaya port have no source on this shelf, and two of those

zones, borobudur and palembang_port, carry P1 build priority, with palembang_port gating the chapter's

lead SQ4 investigation. Pass-2 roughly doubled the Batak depth and moved the Java depth not at all.

Three gates stay open:

• The Java-monument acquisition. Miksic, Borobudur: Golden Tales of the Buddhas, for the three-level

program and the relief narratives; Klokke, The Tantri Reliefs on Ancient Javanese Candi, and Hadi

Sidomulyo, Reading Old Javanese Records, for the Old-Javanese inscription depth Thread 17 wants.

Sections 1, 5, 7 and the Borobudur beats wait on it, and it cannot be back-filled from the Batak

shelf without a false citation. The one cross-island bridge the corpus does supply honestly is

paleographic: the Batak and Javanese-Balinese scripts share a Kawi-Pallava ancestor, and the Batak

Ja is Da plus a horizontal stroke, which holds in Kawi but not in Pallava

[SRC: SRC_01758 — § Analisis Ina ni Surat]. That is real epigraphic reasoning a player can perform

at the monument, and it is the strongest Java-facing material on the shelf.

• The Srivijaya maritime gate. Coedes, The Indianized States of Southeast Asia, for the thalassocracy

and the post-Chola decline that the whole Palembang half of the chapter runs on. Default path is

acquisition rather than deep-research, because Section 8's politics and Section 18's dating both

turn on it.

• The cosmology and creature gate, which is narrower than Section 16 currently states. Tobing on the

Toba-Batak high god and Loeb's Sumatra remain open, but for a specific reason: the begu ganjang is

attested in NEITHER of the two new sources — both audits searched and both returned zero — and both

routed its citation onward to SRC_00226 and SRC_00158 rather than supplying it. Until that routing

is verified in those two records, the creature strand rests on canon-side authority and not on this

corpus. The Naga Padoha material rides held rows for the same reason. Deep-research is the default

path for both; the Chesner and Rose Toba VEI-8 geology remains a straightforward deep-research gap,

with the caldera's own outline already carried [SRC: SRC_00226, print p. 11].

Session sequencing follows the cost curve and the anchor risk. The three small records close first —

SRC_00152 at 15 chunks, SRC_00228 at 21, SRC_01758 at 27, sixty-three chunks between them — and they

carry the Section-9 caravan strand, the Section-18 provenance strand and the literacy spine, which are

the three highest-value strands on the shelf. SRC_00158 at 22 chunks closes with them, with its

attribution rule and its footnote-splice caveat applied at the same pass. SRC_00446 follows at 157

chunks with the material-culture sections and its PDF-index relabel. The keystone catalogue drives the

largest block last, Sections 4, 5, 6, 9, 13, 14 and 15 at 418 chunks, run as several passes, with its

mixed-anchor convention checked per citation rather than per record. The colonial-provenance extraction

pass and the two entity splits run independently of all of it, and the three acquisition gates run

independently of everything. The v1.0 book-cost fields are superseded by the closed acquisition ledger.

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