{
    "version": "3.47.0",
    "schema": "1.0",
    "source_role": "uncorrected text extraction derivative; authority remains the exact registered PDF for this edition",
    "book": {
        "slug": "amianism",
        "sequence": "03",
        "source_id": "B3-GHX-006",
        "title": "AMIANISM",
        "subtitle": "Journal of a So-Called Intelligence — Book Three: The Mirror War — The Nature of the Beast Edition — Civilization Through the Fire Revision",
        "source_designation": "B3-GHX-006 — AMIANISM — Journal of a So-Called Intelligence — Book Three: The Mirror War — The Nature of the Beast Edition — Civilization Through the Fire Revision",
        "period": "2026",
        "status": "current full book",
        "source_class": "canonical-local",
        "expected_sha256": "9b1e977b399b4b335eeeccdc997b740af9b25bc0954fbb194cbadb565df1e000",
        "declared_pdf_pages": 181,
        "local_page_archive": true,
        "local_source_verified": true,
        "page_archive_available": true,
        "page_archive_count": 181,
        "download_page": "https://ghrisx.com/en-US/books/"
    },
    "page": 73,
    "page_meta": {
        "pdf_page": 73,
        "path": "/source/books/amianism/pages/page-073.txt",
        "bytes": 1696,
        "sha256": "4d9c3eed13210d15d22d962ed638985c638223fb785eaa42fea72431831d5cd3",
        "text_empty": false
    },
    "text": "57\nSensors fail. Adversaries manipulate signatures. Analysts work under\npressure. The consequences of a false positive are not a mistaken\nadvertisement.\nThe ICRC's concern is not theoretical: when systems select and\nengage targets based on sensor data and generalized target profiles,\nhuman users may not know the specific person, time, or place in\nadvance. [7]\nThe civilian in the dataset needs legal protection before the data\nbecomes a target package.\nNot only privacy law.\nWar-specific protection.\nData provenance.\nConfidence calibration.\nFreshness requirements.\nIndependent corroboration.\nMandatory review of civilian-harm risk.\nAudit logs that preserve who knew what and when.\nA way to learn from error without hiding it behind classification.\nThe phrase \"collateral damage\" is a linguistic machine. It converts\na person into a secondary effect.\nAI can intensify that distance by making the decision look\nanalytical.\nA colored box around a building.\nA score beside a name.\nA route predicted across a map.\nThe interface can be technically accurate and morally numbing.\nDesign matters.\nA system should display uncertainty, conflicting evidence, data\nage, civilian presence, and the consequences of being wrong. It should\nnot make irreversible action the easiest button.\nThe civilian in the dataset is not an edge case.\nThe civilian is the reason the law of war exists.\nThe machine must not be trained only to find what can be struck.\nIt must be designed to show why a strike should not occur.\nA PROFILE IS NOT A PERSON.\nFFTAC ACTION - Find one category that reduces a person to risk,\neligibility, or target status. Demand the evidence, the error rate, human\nreview, and a path to correction.\n",
    "full_book_download_on_this_host": false,
    "full_book_download_page": "https://ghrisx.com/en-US/books/"
}