烽 燧 · B O N G S U

We do not predict.
We report what is visible now.

BONGSU links a firm's assets to geopolitical, disaster and supply-chain exposure as connected objects, seals every judgment with a hash, and scores it when the verification date arrives. Like the beacon towers of Joseon Korea, we only signal what we have seen.

Request a paid pilot (PoC) Open the console demo

The console is currently in Korean. An English console is scheduled after the first enterprise contract.

12 · 9

12 object types · 9 relation types. One ontology on a single PostgreSQL

3

Operating models — DPRK provocation hazard, Mt. Paektu eruption, market impact. Every figure is read from result files

60

Empirical papers. 810 classes · 898 relations linked as knowledge objects

Sealed

Every judgment is locked with its timestamp, confidence and evidence, then scored after the fact

Why BONGSU

The problem is not too little information. It is silos.

Palantir never competed on better forecasts. It sold connection, not ownership of data. BONGSU does the same. We sell no forecasting product, because our own out-of-sample tests showed none that beats the mean.

We do not sell "our forecasts are more accurate."Brokerage research already distributes top-tier forecasts for free. Selling something better than free is not a business.
"Built from data only we can see, scored, signed, with our own money on the line."BONGSU's sales proposition. An object database and a scoring record cannot be bought with capital; only time builds them.
The value of information falls with the breadth of its distribution.So there is no paid individual subscription. Enterprise contracts only.
Proof of scoring
12/18
Judgments from 1997 & 2002, scored in 2026

We scored judgments written 24 and 29 years ago

Our founder's 1997 undergraduate thesis on Kim Jong-il's succession made 9 forecasts: 6 hits, 2 partial, 1 unscorable (a conditional). His 2002 master's paper on exchange consolidation made 9 more: 6 hits, 2 partial, 1 miss — the claim that "transferring index futures is unrealistic" was proven wrong in 2004.

We did not erase the miss. That is what BONGSU sells. See the score sheet →

What we do

Connect, compute, seal, score

Four steps on one screen. We do not sell any one of them separately.

① Object model

Silos into one object database

Asset, entity, component, facility, route, region, event, indicator, alert, judgment, decision, source. Values are never overwritten; they accumulate in bitemporal form. Merges can be undone.

② Exposure map

Deterministic arithmetic, no probability theatre

Holding weight × dependency × share transiting an axis. You see, in numbers, which assets hang on the Taiwan Strait, the Middle East or North Korea.

③ Alerts

Only declared rules fire

Threshold, direction, duration and cooldown are recorded with their declaration time. Rules cannot be edited afterwards to look right; they can only be retired.

④ Sealed judgment

Our only moat

Each judgment is locked with SHA-256 alongside its time, confidence and evidence objects. The database rejects updates and deletes. When the verification date comes, a score is written.

Provenance

Every attribute carries its source, collection time, reliability grade (A–D) and licence. "As of" queries show what we knew at the time.

Three-axis permissions

Object type, attribute and instance. Client data stays inside the client's boundary; only object definitions and queries cross it.

LLMs draft only

Summaries, translation, extraction. Never judgment. A sentence produced by a model cannot leave without its evidence objects attached.

Object model v2.1

12 object types · 9 relation types

No Kafka, Spark, Neo4j or vector store. One PostgreSQL, a daily batch, and an API first. A graph database is reconsidered when nodes exceed one million.

ASSETAssetWhat the client owns or operates. Where exposure starts
ENTITYEntityFirms, financial institutions, agencies. Supply-chain participants
COMPONENTComponentWhere a supply-chain break actually passes through
FACILITYFacilityPlants, ports, power stations, data centres
ROUTERouteStraits where a blockade equals a cut
REGIONRegionWhere disasters and geopolitical events attach
EVENTEventThings that actually happened. Observed facts only
INDICATORIndicatorGPR, seismicity, transit volume, prices
ALERTAlertIssued by us. A judgment, not an event
JUDGMENTJudgmentA human conclusion. The unit of sealing
DECISIONDecisionWhat was actually done on a judgment
SOURCESourceWhere every value came from. The basis of trust
holds entity→assetdepends on asset→component·route·facilitysupplies entity→entity transits shipment→routelocated in facility→regionaffects event→asset·facility·route triggers indicator→alertbased on alert·judgment→indicator·eventscored by judgment→outcome
Model registry v1.0

Three operating models — figures read from result files only

No number on this page is typed by hand. Each model runs on data, writes a JSON result, and these figures are copied from that file automatically.

BS-M01

DPRK local provocation hazard

79%≥1 event within 3 years

Poisson + Weibull recurrence · 15 events / 30.7 years · last event 2024-10-15

Shape k≈1: indistinguishable from a memoryless process. We know "how often", not "when".
BS-M02

Mt. Paektu eruption probability

15.3%within 30 years (analogue conditional)

1,038 analogue intervals, Weibull · VEI5+ in 10 yrs 0.58% · 1 event detected within 50 km in 35.6 years

The global network saw one event within 50 km in 35.6 years. The monitoring gap matters before the probability does.
BS-M03

Market impact of geopolitical events

+1.47%KOSPI 5 days after DPRK events (t=2.86)

Event study + local projection · 440 monthly · 15,219 daily obs · DPRK 27 · Taiwan Strait 7 events

Out-of-sample R² for returns is negative. Not a forecast — a lens for unpriced risk.

