Add real test methodology document

README.md

@@ -62,6 +62,7 @@ Report reali eseguiti su corpus AIO locale (101k+ records, fonti pubbliche reali
 Artifact pubblicati:
 - `reports/REAL_CASE_TEST_RESULTS_LATEST.md`
 - `reports/REAL_CASE_TEST_RESULTS_LATEST.json`
+- `reports/TEST_METHOD.md`
 
 Scope case use testati:
 - clima mediterraneo ed energia

reports/TEST_METHOD.md

@@ -0,0 +1,38 @@
+# AIO Real Test Method
+
+Generated on: 2026-02-21
+
+## Goal
+Publish real multi-case retrieval tests on the full local AIO corpus, then expose only results in the public showcase.
+
+## Data Scope
+- Corpus file: `aio_archive/records.json`
+- Corpus size at test time: `101257` records
+- Unique sources at test time: `1867`
+- Source family: public knowledge files already ingested locally (wiki-based IT/EN and related public text sources).
+
+## Engine Configuration
+- Runtime: `python app_oracolo.py` (CLI mode)
+- Embedding backend used in test: `sentence-transformers(local-path): aio_models/paraphrase-multilingual-MiniLM-L12-v2`
+- Retrieval path: AIO native hybrid retrieval (vector + lexical overlap + domain coherence bonus).
+
+## Test Procedure
+1. Internal benchmark run:
+   - command: `AIO_CLI_BENCH=1 python app_oracolo.py`
+2. Real use-case runs (8 cases):
+   - command per case: `AIO_CLI_TEST_QUERY=1 AIO_CLI_QUERY="<query>" python app_oracolo.py`
+3. Parse output:
+   - `hits`, `score`, `domain`, `source`
+4. Case pass criteria:
+   - minimum hits threshold (case-specific),
+   - top score threshold,
+   - at least one expected term matched in retrieved context.
+
+## Published Artifacts
+- `REAL_CASE_TEST_RESULTS_LATEST.json`
+- `REAL_CASE_TEST_RESULTS_LATEST.md`
+- timestamped copies (`REAL_CASE_TEST_RESULTS_20260221_222608.*`)
+
+## Licensing Note
+- Free for use and study.
+- Commercial use requires written authorization: `info@rthitalia.com`