# Italian production dictionary reproduction

Morph-it! 0.4.8 (23 February 2009), Marco Baroni and Eros Zanchetta, SSLMIT, University of Bologna, is used under CC BY-SA 2.0. Wikizionario contributors' Italian Wiktionary snapshot 20260901 is used under CC BY-SA 4.0. Adapted semantic decision data and the combined dictionary are CC BY-SA 4.0. See the included source notices and upstream README. Application and parser code remain separate from licensed data.

## Acquire the pinned sources

Use the repository's existing `sources:acquire` command with source IDs `it-morphit-0.4.8 it-wiktionary-20260901`. Set `INSTANT_ANAGRAM_SOURCE_CACHE`, `INSTANT_ANAGRAM_TEMP_ROOT` and `INSTANT_ANAGRAM_REPORT_ROOT` to external directories as for the other production languages. `sources.lock.json` pins both archive identities; `generators/it/source-envelope.json` additionally pins Morph-it's Latin-1 raw entry and the dump's size, MD5, SHA-1 and SHA-256. Never substitute a latest dump. Do not store archives or caches in the repository.

Morph-it's authoritative URL is https://docs.sslmit.unibo.it/lib/exe/fetch.php?media=resources:morph-it.tgz . If it is unavailable, the byte-identical historical recovery is https://web.archive.org/web/20160328041750id_/http://dev.sslmit.unibo.it/linguistics/downloads/morph-it.tgz . Recovery must pass the same size and SHA-256 checks before entering the canonical immutable source cache.

## Reproduce semantic source data

The committed `data-sources/it-wiktionary/lexical-identities.jsonl.gz` contains only lexical classifications and page/revision attribution, not definitions, examples or media. Reproduce it from the pinned dump with Node and Python 3's standard library:

```
node tools/reproduce-it-source.mjs <pinned-dump-file> <fresh-external-directory>
```

Set `INSTANT_ANAGRAM_PYTHON` to an existing Python executable if `python3` is unavailable. The command verifies all dump identities, runs the locked literal lexical-marker parser, and verifies the generated compact ledger against the source envelope. Temporary page/template caches remain external. The parser is the PREQ-G lexical-marker taxonomy, with F1 membership semantics; diagnostic prose does not classify lexical identity. This is reproduction, not a new source audit.

For each identity, `id` starts with the page ID, `title` names the page and `revisionId` pins its revision. Attribution links are `https://it.wiktionary.org/wiki/TITLE`, `https://it.wiktionary.org/w/index.php?title=TITLE&action=history` (URL-encode TITLE), and `https://it.wiktionary.org/w/index.php?oldid=REVISION_ID`. Preserve contributor credit and these references in redistributed adaptations.

## Regenerate production packages

Set the existing production variables `INSTANT_ANAGRAM_SOURCE_CACHE`, `INSTANT_ANAGRAM_TEMP_ROOT`, `INSTANT_ANAGRAM_REPORT_ROOT`, `INSTANT_ANAGRAM_PRODUCTION_ROOT` (fresh external output), and `INSTANT_ANAGRAM_RUN_ID`, then run:

```
npm run production:regenerate -- it
```

The canonical orchestrator verifies both cached sources, joins Morph-it analyses to the pinned semantic ledger, encodes core/display with the shared encoders, and verifies the complete production manifest. It requires no PREQ review archive or PREQ workspace. Repeat with a second fresh output and run ID to compare deterministic output.

Membership is `it-wiktionary-lexical-marker-surface-v2`; surface policy is `it-exact-surface-pos-marker-v2`. Normalization is NFC → Italian lowercase → NFC, 2–30 code points, alphabet `abcdefghijklmnopqrstuvwxyzàèéìïòùú`, without NFKC, transliteration or accent folding. Morph-it families ABL, NPR, PON, SENT, SMI and SYM never confer membership. Separate ordinary/form lexical identities can support homographs; prose keywords never veto them.

Golden: 24,244 lemmas, 226,961 ordered surfaces, standard flag 1 only, rank UNRANKED 0xffffff, zero display overrides. `si` lacks a safe POS-compatible pronominal identity and remains absent; `koala` is outside the Morph-it foundation. No manual additions, frequency source or scoring changes. Italian Phrase is not published.

Modifications: extract Italian lexical markers and snapshot template aliases, match POS-compatible Morph-it lemmas and exact surfaces, normalize, exclude unsupported identities, deduplicate and encode IAG2. Source and adapted data are provided as-is under their license disclaimer terms, without endorsement. Preserve attribution, modification notices and ShareAlike; impose no additional restrictions or DRM on licensed reuse.

The production package differs from the PREQ candidate only in metadata binding: sourceHashes and buildConfigHash refer to the canonical production envelope rather than review-only ledgers/build references. The ordered record/string tables, alphabet, rank, flags and display semantics are identical. The display sidecar binds the resulting new core SHA-256.