1.1 The problem

An address tells a stranger where to go. In Cameroon most homes have none that a stranger can use. The Universal Postal Union’s addressing sheet for the country has no postcode line: its model address is a name, “BP 6000” and “YAOUNDÉ” [1]. Mail goes to post-office boxes (boîtes postales, BP), which identify a counter, not a dwelling; the national daily put it plainly in 2024: the post-office box “is not considered a real address” [2]. In November 2024 the press reported that the government had created postcodes covering the whole territory, assigned per commune and issued through the Campost–FindMe application [3], [4]. Two years later no list, format or example of these codes is public.

Yaoundé and Douala began coding their streets and numbering their doors in 1992 under the World Bank’s second Urban Development Project [5], [6]. The work produced the coded street names still seen in the capital (“Rue 1.014”) and metric door numbers in parts of both cities, but in 2019 fewer than 15 % of Douala’s roads carried a name [7], and Njoh describes the national effort as confined to the two largest cities and to naming rather than physical addressing [8]. Elsewhere, and for most homes even in those cities, directions are given by quarter and landmark: “Bonapriso, behind the bakery”.

The consequences are ordinary and costly. A delivery or a moto-taxi ride ends with a telephone call; an ambulance reaches the quarter and then searches for the gate; a bank, an employer or a school asks for an address that cannot be written down; a family abroad sends money and goods to a description. One compound behind one gate holds several households and rented rooms, and in the villages of the forest and of the north, houses stand along tracks with no name.

Conventional street addressing names streets, numbers doorways and maintains an address directory [5]. It is effective where it is completed and maintained, but it is slow and depends on signs that must be installed and kept. Its average cost in a World Bank review of 21 African cities was US$5.7 per addressed doorway at 2005 prices [5].

1.2 Contribution

This paper presents a national addressing model for Cameroon and the registry built from it. Its contributions are:

1. A code people can say and remember. A code is the name of a place people already use and a number of at most five digits, read in pairs like a telephone number: Bonapriso 14 01.
2. National uniqueness without context. Each name owns one national range of numbers, cut into frozen blocks, one per area that carries the name, so that a code identifies one compound in the whole country without a town, division or device position.
3. Locality and growth. Numbers follow a Hilbert curve over each area, leave gaps for new homes and keep a reserve, so that neighbours have neighbouring numbers and new compounds can be numbered between them.
4. Integrity for systems. A machine form carries a Damm check digit over the name, the arrondissement, the area and the number, which detects every single-digit error and every transposition of adjacent digits.
5. Respect for official addresses. Where a council’s coded street passes the compound, it is shown first; council door numbers and Campost’s codes fill reserved fields the day they are shared or published.
6. Complete, open coverage. All 7,243,272 compounds identified in the country received a code on 21 October 2026. The specification, the reference engine, the registry and a ready-to-deploy server are published under open licences [9].
7. Evaluation and impact. We measure the properties of the issued codes and estimate, under stated assumptions, their effect for citizens, logistics, emergency services and government.

1.3 Structure of the paper

Section 2 reviews Cameroon’s existing postal, addressing and land systems and international practice. Section 3 describes the data. Section 4 sets out the model rule by rule, with a worked example on a real code for each formula. Section 5 describes the implementation and Section 6 the evaluation. Section 7 reports the designs we considered and rejected, with measurements. Sections 8 to 11 estimate the benefits for citizens, logistics and commerce, emergency services, and government. Section 12 discusses privacy, safety and governance, Section 13 the limitations, and Section 14 concludes.

2.1 Territory, population and languages

Cameroon has ten regions, 58 divisions (départements) and 360 subdivisions (arrondissements), each the territory of one council (commune); fourteen city councils (communautés urbaines) group the numbered councils of the larger towns, such as Douala I to VI and Yaoundé I to VII. Law no. 2019/024 of 24 December 2019, the General Code of Regional and Local Authorities, gives the English-speaking North-West and South-West regions a special status [10]. French and English are the official languages, and more than 250 languages are spoken; Cameroonian Pidgin English, Fulfulde, Ewondo and Duala serve as common languages in their regions [11]. The last published census, the third, counted 17,463,836 inhabitants in 2005; the fourth completed its fieldwork on 15 September 2026 and has not yet published results, while the census bureau’s projection before enumeration exceeded 28.8 million [12].

