{
  "meta": {
    "title": "ADRIZ -> INGV Etna -- PROOF OF VALUE (independently verifiable)",
    "generated_utc": "2026-06-27T21:00:00Z",
    "intent": "Show, not claim. Each row places what ADRIZ independently produced next to INGV's OWN public report for the same date, with a live link so INGV can check us. Honest verdicts include where we missed or disagreed.",
    "banner": "INTERNAL / PREVIEW -- ADRIZ self-assessment for INGV review. Not a public product.",
    "primary_ingv_source": {
      "name": "INGV Osservatorio Etneo weekly bulletin, mirrored verbatim (English) by the Smithsonian Global Volcanism Program (GVP), Etna volcano number 211060",
      "archive_url": "https://volcano.si.edu/volcano.cfm?vn=211060",
      "rss_url": "https://volcano.si.edu/news/WeeklyVolcanoRSS.xml",
      "native_bulletin_url": "https://www.ct.ingv.it/index.php/monitoraggio-e-sorveglianza/prodotti-del-monitoraggio/bollettini-settimanali-multidisciplinari",
      "vona_url": "https://www.ct.ingv.it/index.php/monitoraggio-e-sorveglianza/prodotti-del-monitoraggio/comunicati-vona",
      "note": "The GVP Etna archive page carries the full dated weekly archive (2000->present). Navigate to the week label given in each row to read INGV-OE's own words for that date. The native INGV bulletin (PDF) and VONA pages are the primary INGV documents; GVP is the clean machine-readable mirror we ingest."
    },
    "verdict_legend": {
      "agreed": "Our independent satellite-derived call matched INGV's own report for that week.",
      "added_info": "We added information INGV's routine weekly/visual product does not itself carry (e.g. 20 m lava-extent footprint, dNBR burn severity, fused multi-sensor thermal, InSAR LOS, a QA catch).",
      "missed": "We missed it or disagreed -- stated plainly with the reason.",
      "wildfire": "FIRMS-confirmed Sicily wildfire (the flank/Sicily mission, distinct from the volcano)."
    }
  },
  "events": [
    {
      "id": "aug2025_effusion",
      "date": "2025-08-21",
      "ingv_week": "20 August - 26 August 2025",
      "ingv_state": "effusive (lava + ash)",
      "ingv_source_url": "https://volcano.si.edu/volcano.cfm?vn=211060",
      "ingv_source_label": "GVP/INGV-OE weekly, 20-26 Aug 2025 (Etna 211060)",
      "adriz": "Flagged VOLCANIC effusive. S5P SO2 summit plume +3.5 DU; FIRMS NRT active fire on edifice FRP max 94 MW; S2 (33SVB) active-lava core 462 px = 18.5 ha footprint, 424 SWIR-saturated px.",
      "adriz_method": "S5P TROPOMI summit-vs-background SO2; FIRMS VIIRS/MODIS NRT; Sentinel-2 B12 SWIR saturation + Normalized Hotspot Indices (Element84 S2 STAC).",
      "adriz_timestamp": "2025-08-19..23 overpasses",
      "corroboration": "Independent: FIRMS NRT (94 MW), S2 lava-extent 18.5 ha, fused multi-sensor thermal (only 5-sensor-agreement day in the set incl. MODIS).",
      "verdict": "added_info",
      "verdict_text": "Agreed with INGV AND added a quantitative 20 m active-lava footprint (18.5 ha) the weekly bulletin's narrative does not provide."
