diff --git a/Uruguayan-eID.md b/Uruguayan-eID.md
index 5d110d1..fec6eab 100644
--- a/Uruguayan-eID.md
+++ b/Uruguayan-eID.md
@@ -31,6 +31,10 @@ run by [AGESIC](https://www.gub.uy/agencia-gobierno-electronico-sociedad-informa
 Resources:
 
 * Official digital-identity information: [AGESIC](https://www.gub.uy/agencia-gobierno-electronico-sociedad-informacion-conocimiento).
+* AGESIC's [technical documentation for the chip card](https://www.gub.uy/agencia-gobierno-electronico-sociedad-informacion-conocimiento/book/7192/download)
+  (PDF, in Spanish), which specifies the APDUs and the data-object layouts, and
+  the [reference APDU code](https://github.com/eIDuy/apdu-services) it publishes
+  alongside it.
 * [firmauy](https://github.com/carlosplanchon/firmauy): a third-party command-line
   tool to sign and verify PDF (PAdES), XML (XAdES) and arbitrary files
   (CAdES/.p7s) with the cédula over PKCS#11, with local certificate-chain
@@ -110,9 +114,129 @@ Data object 'MRZ'
     Data (93 bytes): <machine-readable zone, redacted>
 ```
 
+## Decoding the identity data objects
+
+`pkcs15-tool -D` prints the objects as raw bytes, and `-R <label> -o <file>`
+writes a single one to a file. Each object is a BER-TLV wrapper (tag, length,
+value), so getting the human-readable content means stripping the header and
+decoding the value:
+
+| Object label on card | EF, under DF `7000` | Tag | Length form | Value |
+|---|---|---|---|---|
+| Numero de documento | `7001` | `5F 01` | short, 1 byte | ASCII digits |
+| Datos biograficos | `7002` | `1F xx`, one per field | short, 1 byte | text, see below |
+| Fotografia | `7004` | `3F 01` | **long**, `81`/`82` + N bytes | JPEG |
+| MRZ | `700B` | `7F 01` | short, 1 byte | ASCII, fixed-width lines |
+
+The `pkcs15-tool` output above prints these as full paths (`3f0070007001` and so
+on), the EF column is the last two bytes of each.
+
+Only the photo exceeds 255 bytes, so it is the only object using a long-form
+length. These lengths were obtained by reading official documentation and experimentation. **Always read the length byte instead of assuming a fixed header size**.
+
+### Document number (`7001`)
+
+Tag `5F 01`, one length byte, then that many ASCII digits, so the value starts at
+offset 3. The specification fixes that length at 9.
+
+This is a different field from the ID number in `7002` (tag `1F 07`). The two
+usually differ in length and are not interchangeable.
+
+### Biographic data (`7002`)
+
+A flat run of TLVs, each with a two-byte tag `1F xx` and a one-byte length. Walk
+the buffer from offset 0: read the tag and length, take that many bytes as the
+value, advance by 3 + length, and stop once the byte at the current offset is no
+longer `1F` (end of data or padding) or a declared length would overrun the
+buffer.
+
+Zero-length fields are normal, only the header is present.
+
+The specification describes these values as ASCII, but that does not hold on real
+cards. The text fields are **UTF-8**, and decoding them as ASCII or latin-1 gives
+mojibake on accented names, which are common here. Not every field is text
+either, see `1F 08` below.
+
+| Tag | Field | Format |
+|---|---|---|
+| `1F 01` | Surname(s) | UTF-8 text |
+| `1F 02` | Second surname | UTF-8 text, might be empty |
+| `1F 03` | Name(s) | UTF-8 text, all given names in one field |
+| `1F 04` | Nationality | ISO 3166-1 alpha-3 (e.g. `URY`) |
+| `1F 05` | Date of birth | `DDMMYYYY` |
+| `1F 06` | Place of birth | `CITY/COUNTRY`, country as alpha-3 |
+| `1F 07` | ID number | digits in ASCII, check digit last |
+| `1F 08` | Date of issue | 4 bytes, **packed BCD**, `DD MM YY YY` |
+| `1F 09` | Expiry date | `DDMMYYYY` |
+
+The specification only lists `1F 01` through `1F 06`, and fixes the length of two
+of them: `1F 04` is 3 bytes and `1F 05` is 8. The rows for `1F 07` to `1F 09`
+come from reading cards, not from the document.