2.2 Post and the post-office box

The UPU sheet for Cameroon gives a model address with a box number and a town and no postcode [1]. Campost and the private company FindMe signed a partnership on 1 November 2024; FindMe had operated in Cameroon since July 2024, and its address takes the form “occupancy number, street name, postal code, city” [3]. The press reported that Cameroon “had no postal codes for its communes” and that codes are now assigned per commune [4]. A certified address, the plan de localisation postale (PLP), costs 2,000 FCFA [2]. On 21 October 2026 neither Campost’s website nor any public document showed a list, a format or an example of the commune codes [13].

2.3 Coded streets and door numbers in Yaoundé and Douala

The World Bank’s review documents the programme of the two cities in detail [5]. Each urban area was divided into six major zones; a street is identified by two figures, the zone and a sequential number (street 4.15 is the fifteenth street of zone 4); doors are numbered metrically, with even and odd numbers on opposite sides; and the survey recorded the occupant’s name, plot use and cadastral references. A presidential circular on street names dates from 1971. This explains the form of the addresses in use today: the United Nations’ offices in Yaoundé give their address as a door number on a coded street, a building name and a quarter (“Rue 1794, Ekoudou, Bastos”) [14]. OpenStreetMap names 8,625 streets of the two cities, 7,531 of them with their code. We found no public look-up and no open data of either city’s door numbers.

2.4 Land, mapping and the absence of an address register

Land titles and cadastral plans are kept by the Ministry of State Property, Surveys and Land Tenure (MINDCAF); they identify land, not the gate a courier stops at, and most homes have none. The national geodetic reference used in the cadastre has moved from the Douala 1948 datum to Manoca 1962 / UTM zone 32N [15]. A World Bank project approved in May 2025 addresses land administration and urban infrastructure, not street addressing [16]. The Institut National de Cartographie (INC) is the national map producer; by the country’s report to the United Nations Group of Experts on Geographical Names, toponymic work in 2023–2024 remained hampered by the economic situation and by insecurity in the North-West and South-West [17]. We found no national address register, no national addressing law and no national geoportal of addresses.

2.5 International practice

The UPU’s S42 standard defines how postal addresses are structured and rendered in each country [18], and ISO 19160-1 provides a conceptual model of addresses [19]. Both describe addresses; neither creates them where streets have no names. The UPU’s report Addressing the World argues that an address for every dwelling is a precondition for postal, financial and civil services [20]. The World Bank’s programme of street addressing in African cities, which included Yaoundé and Douala, is the reference for conventional practice [5]. Several countries of the region have adopted national digital address systems based on grid codes; Ghana’s, introduced in 2017, derives a code from the location’s grid cell [21].

2.6 Other code systems

Grid codes such as Open Location Code (Plus Codes) give any point on Earth a short code from its latitude and longitude [22]. They work offline, but their codes mix letters and digits, carry a plus sign and say nothing that a resident recognises: “6FR5JCH8+5QV” is hard to dictate in French, English or Pidgin, and no one remembers it. Proprietary word-based systems are not open. Neither type is unique to a home: two gates of one compound have two codes, and one code can cover two homes. The model presented here uses a grid only for points that are not homes (Section 4.10), and keeps names people say for homes.

3.1 Boundaries and named places

The 360 arrondissements are taken from OpenStreetMap (administrative level 8), and the named places from OpenStreetMap (city, town, suburb, quarter, neighbourhood, village, hamlet and locality) and from GeoNames populated places [23], [24], [25]. Each arrondissement receives a key of four characters (YDE6 for Yaoundé VI, DLA1 for Douala I, BDA2 for Bamenda II) and a frozen index from 0 to 359. Figure 1 shows the arrondissements shaded by the density of compounds.

3.2 Building footprints

Building footprints come from OpenStreetMap and, where OpenStreetMap has none, from Google Open Buildings [26] and Microsoft’s global footprints [27], both through the Overture Maps Foundation [28]. A footprint from Google or Microsoft is kept only where no OpenStreetMap building overlaps it. Footprints under 1 m² are dropped, as are 182,507 footprints outside the border. The result is 9,293,569 footprints: 3,836,259 from OpenStreetMap and 5,457,310 from the other two datasets. The median footprint is 51 m². Figure 2 shows the sources by region: OpenStreetMap is the majority source in the Centre, the South and the South-West, and the machine-detected footprints dominate in the Far North, where 70 % of footprints come from them.