Candidates: a Korean-language geopolitical index · public-project feasibility · DPRK/China trade risk · corporate default. Nothing without out-of-sample evidence enters the registry.

Knowledge library

Sam-Man's Infoshop — 60 papers connected as objects

Every empirical paper gets an ontology: classes, relations, constraints and competency questions. Papers sharing classes are linked automatically and grouped by theme into books. Information is reproduced and expanded.

The information business that began thirty years ago under the name "Infoshop" continues as BONGSU.

810 classes · 898 relations · 673 competency questions · 565 constraints. The most shared class is Scenario (20 papers).

Papers are searchable by theme, class and question in the console's Infoshop tab. Full texts are provided to contract clients only.

Open the Infoshop
Plans

Enterprise contracts only — no paid individual subscription

Annual fees depend on organisation size and integration depth. Every contract starts with a paid pilot (PoC), whose fee is credited against the full contract. Prices in Korean won, VAT excluded.

Small

₩15M / year
  • Exposure map on 3 axes (public data)
  • Indicators, events, alerts
  • Results of 3 models
  • Full-text library · 3 seats
  • Read API

Medium

₩30M / year
  • Everything in Small
  • Your asset list as objects (originals stay inside your boundary)
  • Client-specific exposure and alert rules
  • Shared sealed judgments and scores
  • 10 seats · quarterly briefing

Large

₩60M / year
  • Everything in Medium
  • Internal data integration (suppliers, components, contracts)
  • Entity resolution · reversible merges
  • Object- and attribute-level permissions · audit trail
  • Unlimited seats · monthly briefing · advisory

Individual membership — free

Solo founders, students, professors and professionals can register. Members see scored judgments (starting with 18 from the 1997 and 2002 papers) and paper summaries with ontologies. Live alerts, exposure maps and pending judgments are for contract clients only. Registration data is used for demand research.

Register

What each tier can access

Individuals have one free tier; enterprises have Pilot, Small, Medium and Large. Data is classed Public (C0), Member (C1), Contract (C2), Customer-only (C3) and Internal (C4). Access is enforced in the database, not in the screen. When a contract ends, the organisation drops to the individual tier automatically.

Paid pilot (PoC): 8 weeks, ₩5M, up to 20 client assets, producing a draft exposure map. Regardless of outcome, one judgment is sealed for you. Prices as of September 2026.

Roadmap

Every stage has a pass condition and a failure path

We declare the condition for the next stage before starting. If it is not met, we do not proceed.

Stage 0 · now
Demand measurement

Interviews with 20 prospects. Proceed if 3+ show intent for a paid pilot; stop if none.

Stage 1 · 9 mo
Object model & MVP

Exposure map on 3 axes from public data. One paying pilot client.

Stage 2 · 6 mo
Provenance & permissions

Source and grade on every value. Pass one client security review.

Stage 3 · 12 mo
Client data integration

Originals stay inside the client. 3 integrations, 1 renewal.

Stage 4 · 12 mo
Public scoring

First published score sheet from sealed records. This becomes the moat.

Stage 5
Public sector & overseas

Procurement pricing and payback verified first. 60% renewals.

Company

SSGlobal Inc. · brand BONGSU

Legal name
SSGlobal Inc. (에스에스글로벌 주식회사)
Brand
BONGSU (烽燧) — Joseon Korea's beacon network, which did not predict but relayed what was visible now
CEO
Sam-Man Cho
Business reg.
866-88-01300
Head office
180 Ujeon-ro, Wansan-gu, Jeonju, Jeonbuk, Republic of Korea
Activities
Information services · software · systems integration · management advisory · publishing
Sam-Man Cho

Sam-Man Cho, CEO

Formerly Head of Alternative Investments at Hyundai Asset Management after careers in securities and asset management (asset management). He writes under the pen name Cho Bong-su. In 2026 he designed ontologies for 60 empirical papers. Judo 4th dan; runs for a hobby.

An interest in national security — he prepared for the intelligence-agency examination after studying political science — met twenty years of asset-management practice and became BONGSU. The goal is not to hold more information but to put judgments on record and be scored.

Thirty years of preparation — from a 1998 résumé to BONGSU

1993–95Army service on the DMZ front (Cheorwon GOP, 26 months). Seeing the border first-hand
1995–97Five prize-winning unification papers, including "Kim Jong-il's Consolidation of Party Power and Regime Outlook" (1997, top prize) — the ancestor of 2026 Paper 20 and model BS-M01 — and "Improving the Ministry of Unification Website" (national top prize)
1997Built the website of the Local Autonomy Research Institute at Chonbuk National University; ran a BBS internet community — his first information platform
1999Graduated in political science; prepared for the National Security Planning Agency (now NIS) examination
2002Master's paper on exchange consolidation and a "Korea Exchange" holding company; 6 of 9 forecasts confirmed by KRX (2005) and the alternative trading system launch (2025)
2000s–Top sales performer at Hyundai Securities (cover of the in-house magazine) → asset management → Head of Alternative Investments, Hyundai Asset Management
2026Ontologies for 60 papers · security trilogy (geopolitics, Kim Jong-un, Mt. Paektu) · BONGSU founded
Contact

Paid pilot · contracts · data requests

Tell us your organisation and the axis you care about. We reply within two business days. We do not take individual subscription enquiries.

Your details are used only to respond to this enquiry, under our privacy policy (Korean).