How this was built, and what it cannot tell you.

• page header reads '2025 MONTHLY FIGURES' but the file is stats_monthly2026.htm; year taken from the URL

• Primary filings for the two licensed Maldives telecom operators. The portal's coverage begins in 2023; earlier years are taken from the five-year summaries inside the annual reports, which is recorded on each derived series.

• Ooredoo publishes rate plans on its website but sells most prepaid add-on packs through its app, USSD and self-care, which are not public. Dhiraagu publishes its full prepaid pack ladder. The two catalogues are therefore not symmetric, and any 'portfolio size' comparison between them measures disclosure, not product count.
• A snapshot cannot describe a portfolio's history; no operator publishes withdrawn products or past prices.
• Headline data allowances on very short-validity packs are fair-use ceilings, not volumes a customer can consume.

• Used only as a penetration denominator. Intercensal monthly population is a model; no official monthly series exists.

• Used to test whether the mobile subscription base moves with the visitor cycle. Arrivals count entries, not SIM purchases.

February 2012 is the material case: the published total (508,971) is lower than its own components (538,174) and lower than both the month before and the month after, which is the signature of a transcription error in the Total row rather than a real 5% one-month collapse. The remaining breaks are small.

Action: flagged, not corrected; February 2012 is excluded from growth-rate calculations and marked on every chart that shows it

Small residuals of ±1 to 200 lines against a base of 11,000–26,000. Consistent with rounding or single-cell typos in a hand-maintained table.

Action: left as published; regional shares are computed on the components, not the total

Adding depreciation and amortisation back to the reported operating result gives MVR 1,648m for Dhiraagu in 2025 against MVR 1,638m reported — a 0.6% difference. The same arithmetic gives MVR 1,167m for Ooredoo against MVR 1,263m reported — 8.3%. Ooredoo's reported EBITDA therefore excludes roughly MVR 96m of costs that its own income statement charges above the operating line, or is struck on a different consolidation. Comparing the two headline margins (57% and 58%) without saying so would be a like-for-unlike comparison.

Action: the site reports a consistently derived EBITDA for both operators alongside each operator's own reported figure, and never mixes the two in one comparison

Dhiraagu publishes its complete prepaid pack ladder on its website; Ooredoo publishes rate plans but sells most prepaid add-ons through its app, USSD (*929#) and self-care, which are not public. A count of published products measures disclosure practice, not portfolio breadth.

Action: product-count comparisons are not made; price comparisons are restricted to segments where both operators publish

Dhiraagu's FLEXIHOUR 20 advertises 100GB + 25GB with one hour of validity. On the headline it costs MVR 0.20 per GB, which would make it the cheapest data sold in the Maldives by a factor of fifty. It is an unlimited hour, priced as access.

Action: a consumption-capped price per GB is computed alongside the headline one, and the cap is stated as an assumption

Large single-month moves in a subscription series are usually administrative (a base clean-up, a reclassification) rather than behavioural. April 2022's mobile-broadband reading is 22% below both neighbours and returns to trend the following month.

Action: retained; each is listed so a reader can judge it, and none is smoothed

Large single-month moves in a subscription series are usually administrative (a base clean-up, a reclassification) rather than behavioural. April 2022's mobile-broadband reading is 22% below both neighbours and returns to trend the following month.

Action: retained; each is listed so a reader can judge it, and none is smoothed

Mobile subscriptions peaked at 907,763 in September 2017 and then fell for four years. The 2018 decline is gradual — prepaid falls month after month while postpaid rises — so it reads as substitution, not a one-off base clean-up. The 2020 fall is abrupt and coincides with the border closure and the departure of migrant workers and tourists. Neither is a data error.

Action: the forecasting work fits from 2022 onward and reports what happens when the pre-COVID period is included instead

Voice, SMS and roaming packs often state minutes rather than gigabytes, and recurring rentals have no expiry. Those products have no price-per-GB and are excluded from data-price charts rather than given a zero.

Action: excluded from the affected metric, retained everywhere else

CAM publishes one page per year; a missing month means the page cell was blank.

Action: none required

This is a definitional finding, not just a check. CAM's 'Broadband (Mobile)' series includes LTE fixed-wireless broadband — a household product sold on a router, not a mobile subscription. Anyone reading the headline as mobile-only data users overstates them by the LTE component (13,284 of 342,800 in December 2025).

Action: the analysis reports mobile broadband excluding LTE fixed wherever the distinction matters, and says which is being used

The revenue split is the analytically valuable part of these filings and it reconciles, which is the main evidence that the extraction read the right columns.

Action: any break is listed and re-read from the filing

The quarterly income statements are parsed out of PDFs; the annual reports state the year total independently. Agreement to the rounded million is the strongest available evidence that the parse is right.

Action: no adjustment

The monthly subscription series contains two regime changes: a four-year decline from the September 2017 peak, driven by prepaid, and an abrupt fall in 2020 when the border closed and migrant workers and visitors left. Fitting a trend model across either produces a confident and wrong answer. The published models are estimated from January 2022 forward; the full-sample backtest is computed too and stored alongside, so the choice of window is visible rather than assumed. This follows the house rule on structural breaks: they are handled explicitly, never silently fitted over.

Models are scored on MAE, RMSE, MAPE and MASE across a rolling origin with a twelve-month horizon. MASE is scaled by the in-sample seasonal-naive error, so below 1.0 means the model beats repeating last year's month. Prediction intervals are empirical — the 80th percentile of the model's own backtest errors at each horizon step, widened with the square root of time beyond the backtested horizon — rather than the model's parametric interval, because a model selected by backtest should report the uncertainty it actually produced. Bull and bear scenarios are not separate models: they are the base path shifted by a stated ±2 percentage points of annual growth, so the reader can see exactly which assumption creates the spread.

• No market shares. The operators' customer counts and the regulator's subscription counts are different concepts, and no operator publishes its own definition.
• No churn. Not disclosed by either operator, not required by the regulator or the exchange.
• No product volumes. Which means no measured elasticity, no revenue mix by product, and no cannibalisation measurement — only frameworks with stated assumptions.
• No usage data. CAM counts subscriptions, not traffic. There is no published Maldivian series for data volume per subscriber.
• No product history. Operators publish what is on sale today. A snapshot cannot describe last year's portfolio, so the product data on this site is explicitly a point-in-time observation.
• Almost no geography. The only geographic split the regulator publishes is fixed telephone lines, Malé region against everywhere else.

Where a section needed one of these and could not have it, the site builds the framework, labels it, and names the dataset that would make it real. That is the difference between a gap and a fabrication, and it is the only defensible way to do this work from outside.