3.3 Official streets, walls and gates

From OpenStreetMap we also take the named streets of Yaoundé and Douala with their codes, the walls, fences and hedges that enclose compounds (854 mapped enclosures with buildings inside), and the mapped gates and entrances. Every compound records the nearest official street within 30 m.

3.4 Verification of existing systems

The existing systems were verified between 17 and 21 October 2026 by opening each source and reading it; the full record, with each source marked as read or seen only in search results, is in docs/existing-systems.md of the repository [9].

4.1 The forms of one code

One compound has several forms of one code, each for a different reader.

Let $P$ be the area’s plate name, $N$ the number of $n \in \{3,4,5\}$ digits, $Z$ the arrondissement’s key, $I$ the area’s three-digit index within it, and $c$ the check digit (Section 4.9). The two principal forms are

where $\sqcup$ is a space and $g$ groups the digits as they are said: $g(137) = \text{137}$, $g(1401) = \text{14 01}$, $g(18834) = \text{18 834}$. A plate name never contains a digit, so the number always begins at the first digit of the text.

4.2 Units and code areas

A code area is a named place of one arrondissement: an OpenStreetMap place, or a GeoNames populated place unless it lies within 1 km of an OpenStreetMap place of the same name. A second place of one name within 300 m of the first is dropped. This yields 24,213 areas, of which 23,208 contain homes. Every footprint $f$ takes the nearest place point of its own arrondissement $U(f)$:

with $\lambda$ the longitude and $\varphi$ the latitude, and ties broken by the lower area key. Each area is frozen at first issue: its key, its block of numbers and its frame, the bounding box $(x_0, y_0, x_1, y_1)$ of its footprints’ centres, over which its numbers are laid out. Yaoundé has 307 areas and 346,523 compounds, Douala 267 areas and 410,644 compounds; the largest area, Alphonse (Carrefour Alphonse, Douala III), holds 19,107 compounds. Figure 3 shows the distribution of area sizes.

4.3 Plate names

A name on a gate, a parcel or a telephone call must be short. A plate name $s$ is admissible when

where $\lvert\cdot\rvert$ counts Unicode characters after canonical composition and $L$ is the set of articles and prepositions that are not counted: $L_{\text{fr}} = \{$de, du, des, d, la, le, les, l, et, sur, en, au, aux, à$\}$ in the 295 French-speaking arrondissements and $L_{\text{en}} = \{$of, the, and, on, at$\}$ in the 65 arrondissements of the North-West and South-West, where plate names follow English usage.

An official name that fails Eq. (3) is shortened by fixed steps, and the first step that succeeds is kept: cleaning (brackets; the generic words Quartier, Carrefour, Lotissement, Cité, Village, Quarter, Layout, Junction: “Carrefour Intendance” becomes Intendance); “Saint” becomes “St”; then the most distinctive run of whole words, taken only when it is clearly ahead of the alternatives. A word is never cut.

Cameroonian place names often join a landmark word to the place’s own name: “Carrefour Laboratoire Meka”, “Mission Catholique Mokolo”, “Place du Président Ahmadou Ahidjo”. The rule therefore carries a list of 186 ordinary French and English words (laboratoire, mission, catholique, président, cimetière, poteau, school, junction, …) that count as weak, so that a short form keeps the proper name (Meka, Mokolo, Ahidjo) and never the landmark. Words of the national languages that form part of names (Ouro, Mayo, Hosséré, Djaouro, Nkol, Mvog, Elig) are names, and so are the names of peoples (Haoussa, Foulbé). Where the only short form would contain a digit, consist of ordinary words only (“Vallée de la Mort” would give “Mort”), cut a phrase (“Yenga Sans” from “Yenga Sans Famille”) or name another town, the area carries its arrondissement’s name until the council chooses.

The 1,236 areas flagged for their councils carry a provisional plate name; 487,712 codes were issued under such a name, and a council may change it once. The official name always remains an alias (1,773 aliases are kept; 37 are dropped because two plate names claim them), so typing the full official name finds the same compound.