+
+`1F 08` is the field the specification warns about without ever naming it: it
+says every value is ASCII *except* the date of issue, yet lists no tag for that
+date. It holds four bytes of packed BCD, two digits per byte, in `DD MM YY YY`
+order, and decoding it that way yields the date of issue printed on the card.
+
+> **`1F 08` will fool a non-ASCII check.** BCD bytes all stay below `0x80`, so
+> scanning for high bytes reports the field as ASCII, and decoding it as text
+> succeeds and returns punctuation instead of raising. A card issued this century
+> even carries a literal `0x20` for the `20` of the year, which reads as a space.
+> Decode this field as BCD, not as text, and do not infer the encoding of a field
+> from whether high bytes are present.
+
+`1F 0A` has also been seen, zero-length on every card read so far, with no
+documented meaning. AGESIC's reference code models one field that no tag accounts
+for, *observations*, which would fit. Surface unknown tags rather than dropping
+them, and do not assume any of them are ASCII.
+
+The two applet versions carry the same layout. A v4 and a v5 card read for this
+page both hold the same ten tags, `1F 01` to `1F 0A`, in the same order, with
+`1F 08` four bytes wide on both. Nothing here is specific to one generation.
+
+> **The surname split is not reliable.** The specification names `1F 01`
+> *PrimerApellido* (first surname) and `1F 02` *SegundoApellido* (second
+> surname), but cards have been observed writing *both* surnames into `1F 01` and
+> leaving `1F 02` zero-length. That was the case on both cards read for this page,
+> one of each applet version. An empty `1F 02` therefore does not mean the holder
+> has a single surname, and splitting `1F 01` on whitespace is ambiguous with
+> compound surnames (*De León*, *Da Silva*). Label `1F 01` as *surname(s)*.
+
+### Photo (`7004`)
+
+Tag `3F 01`, followed by a long-form length. The byte at offset 2 gives the
+header size: below `0x80` it is the length itself and the value starts at offset
+3, `0x81` means one more length byte and the value starts at 4, `0x82` means two
+and the value starts at 5. Portraits are a few tens of kilobytes, above the
+255-byte ceiling of `0x81` and below the 65535 of `0x82`, so `0x82` is the usual
+case.
+
+The value is a complete JPEG, no re-encoding needed: write it out as-is. It
+should start with `FF D8 FF` (SOI) and end with `FF D9` (EOI); if it does not,
+the length form was misread and the value offset is wrong.
+
+### MRZ (`700B`)
+
+Tag `7F 01`, one length byte, then that many ASCII characters starting at offset
+3. The value carries no line separators, so split it positionally by total
+length: 90 characters is TD1 (3 lines of 30, the ID-card format), 88 is TD3
+(2 lines of 44, the passport format). Trailing runs of `<` are MRZ filler, not
+data.
+
+The specification does not state which layout the card carries, both figures are
+the ICAO 9303 ones. The 93-byte object in the dump above is a TD1 value plus its
+3-byte header.
+
 ## Notes
 
 * The ATR differs between card batches (the applet version and batch bytes), so
-  the driver matches it with a mask.
+  the driver matches it with a mask. The applet version is not derived from it,
+  it is read with `GET DATA` as described at the top of this page.
 * The identity data objects are readable without authentication by design, since
-  the same data is printed on the card itself.
+  the same data is printed on the card itself. Anything read out of them is still
+  personal data, treat it accordingly.
+* **Reading these objects proves nothing about authenticity.** The ICAO Document
+  Security Object that signs them, EF `711D`, is absent on earlier card batches,
+  so the driver does not expose it, and AGESIC's own reference code reports the
+  data as unsigned whenever the file is missing. A portrait or an MRZ read this
+  way is unverified data.