GeoForge

---

GeoForge is a reproducible GeoIP database compiler that builds a local MaxMind-compatible MMDB from multiple independent geolocation sources.

The pipeline combines DB-IP Lite, MaxMind GeoLite2, IP2Location LITE, Sypex Geo, operator geofeeds, RIR delegated statistics, and GeoNames reference datasets into a normalized consensus-based geolocation layer.

Primary outputs:

• release/geo.mmdb
• release/geo.csv
• release/geo-quality-report.txt

---

Overview

Free GeoIP datasets typically optimize for broad coverage, not cross-source validation.

GeoForge approaches geolocation as a consensus problem:

prefix seed
-> multi-source candidate collection
-> confidence scoring
-> normalization
-> conflict resolution
-> reproducible MMDB output

The builder merges independent signals, downranks inconsistent records, applies conservative normalization rules, and produces an auditable local database suitable for gateways, analytics, enrichment services, fraud systems, and infrastructure tooling.

The project is designed for:

• offline local lookups
• deterministic rebuilds
• source transparency
• quality regression tracking
• operational GeoIP enrichment

---

Architecture

 Source Databases
│
┌──────────────────┼──────────────────┐
│ │ │
▼ ▼ ▼
GeoLite2 IP2Location Sypex Geo
│ │ │
└──────────────┬───┴──────────────────┘
▼
Consensus Engine
scoring / merge / weighting
▼
GeoNames Enrichment
postal / city normalization
▼
Output Cleanup
timezone / precision / QA
▼
MMDB Export

---

Repository Layout

cmd/
├── builder/ Main database compiler
└── qualitycheck/ Post-build validation
internal/
├── consensus/ Merge and scoring logic
├── geofeed/ RFC 8805 parser/index
├── geozip/ GeoNames enrichment
├── output/ Final normalization
├── refdata/ Country/currency metadata
├── rirstats/ RIR delegated statistics
└── strnorm/ String normalization
data/
release/
scripts/

Downloaded source datasets and generated outputs are intentionally gitignored.

---

Data Sources

Source	Role
DB-IP Lite	Required prefix seed
MaxMind GeoLite2 City	Consensus baseline
IP2Location LITE DB5	Independent geo signal
Sypex Geo City	CIS-focused enrichment
RFC 8805 geofeeds	Operator-published corrections
RIR delegated stats	Registry country attribution
GeoNames postal dump	Postal enrichment
GeoNames cities1000	City geoname resolution

GeoForge separates source collection from output generation, allowing partial builds when some datasets are unavailable.

---

Build

Create a local environment file:

cp admin.env.example admin.env

Run the pipeline:

./geo.sh

The build workflow:

1. Acquires a release lock
2. Downloads updated datasets
3. Detects source changes
4. Runs Go tests
5. Compiles builders
6. Generates MMDB + CSV outputs
7. Runs quality validation

Force rebuild:

FORCE_BUILD=1 ./geo.sh

Disable downloads:

AUTO_DOWNLOAD=0 ./geo.sh

---

Outputs

File	Description
release/geo.mmdb	MaxMind-compatible GeoIP database
release/geo.csv	CSV audit/export copy
release/geo-quality-report.txt	Post-build quality analysis
release/geo.previous.csv	Previous snapshot for diffing

---

Record Schema

Each MMDB entry contains top-level metadata plus a nested location object.

Top-Level Fields

Field	Description
matched_prefix	CIDR written into MMDB
confidence	Consensus confidence score
source_updated_at	UTC build timestamp
country_metadata	Country/currency/calling metadata
location	Final geolocation object

Location Fields

Field	Description
continent_code	Continent code
country_code	ISO country
registry_country_code	RIR-derived registry country
subdivision_name	Normalized admin region
city_geoname_id	GeoNames city identifier
city_name	Normalized city
postal_code	Consensus postal code
latitude	Rounded latitude
longitude	Rounded longitude
time_zone	Derived timezone
accuracy_radius_km	Conservative accuracy estimate

---

Example Record

{
"ip": "1.208.10.20",
"matched_prefix": "1.208.0.0/12",
"confidence": 85,
"source_updated_at": "2026-05-20T00:00:00Z",
"location": {
"country_code": "KR",
"country_name": "South Korea",
"city_name": "Seoul",
"postal_code": "04524",
"latitude": 37.56631,
"longitude": 126.9772,
"time_zone": "Asia/Seoul",
"accuracy_radius_km": 20
}
}

---

Quality Model

GeoForge is designed to improve operational quality through source consensus rather than raw source replacement.

Expected improvements over single-source lite datasets:

• better country stability
• improved city consistency
• stronger CIS coverage
• cleaner normalization
• more conservative precision signaling
• reduced malformed text artifacts

The builder intentionally favors stable consensus over aggressive precision claims.

---

Quality Gate

After each build, the pipeline runs a post-build validation stage and generates:

release/geo-quality-report.txt

Validation includes:

• coverage statistics
• confidence distribution
• added/removed prefixes
• country/city regressions
• mojibake detection
• MMDB smoke lookups

Enable strict mode:

QUALITY_STRICT=1 ./geo.sh

Strict mode fails the build on large regressions or suspicious output anomalies.

---

Normalization

Final records are normalized immediately before export.

Normalization includes:

• coordinate rounding
• mojibake repair
• subdivision cleanup
• duplicate collapse
• timezone derivation
• conservative multilingual cleanup

Normalization intentionally avoids broad transliteration or aggressive geopolitical rewriting.

---

Geofeeds

Allowlisted RFC 8805 feeds are configured in:

data/geofeeds/allowlist.tsv

Supported formats:

prefix,country,region,city
prefix,country,region,city,postal

Default IPv4 floor:

GEOFEED_MAX_IPV4_BITS=24

This prevents excessive host-level fragmentation from narrow geofeed entries.

---

Update Semantics

The downloader uses content hashing and atomic replacement semantics.

Tracked state files:

data/download-state.tsv
data/download-changed.txt

If no source changed, the builder preserves the existing MMDB unless forced.

---

Use Cases

Domain	Example
Fraud Detection	Geo consistency checks
SIEM Enrichment	Country/city attribution
Analytics	Geographic aggregation
Gateways	Local GeoIP lookups
Data Pipelines	IP enrichment
Infrastructure	Region-aware routing

---

Operational Notes

• City-level geolocation remains probabilistic
• Mobile and VPN accuracy may vary substantially
• Prefix coverage depends on DB-IP Lite seed availability
• Postal enrichment should be treated as opportunistic
• Geofeeds improve operator-owned allocations but are uneven globally

---

Publication

The repository is designed so code can be published independently from downloaded datasets.

Do not commit:

• admin.env
• downloaded provider databases
• generated release artifacts
• API credentials or license tokens

Review THIRD_PARTY_DATA.md before redistributing derived outputs.

---

Roadmap

Planned additions:

• ASN-aware geo heuristics
• confidence-weighted source tuning
• regional regression dashboards
• IPv6 quality scoring
• compressed bulk exports
• build reproducibility attestations

---

License

See LICENSE.

Additional redistribution guidance:

THIRD_PARTY_DATA.md

---

Disclaimer

GeoForge aggregates third-party geolocation datasets into derived operational outputs. IP geolocation should be treated as probabilistic infrastructure metadata, not physical-user attribution.