4.4 The name key

Two spellings of one name must decode to the same range. The key of a name is the composition

applied right to left: $\mu$ removes accents and $\upsilon$ converts to upper case; $\tau$ expands St and Ste to Saint and Sainte; $\eta$ drops a leading head word (Q, Quartier, Quarter, Hameau, Village de) and $\alpha$ a leading article (le, la, les, the, l’); $\jmath$ removes every character other than A–Z and 0–9; $\phi$ folds spelling variants common in Cameroonian orthography (tch and ch to c; dj to j; gh to g; ph to f; ck to k; ou and w to u; gu before i or e to g); and $\sigma$ writes a doubled letter once. Digits are never squeezed, so “Mile 22” stays apart from “Mile 2”. By construction $\kappa$ can only merge spellings, never split one name into two: Djoungolo and Joungolo, Tchinga and Chinga, Douala and Duala, Ekellé and Ekele each share a key. The 24,213 areas carry 17,240 plate names; 3,886 names are shared by 10,859 areas, among them 30 areas named Adjap (one spelt Ajap), 27 Mélen, 27 Yaoundé (a provisional name), 26 Akak and 24 Akom.

4.5 One national range per name

Every name owns one national range of numbers of a fixed length: 100–999, 1000–9999 or 10000–99999. Let a name be carried by $k$ areas with $c_1, \dots, c_k$ compounds, in the frozen order of their keys (arrondissement, then area index). Each area receives a block of size

twice its compounds rounded up to a whole ten, and at least 20. The total size of the name’s blocks and the number of digits of its range are

so that the top tenth of every range (beyond 810, 8,100 or 81,000 numbers) is always spare. The blocks are laid end to end from the bottom of the range:

Worked example: Bonapriso. The name has one area, in Douala I, with 962 compounds. By Eq. (5), $B = 10 \lceil 1924/10 \rceil = 1930$. Since $810 < T = 1930 \le 8100$, Eq. (6) gives four digits, and Eq. (7) gives the block 1000–2929; the numbers 2930–9999 are spare.

Worked example: Alphonse, the largest area, with 19,107 compounds: $B = 10\lceil 38214/10 \rceil = 38220$, and $8100 < 38220 \le 81000$ gives five digits and the block 10000–48219.

Worked example: Adjap, the most repeated name, carried by 30 areas in 17 arrondissements of the South and Centre regions. Their blocks sum to $T = 3950 \le 8100$, so the name has four digits; the blocks run from 1000 to 4949, and the numbers 4950–9999 are spare (Figure 5). Adjap 10 50 lies in Oveng and Adjap 48 50 in Olamzé: the number alone says which.

Figure 4 shows the distribution of $T$ over all names. 12,437 names fit three digits, 4,057 four and 213 five. No name needs a qualifier and no area had to be split: the fullest range, that of the 27 areas provisionally named Yaoundé, uses 49,000 of its 81,000 numbers.

Growth. A new area of an existing name, or an area whose block is full, takes the next block above every block already given to its name, from the spare. Blocks never move, shrink or get renumbered.

4.6 Numbering within an area: the Hilbert curve

Numbers should be local: neighbouring compounds should have neighbouring numbers, so that a courier who finds 14 01 knows that 14 02 is close. We order the compounds of an area along a Hilbert curve [29], whose clustering properties are the best known among space-filling curves [30]. Each compound’s reference point $(\lambda, \varphi)$ (the centre of its largest building until a gate is confirmed) is mapped to a cell of a $2^{16} \times 2^{16}$ grid over the area’s frame,

and the cell’s distance $h(g_x, g_y)$ along the curve is computed by the standard recursion over the bit levels $s = 2^{15}, 2^{14}, \dots, 1$: with $r_x = [g_x \wedge s \ne 0]$ and $r_y = [g_y \wedge s \ne 0]$, add $s^2 \cdot \big((3 r_x) \oplus r_y\big)$ to $h$; then, if $r_y = 0$, reflect the cell ($g \mapsto 2^{16} - 1 - g$ on both axes) when $r_x = 1$ and swap $g_x$ and $g_y$. Panel (a) of Figure 6 shows the order of the cells for a curve of order 3.