    },
    {
      "id": "feb2025_eruption",
      "date": "2025-02-11",
      "ingv_week": "12 February - 18 February 2025",
      "ingv_state": "effusive (lava + ash), ACC RED that week",
      "ingv_source_url": "https://volcano.si.edu/volcano.cfm?vn=211060",
      "ingv_source_label": "GVP/INGV-OE weekly, 12-18 Feb 2025 (Etna 211060); VONA ACC raised to Red",
      "adriz": "Flagged VOLCANIC via FIRMS NRT on edifice (FRP max 43.9 MW). S2 (2025-02-12) active-lava core 255 px = 10.2 ha, 465 saturated px. S5P single overpass showed NO plume (-0.76 DU) -- honest single-pass dropout.",
      "adriz_method": "FIRMS VIIRS/MODIS NRT; Sentinel-2 SWIR lava-extent; S5P SO2 (single pass).",
      "adriz_timestamp": "2025-02-11..12 overpasses",
      "corroboration": "Independent: FIRMS NRT (43.9 MW), S2 lava-extent 10.2 ha.",
      "verdict": "added_info",
      "verdict_text": "Agreed on activity AND mapped a 10.2 ha lava footprint. HONEST: our single S5P overpass that day registered no SO2 plume (-0.76 DU) -- a known single-pass advection dropout, not a state error; the rolling-window prior absorbs it."
    },
    {
      "id": "jun2025_paroxysm",
      "date": "2025-06-02",
      "ingv_week": "28 May - 3 June 2025",
      "ingv_state": "paroxysm (2 June lava fountain)",
      "ingv_source_url": "https://volcano.si.edu/volcano.cfm?vn=211060",
      "ingv_source_label": "GVP/INGV-OE weekly, 28 May-3 Jun 2025 (Etna 211060); 2 Jun 2025 paroxysm",
      "adriz": "S2 caught the cooling flow at +2 days (2025-06-04, 6 active-core px) then MISSED it at +3 days (2025-06-05, 0 active-core px -- surface cooled below the high-T SWIR threshold).",
      "adriz_method": "Sentinel-2 SWIR active-lava core (B12>=4000, NHI_SWIR>0.20).",
      "adriz_timestamp": "2025-06-04 (+2d) and 2025-06-05 (+3d) overpasses",
      "corroboration": "Self-bracketed temporal sensitivity floor (no independent corroborator needed -- this is OUR miss).",
      "verdict": "missed",
      "verdict_text": "HONEST MISS. A short paroxysm (brief fountain, not sustained effusion) is detectable by a 5-day-revisit optical sensor for only ~2-3 days; by +3 days the incandescent surface had cooled out of the SWIR detection band. Concrete proof S2 cannot be the alert layer for short-lived events."
    },
    {
      "id": "aug2024_paroxysm",
      "date": "2024-08-15",
      "ingv_week": "14 August - 20 August 2024",
      "ingv_state": "paroxysm (lava + ash)",
      "ingv_source_url": "https://volcano.si.edu/volcano.cfm?vn=211060",
      "ingv_source_label": "GVP/INGV-OE weekly, 14-20 Aug 2024 (Etna 211060)",
      "adriz": "Flagged VOLCANIC. S5P SO2 summit plume +5.08 DU (2024-08-13); FIRMS NRT on edifice FRP max 33.9 MW. Our per-overpass label read 'strombolian' where the weekly window peak was 'paroxysm'.",
      "adriz_method": "S5P TROPOMI SO2; FIRMS NRT.",
      "adriz_timestamp": "2024-08-13 overpass",
      "corroboration": "Independent: S5P SO2 plume + FIRMS NRT both positive same window.",
      "verdict": "agreed",
      "verdict_text": "Agreed on volcanic-activity-present (both SO2 plume and FIRMS thermal positive). HONEST: our per-overpass severity label ('strombolian') is one level below the weekly peak ('paroxysm') -- a weekly-vs-daily granularity difference, not a sensing disagreement."
    },
    {
      "id": "nov2023_qa_catch",
      "date": "2023-11-20",
      "ingv_week": "22 November - 28 November 2023",
      "ingv_state": "Strombolian, lava overflowing crater from 24 Nov; paroxysm 1 Dec",
      "ingv_source_url": "https://volcano.si.edu/volcano.cfm?vn=211060",
      "ingv_source_label": "GVP/INGV-OE weekly, 22-28 Nov 2023 (Etna 211060)",
      "adriz": "Our internal activity-label table coded this date QUIESCENT (a labelling error). The S2 detector independently flagged 50 high-T + 33 SWIR-saturated px at the summit -- i.e. it caught OUR own mislabel and matched INGV's report.",
      "adriz_method": "Sentinel-2 SWIR lava-extent (detector-vs-label QA cross-check against GVP/INGV).",
      "adriz_timestamp": "2023-11-20 overpass",
      "corroboration": "INGV weekly (22-28 Nov 2023) confirms lava overflow -- the detector, not the label, was right.",
      "verdict": "added_info",
      "verdict_text": "The detector caught a labelling error in our own ground-truth table by agreeing with INGV's report over our label. A QA result that demonstrates the detector is independent of (and can correct) the label set."
