Case study · Jurisdiction DE · Anonymized

    One monitored account. 131,781 comments. A deadline-sorted caseload.

    A German public figure's Instagram account, monitored continuously by Finium. The engine analyzed 81,101 comments, isolated 5,951 that may potentially violate German law (subject to legal review), traced them to roughly 4,160 distinct authors, and surfaced a prioritized, advisory-deadline-sorted work-list for counsel — triage that is effectively impossible to do by hand.

    Figures from a live read-only production audit, July 2026. All client and author identities removed.

    81,101

    Comments machine-analyzed

    of 131,781 collected

    5,951

    Flagged as potentially unlawful

    pending legal review

    896

    Repeat harassers identified

    ≥ 2 flagged comments each

    39

    Ready-to-review perpetrator cases

    31 statute-mapped, advisory

    The account-analysis funnel

    From a raw comment feed to the people behind it.

    One account, 1,575 posts, monitored continuously. About one in every 22 collected comments was flagged as potentially unlawful — insult, defamation, threats, or incitement under German law, each pending qualified review. Behind those comments sit thousands of identifiable authors, and a core of repeat offenders whose conduct forms a pattern over time.

    No human team realistically reads 131,781 comments. The point of the funnel is not volume — it is that the caseload at the bottom is limited by review capacity, not by a shortage of potentially unlawful material.

    Comments · one Instagram account

    Comments collected131,781
    Machine-analyzed81,101 · ≈ 62%
    Flagged potentially unlawful5,951 · ≈ 1 in 22 collected

    Authors behind the flagged comments

    Distinct authors4,159
    Repeat harassers (≥ 2 flagged comments)896
    Persistent harassers (≥ 3 flagged comments)354
    Author bars use their own scale (max 4,159). Classifications are AI-assigned risk categories, each a candidate for lawyer review — not legal determinations.

    The pipeline

    Monitor → classify → map → prioritize → review.

    Every stage is deterministic and inspectable. The engine never makes a legal decision: it preserves material, assigns risk categories, maps candidate statutes, and computes advisory windows — then routes the highest-priority items to a human review queue where counsel decides.

    1. 01

      Monitor

      131,781

      comments collected

      Continuous capture across the account’s posts, preserved with source context before material changes or disappears.

    2. 02

      Classify

      5,951

      flagged potentially unlawful

      AI classification by harm type and severity — insult, defamation, threat, incitement — pending qualified review.

    3. 03

      Map statutes

      31

      cases statute-mapped

      Each case mapped to the German statute it may potentially violate (e.g. § 185 StGB), deterministically from risk categories.

    4. 04

      Prioritize

      16

      live advisory windows

      Advisory filing windows computed per case, anchored conservatively to the comment’s post date, sorted by urgency.

    5. 05

      Review

      120

      items queued for counsel

      A reason-coded queue for lawyer review. The lawyer starts at the answer, not the raw feed — and stays the legal actor.

    From analysis to caseload

    A conservative slice became 39 cases.

    A narrower, client-attributed slice of the surface — 1,516 comments — was pushed through statute mapping and case generation. Even that conservative slice produced 39 ready-to-review perpetrator cases, 31 of them mapped to a specific German statute with an advisory filing window. The account-wide numbers above show how much higher the ceiling sits: roughly 900 repeat harassers against 39 promoted cases.

    Every mapping reads “may potentially violate,” and every deadline is advisory — produced deterministically from risk categories as a prioritization signal for counsel, never as an automated legal decision.

    Attributed slice · statute mapping

    Comments in attributed legal pipeline1,516
    Flagged potentially unlawful967
    Distinct authors833
    Perpetrator cases generated39
    Mapped to an actionable statute (advisory)31

    Potential statutes · 39 cases (advisory)

    Potential statuteCases
    § 185 StGB — Beleidigung (insult)26
    § 186 StGB — Üble Nachrede (defamation)5
    § 130 StGB — Volksverhetzung (incitement)1
    No actionable statute (classified not relevant)7

    Advisory-deadline posture · 31 statute-mapped cases

    • Open, inside ~30 days9
    • Open, ~31–47 days7
    • Past advisory criminal window15
    Windows anchored conservatively to each comment’s post date. Under § 77b StGB the clock runs from the victim’s knowledge, so counsel may find ‘expired’ items still open — and civil remedies may remain either way. One § 130 case carries no filing window at all (prosecuted ex officio).

    The review queue

    The lawyer starts at the answer, not the feed.

    The engine routed 120 highest-priority items into a reason-coded review queue for this one client. Each item says why it is there — severity consensus, an approaching advisory window, or a lapsed criminal window where civil remedies may remain.

    The largest reason code is the sobering one: for 86 items the advisory criminal window had already lapsed by review time. Without continuous monitoring, potentially actionable complaints quietly time out. Speed is the product.

    Queue reasons · 120 items

    Criminal window lapsed — civil-only review86
    High-severity consensus32
    Advisory deadline warning2
    Reason codes are routing signals for counsel, not legal conclusions. Whether any item supports a complaint, a civil claim, or no action is a decision for qualified lawyers.

    Method and limits

    What these numbers are — and are not.

    01

    Advisory, not adjudicated. Every statute mapping and deadline is a candidate for lawyer review, produced deterministically from AI risk categories. Nothing here is an automated legal decision, and no content is asserted to be unlawful.

    02

    Deadlines are anchor-sensitive. Windows shown are anchored to each comment’s original post date — the conservative reading. Under § 77b StGB the legal clock runs from the victim’s knowledge, so counsel may set different dates. Treat every window as an illustration of prioritization, not a legal fact.

    03

    Evidence tier: the captures behind these cases are reconstruction-tier bundles — database-reconstructed documents rendered to PDF/PNG — not live-page forensic captures, and not yet a full tamper-evident Evidence Pack with root-hash manifest and independent timestamp.

    04

    Repeat-offender data (about 900 accounts with two or more flagged comments) supports a pattern-of-conduct narrative for counsel to assess. The cases shown here each cover a single incident; multi-incident consolidation is on the roadmap, not shipped.

    05

    This page is anonymized aggregate reporting from a production audit. It is not legal advice, and it does not promise any court, platform, investigative, or enforcement outcome.

    Related resources and next steps

    FINIUM LEGAL

    How many cases are sitting in your client's comments?

    Send one public account or representative matter and see the kind of prioritized, source-aware caseload Finium prepares for counsel.