The $m$ compounds of an area with block $[F, F + B)$, sorted by $(h, \text{stable identifier})$ and indexed $i = 0, \dots, m-1$, receive

so that the first two thirds of the block, the main part $[F, F+M)$, carry the first issue with gaps of $\Delta - 1$ numbers for infill, and the last third, the tail reserve $[F + M, F + B)$, is kept for later.

Worked example: Bonapriso 14 01. The compound’s reference point is $\varphi = 4.0243022$, $\lambda = 9.6904852$; Bonapriso’s frame is $(9.687462,\, 4.017162,\, 9.697280,\, 4.031143)$. Eq. (8) gives

$$g_x = \Big\lfloor \frac{0.0030232}{0.009818} \cdot 65535 \Big\rfloor = 20179, \qquad g_y = \Big\lfloor \frac{0.0071402}{0.013981} \cdot 65535 \Big\rfloor = 33469,$$

and $h = 2\,062\,693\,980$, the 402nd of the area’s 962 positions along the curve ($i = 401$). With $B = 1930$, Eq. (9) gives $M = 1286$ and $\Delta = \max(1, \lfloor 1286/1443 \rfloor) = 1$, hence $N = 1000 + 401 = 1401$; the tail reserve is 2286–2929. Panel (b) of Figure 6 shows all 962 compounds of Bonapriso coloured by number.

4.7 A new home

A compound built after first issue is placed between its two neighbours on the curve. Let $a$ be the number of the last issued compound at or before its position $h$ and $b$ the number of the first after it ($a = F - 1$ and $b = F + M$ at the ends). The new compound receives

the free number $x$ (in the set $\mathcal{F}$ of numbers never issued) nearest the midpoint, the lower on a tie; if there is none, the lowest free number of the tail reserve; then of the area’s later blocks; and if the area is full, a new block of its name from the spare (Section 4.5). In Bonapriso $\Delta = 1$, so the main part has no gaps between neighbours, and a house built between two of them receives Bonapriso 22 86, the first number of the tail reserve. A number is never reused, even after its code is retired.

4.8 Decoding without context

The decoder reads a text as a name followed by three to five digits, with or without spaces, and returns the set of areas

Because the blocks of one name are disjoint, $\lvert D \rvert \le 1$ for every input. A number in no block, in a name’s spare or with the wrong number of digits is reported as not issued; it is never corrected or guessed. A name that also exists in another country is resolved by the number, then by the user’s position; when neither decides, no country is assumed.

4.9 The check digit

The machine form ends with a check digit computed by Damm’s algorithm [31], [32], a decimal method that, unlike the Luhn and Verhoeff schemes [33], uses a single table and detects every single-digit error and every transposition of two adjacent digits. Let $u = u_1 u_2 \cdots u_\ell$ be the concatenation of the name’s national index on five digits, the arrondissement’s index on three, the area index on three and the number. Then

where $Q$ is the table of a totally anti-symmetric quasigroup of order 10:

Worked example. Bonapriso’s national index is 02616, Douala I’s index 020, the area’s index 015 and the number 1401, so $u = 026160200151401$:

The check digit is 6 and the machine form CM-DLA1-015-1401-6. Because the name’s index, the arrondissement and the area enter the check, a number moved to another area of the same name fails both the check and the block test of Eq. (11). The check digit lives only in the machine form: it is never shown to people, never typed by them, and verified whenever a system sends a code.

4.10 Spot codes

Any point, home or not, has a spot code: its arrondissement’s key and its distance east and north of the arrondissement’s frozen origin $(\lambda_0, \varphi_0)$, the south-west corner of its bounding box floored to 0.01°, in units of 10 m:

where $k_x$ and $k_y$ are metres per degree of longitude and latitude and $\varphi_c$ is the frame’s reference latitude (the centre of the box, rounded to 0.01°). Each axis has four digits, or five where the arrondissement spans more than 99.99 km. Decoding returns the centre of the cell, so the error is bounded by half the cell’s diagonal:

Worked example. For Bonapriso 14 01, in Douala I’s frame ($\lambda_0 = 9.65$, $\varphi_0 = 4.00$, $\varphi_c = 4.05$): $k_x = 111\,042$; $E = \lfloor 0.0404852 \times 111\,042 / 10 \rfloor = \lfloor 449.56 \rfloor = 449$ and $N = \lfloor 0.0243022 \times 110\,574 / 10 \rfloor = \lfloor 268.72 \rfloor = 268$, hence the spot code CM-DLA1 E0449 N0268. Its decoded centre lies 2.3 m from the point (Figure 7). A spot code is shared by link; it is never needed to find a home.

