15h Hradčany 15h brick red · POFIS 7 · 1919
CZENFRDEES

About the collection

This site is based on many years of specialised study of the 15h value of the Hradčany issue. So that the figures and reconstructions carry weight, this page explains where the data come from and how they are produced.

Scope of the material

The statistics are based on records of 21,239 attributed 15h stamps (as of June 2026), 6,243 of which carry a legibly dated cancellation. The plate reconstructions rest on 8,012 stamps attributed to their stamp field — from the complete, deep reconstructions of plates I and II (20–55 copies per position) to the reference library of the rare plate VII (122 copies over 70 positions).

How the reconstructions are made

Every copy is scanned and registered on the red frame of the design to a uniform format, so the design details sit in the same place on every stamp and positions can be compared side by side (see the comparator). The material comes from album pages, reconstruction sheets, strips and blocks as well as loose sets; for strips and blocks the attribution is verified by field adjacency (horizontal neighbour +1, vertical +10), which has several times caught an attribution typo.

Quality is guarded by multi-stage checks: montage reviews of every batch, overlap detection, cross-checks between plates and — for plates with enough material — statistical position models that allow blind verification of attributions.

Manual and machine attribution complement each other. Every stamp is first of all attributed the classical way — with a magnifier, comparing the plate characteristics against the reference library, just as Hradčany specialists have done for decades. Alongside this I also use automated recognition based on artificial intelligence, which serves as a second, independent opinion and as fast pre-sorting of larger batches of material.

How does the automated recognition work? Put simply: the computer has “learnt” to recognise the individual stamp fields from thousands of manually attributed scans. Each scan is first precisely registered on the frame of the design so that the details always sit in the same place. From the image it then extracts the tiny deviations of the drawing that the eye easily overlooks — exactly the ones by which the fields of a letterpress plate differ — and statistically compares them with all the reference positions. The result is not a verdict but a list of the most probable positions with a confidence level; for plates with enough material the model is right first time in roughly 9 cases out of 10. The final word, however, always belongs to manual checking — no attribution counts as valid until I have confirmed it myself.

Sources

Author: Josef Matoušek, philatelist and Hradčany specialist — pepa.matousek@gmail.com. My thanks for cooperation and material go to colleagues, especially Jozef Džubák, Jaroslav Michalík, Johan Sevenhuijsen and Mark Wilson.