    },
    {
      "id": "jun2026_ash_vona",
      "date": "2026-06-14",
      "ingv_week": "11 June - 17 June 2026",
      "ingv_state": "ash_emission, ACC raised Yellow->Orange (diffuse ash)",
      "ingv_source_url": "https://volcano.si.edu/volcano.cfm?vn=211060",
      "ingv_source_label": "GVP/INGV-OE weekly, 11-17 Jun 2026 (Etna 211060); 14 Jun VONA",
      "adriz": "Ingested and classified the week ash_emission / ACC=yellow, matching the 14-Jun VONA. This is the single INGV weekly report in the last 120 days -- the cadence finding itself.",
      "adriz_method": "GVP RSS + Etna 211060 narrative ingest into machine-readable ingv_truth.json.",
      "adriz_timestamp": "2026-06-17 (weekly window end)",
      "corroboration": "Spot-audited against the real 14-Jun VONA (ACC Yellow->Orange).",
      "verdict": "agreed",
      "verdict_text": "Our classifier matched INGV's own VONA (ash, ACC raised). It is also the ONLY weekly Etna report in the last 120 days -- Etna was quiet Feb->early-Jun 2026, so GVP/INGV issued no weekly report; this is the activity-triggered-cadence latency finding, surfaced honestly."
    },
    {
      "id": "sicily_wildfire_caltanissetta_2025",
      "date": "2025-07-22",
      "ingv_week": "N/A -- wildfire, not volcanic",
      "ingv_state": "FIRMS-confirmed Sicily interior wildfire (Caltanissetta)",
      "ingv_source_url": "https://firms.modaps.eosdis.nasa.gov/map/#d:2025-07-22;@14.62,37.32,11z",
      "ingv_source_label": "NASA FIRMS active-fire archive (VIIRS NOAA-20 + S-NPP), 22 Jul 2025, 37.32N 14.62E",
      "adriz": "FIRMS NRT active fire FRP max 63 MW. Sentinel-2 pre/post dNBR burn scar: fire-pixel dNBR median 0.428 vs background 0.123; AUC(fire-vs-background) = 0.867 [0.842, 0.887] -- a statistically significant, co-located burn scar (moderate-high severity).",
      "adriz_method": "FIRMS NRT; Sentinel-2 B8A/B12 dNBR pre/post (Element84 S2 STAC) with FIRMS-co-located AUC + bootstrap CI.",
      "adriz_timestamp": "2025-07-22 (FIRMS) + pre 06-30 / post 08-09 S2",
      "corroboration": "Independent: FIRMS active fire (the fire truth) + S2 dNBR scar AUC 0.867.",
      "verdict": "wildfire",
      "verdict_text": "The flank/Sicily mission: a FIRMS-confirmed interior wildfire with an independently mapped 20 m burn-severity scar (AUC 0.867, CI excludes 0.5). One of 5/5 Sicily 2025 sites where S2 dNBR gave a significant co-located scar."
    }
  ],
  "scoreboard": {
    "source": "comparison.json (auto-generated; INGV-OE weekly bulletin via GVP 211060 vs ADRIZ satellite-derived calls)",
    "note": "Read the 95% Wilson CI, not the point estimate. Small, historical overlap (our S5P/EMIT campaign sampled ~30 dates 2023-2026).",
    "rows": [
      {
        "dimension": "Volcanic activity present (our SO2-plume OR on-edifice FIRMS fire vs INGV non-quiescent week)",
        "independent": true,
        "agree": 19,
        "n": 19,
        "rate_pct": 100.0,
        "ci95_pct": [83.2, 100.0],
        "caveat": "RECALL-ONLY: the graded overlap contains 0 INGV-quiescent weeks (no negatives tested), so this is recall, not specificity. Quote the lower CI bound (83.2%)."