Points in no outline. The arrondissement outlines are generalised, so 240 issued compounds lie in no outline: 67 in slivers between arrondissements, 172 just over an outline at the shoreline or the national border, all within 62 m of one, and one industrial structure at sea, 10.9 km off Kribi. Such a point takes its arrondissement from those whose outline lies within 12 km: among them, the one with the nearest outline whose frame holds the point, distances being measured to the nearest edge with longitude differences scaled by $\cos\varphi$ and ties going to the lower key; if no such frame holds it, the nearest cell of a frame at most 5,000 m away. Frames are never moved or widened, because either would change spot codes already published, and the decoder is unchanged, since a spot code names its arrondissement. Every issued compound therefore has a spot code, and the registry keeps its exact entrance point. For example, a compound of Yoyo Bar (Yaoundé II) at $\varphi = 3.8751747$, $\lambda = 11.4844409$ lies in a sliver 5.6 m from Yaoundé VII’s outline and 9.7 m from Yaoundé II’s, falls in Yaoundé VII’s frame and receives CM-YDE7 E0826 N0499. The structure at sea lies 4.49 km beyond the frame of Kribi I and receives its nearest edge cell, CM-KRI1 E0000 N3321.

4.11 Compounds: grouping structures

The registry numbers compounds, not buildings. Before any number is given, the footprints of each area are grouped by four rules, in a fixed order by stable identifier. Distances are measured between outlines,

Only annexes join a house, and a hut cluster never joins two houses, so groups cannot chain across a quarter. The rules produced 7,243,272 compounds from 9,293,569 footprints, 1.28 footprints per compound: 5,795,550 single buildings, 1,155,565 houses with their annexes, 290,494 hut clusters and 1,663 compounds within mapped walls. By size, 5,796,261 compounds have one structure, 1,062,820 two, 256,178 three and 128,013 four or more.

Validation. We re-ran the rules on the 854 mapped enclosures with the enclosures hidden from them: 10.5 % of the enclosed compounds come out whole and 9.3 % exactly, and a further 40.4 % miss only one building. Behind a wall, the main house, the rooms and the kitchen are separate buildings, and with no wall mapped the rules keep them apart deliberately: merging two codes later costs nothing, whereas splitting a wrong merge gives a household a new code. Residents’ answers in the N-Ways application (“these buildings are one home”, “here is the gate”) complete the grouping.

4.12 The building profile

Every structure carries a use class with its confidence and source, its size, its floors where known and a status. Classes are inferred in a fixed order: the OpenStreetMap building type (confidence 0.85), a shop, amenity or office tag on the footprint (0.75), an N-Ways place inside it (0.65), the footprint’s size (0.3–0.4), and otherwise residential by default (0.45). Of 9,293,569 structures, 9,156,024 are classed residential, nearly all by the default; 82,878 are annexes, 14,633 under construction, 14,446 commercial, 8,090 educational, 3,618 industrial, 3,551 governmental, 2,922 religious and 2,635 health buildings. OpenStreetMap gives floors for 1,835 buildings; the heights in the machine-detected datasets are not used. A home never carries a name.

5.1 The reference engine

The reference engine (engine/reference/cm_engine.py and plate_names.py) implements every rule of Section 4 in Python with the standard library only: parsing of every form, the name key, plate names, blocks and ranges, the Hilbert numbering, insertion, decoding, the Damm check and spot codes. The normative specification is engine/SPEC.md. The engine ships with its data: engine/data/register.json (every arrondissement with its spot frame, every area with its plate and official names, name index, blocks and frame, and the aliases) and engine/data/units.json (the arrondissement outlines). It decodes offline, with no server. Conformance vectors (engine/reference/vectors.json) give 722 name keys, 500 code forms, 2,394 point look-ups, 698 spot codes, 470 points in no arrondissement outline (the 240 issued compounds among them, points beyond a frame and refusals), Hilbert indices and grouping cases; any other implementation must reproduce them.

5.2 The registry

The registry (registry/) publishes every issued code: one gzip-compressed CSV file per arrondissement (360 files, 7,243,272 rows), the frozen area table areas.csv.gz (24,213 areas with their blocks, names, frames and counts) and MANIFEST.json, which gives the count of each file and the SHA-256 of each file uncompressed. Each row holds the three forms, the area, the number and check, the status (provisional until residents confirm it), the issue date, the entrance point and its kind, the spot code, the official address where one applies, and the building profile. Building outlines are not republished: they remain in their source datasets, from which they can be joined.

5.3 The server and its API

The repository includes a deployable server (server/), written in Python with the standard library only and reusing the reference engine. import streams the registry into SQLite, verifies each file against the manifest and refuses a mismatch, and builds indexes on every form and an R*-tree spatial index [34], [35]. serve exposes a JSON API under /v1, described by an OpenAPI 3.1 contract [36]: health and information, the full card of a code, search by any form as people type it, the code at a point, codes near a point, verification, spot codes, arrondissements and areas, and paged listings. export writes the registry back in the published format; after an import of the published registry the export is byte-identical, file by file, when uncompressed. A Docker image imports the bundled registry on first start, so that docker compose up is a complete deployment.

The server can also issue codes by the issuer’s own rules: the duplicate checks (an entrance within 3 m, or a point inside an issued outline), the outline limits (6 m² to 2 ha), the placement rule (the area of the nearest issued code within 250 m, else the area containing the point), numbering by Eq. (10) with extension blocks, the 80 m rule for further doors and the per-key daily limit. Issuing must come from one issuer per country, because two issuers working from the same registry would give the same free number to two homes; a deployment therefore issues only when the State or N-WAYS LIMITED has designated it, and otherwise runs read-only.

5.4 Conformance

Parity with the issuer is tested case by case: 253 issue requests posted in order (points near real compounds, points already coded, points far from any code, points outside the country, and 215 doors of one compound that fill its area and open an extension block), 12 single insertions and 83 point look-ups, each computed by the issuer’s own implementation from the same register; the server reproduces all of them. On the full registry, 50 random codes in 45 arrondissements agreed field by field with the issuer’s service, and the full import took 166 s and produced a 3.4 GB database.

11.1 Measures of impact

The estimates above rest on assumptions. We will measure the following with partners and publish the results.

12.1 Privacy

Codes describe places, never people. A home carries no name and no data about its occupants; business and public buildings carry their public names. Photographs taken when a resident adds a home are kept for verification only, and a household can ask for its gate to be hidden. No identity number is collected. Law no. 2024/017 of 23 December 2024 on the protection of personal data has applied since 23 June 2026; it subjects transfers abroad to prior authorisation and creates a supervisory authority [44]. N-Ways declares to that authority its processing of the personal data its application holds (telephone numbers and photographs) and makes transfers abroad on the terms the law sets.

12.2 Safety

The registry publishes every issued code in all 360 arrondissements. The United Kingdom’s travel advice counsels against all travel to the North-West, to most of the South-West and the Far North and to the border bands with Nigeria, Chad and the Central African Republic, and against all but essential travel to the North and parts of Adamawa [45]. The area table flags the 187 arrondissements of the North-West, South-West, Far North, North, Adamawa and East regions (alert_zone; 13,824 areas and 4,046,885 codes) as context for relief and emergency planning. Relief agencies, health services and emergency responders need addresses most where access is hardest, and a code describes a place, not who lives there. Data about people (who lives where, who confirmed a code) is never published and is released only under a court order, with a public report.

12.3 Governance

One issuer gives codes for the whole country, so that no two homes share a code: N-WAYS LIMITED today, or the body the State designates. Areas, blocks and frames are frozen and codes never move; regrouping buildings never changes an issued code. A council may rename an area (the old name remains an alias), choose a provisional name once, or add areas. The councils’ addresses are theirs: they are shown as written, never edited, and any mismatch observed on the ground is reported to the council. Every code starts as provisional and becomes confirmed when residents agree in the application (two telephones, at least one at the gate, and a majority of answers); no code is reviewed by hand. The repository can be transferred to the State or a body it designates under its open licences, without forking the engine.