      },
      {
        "dimension": "SO2 plume vs INGV activity (independent S5P TROPOMI)",
        "independent": true,
        "agree": 3,
        "n": 5,
        "rate_pct": 60.0,
        "ci95_pct": [23.1, 88.2],
        "caveat": "n=5, wide CI. The 2 misses are single-overpass SO2-advection dropouts (2025-02-11, 2023-08-25), not classifier errors."
      },
      {
        "dimension": "FIRMS fire vs INGV thermal/effusive (independent)",
        "independent": true,
        "agree": 12,
        "n": 19,
        "rate_pct": 63.2,
        "ci95_pct": [41.0, 80.9],
        "caveat": "The 7 disagreements are mostly INGV summit-Strombolian weeks with intra-crater activity below FIRMS' satellite thermal floor -- a real sensor-floor gap, not an error."
      },
      {
        "dimension": "State label within 1 severity level (NON-independent transcription)",
        "independent": false,
        "agree": 9,
        "n": 19,
        "rate_pct": 47.4,
        "ci95_pct": [27.3, 68.3],
        "caveat": "Our label is compiled FROM GVP/INGV, so this measures weekly-vs-daily granularity, not sensing skill. Excluded from any accuracy claim."
      }
    ]
  },
  "model_proof": {
    "headline": "Volcanic false-alarm on INGV's OWN camera frames",
    "value_pct": 9.7,
    "ci95_pct": [4.5, 19.5],
    "n": 62,
    "set": "143-frame REAL bulletin-confirmed INGV-OE camera held-out set (MarkL02/ingv-etna-camera-historical, split=confirmation_heldout); 62 visible held-out volcanic/false-situation frames; 0 phash leakage vs training.",
    "operating_point": "At conf 0.40-0.50 the volcanic false-alarm drops to 4.8-6.5% with NO loss of wildfire recall (96.8% on the synthetic slice).",
    "framing": "Tested against INGV's own camera frames, labelled from INGV's own bulletins. The model is volcano-hardened: it learns lava/ash/degassing as their own classes so a flank wildfire is not confused with summit activity. Wildfire-vs-volcanic is resolved by geometry + off-vent independent corroboration, not by spectrum alone (which INGV's own literature says is spectrally impossible)."
  },
  "honest_limits": [
    "Where we MISS: FIRMS' satellite thermal floor does not catch summit-Strombolian intra-crater activity that INGV's ground/webcam sensors see (the 7 FIRMS-vs-thermal disagreements). S2 lava-extent misses a short paroxysm seen >2-3 days late (Jun-2025 +3d miss).",
    "RECALL-ONLY volcanic-activity metric: the graded overlap has 0 INGV-quiescent weeks, so our false-positive rate against INGV-quiet weeks is untested here. Quote the 83.2% CI lower bound, not 100%.",
    "Single-site: all volcanic evaluation is Etna (+Eolie cameras). Every rate is small-n (5-62); read the CI.",
    "Synthetic wildfire positives: real on-camera Etna wildfires are rare, so the camera model's wildfire positives are largely synthetic composites; only the real-only slice is the headline.",
    "Weekly-resolution oracle: INGV's machine-readable product (the GVP-mirrored weekly bulletin) is a WEEKLY summary -- it cannot resolve the sub-weekly timing our per-overpass FIRMS (~3-6 h) and S5P feeds do, so per-day vs weekly-window labels legitimately differ at episode boundaries.",
    "Image/PDF-locked INGV data: tremor-RMS and SO2-flux t/day series are published as plots/PDF, so we do not ingest those numbers; VONA per-event UTC and HOTSAT FRP have no stable public per-day endpoint we could find. These are data-access gaps, not algorithm gaps."
  ]
}