Home/Projects/Conflict Collection
2.e · Flagship Project

Conflict Collection & Neutral Analysis Platform.

Ingest text threads, emails, screenshots, and notes. Normalize into conflict events. Run each event through multiple AI models. Surface the consensus and the disagreement. Build court-ready exhibits with a documented provenance chain. Targeted at co-parents, family-law attorneys, and mediators.

Stage
In development
Users
Co-parents · Attorneys · Mediators
Differentiator
Multi-model consensus = neutral
Output
Timeline · Dashboard · Exhibit packet
Eventual home
Dedicated subdomain / product
Why this exists

Conflict documentation in family-law contexts is fragmented across text threads, email, screenshots, voicemail transcripts, calendar entries, and people's memories. Legal exhibit work requires clean timelines, neutral framing, and verifiable provenance. Doing that manually is expensive — and lawyers bill for it. Doing it with a single AI model invites accusations of bias. Doing it with multiple models and surfacing their agreement gives the neutrality a structural basis rather than a vendor claim.

Workflow

Six stages, one through-line.

  1. Ingest

    Accept iMessage/SMS exports, Gmail thread exports, screenshot OCR, manual notes, calendar entries, voicemail transcripts. Each source gets tagged with its provenance — where it came from, when it was collected, whether it's verifiable against the original.

  2. Normalize into conflict events

    Every piece of ingested content becomes one or more event records: timestamp, parties, channel, content, attachments, tags. This is the canonical object the rest of the system operates on.

  3. Multi-model neutral analysis

    Each event is run through multiple AI platforms (Claude, GPT, Gemini). Outputs are structured and compared. Where models agree on a characterization — tone, behavior pattern, escalation marker — that's surfaced as consensus. Where they diverge, the divergence itself is shown. Neutral is emergent from the consensus, not asserted by the system.

  4. Dashboard

    Timeline view, party/relationship view, pattern detection — escalation, stonewalling, frequency shifts, response-time drift. The dashboard is built for a human reviewer (user or attorney) to scan quickly and drill down selectively.

  5. Exhibit builder

    Export court-ready exhibit packets — numbered, timestamped, redacted where needed, with a documented provenance chain for every item. This is the piece attorneys care about most: exhibits that don't get challenged on foundation.

  6. Case continuity

    Ongoing ingestion as the conflict continues, with pattern drift reporting over time and re-export of updated exhibit packets.

Data model

The conflict event record.

Everything the system does operates on a normalized event. Rough shape:

ConflictEvent {
  id: uuid
  timestamp: iso8601
  parties: [PartyRef]
  channel: enum { sms, imessage, email, screenshot, note, voicemail, calendar }
  source: SourceRef { raw_filename, ingest_timestamp, provenance_chain[] }
  content: {
    raw_text: string
    attachments: [AttachmentRef]
  }
  tags: [string]
  analyses: [
    { model: enum, prompt_version: string, output: structured_json, timestamp: iso8601 }
  ]
  consensus: ConsensusSummary | null
  exhibit_refs: [ExhibitId]
}
Target users

Who's actually using this.

Co-parents

Documenting an ongoing conflict for their own sanity and for future legal readiness. The timeline view and exhibit packets are the primary value.

Family-law attorneys

Receiving a clean, neutral-framed conflict record from their client instead of a chaotic drop of screenshots. Court-ready exhibits straight out of the platform.

Mediators

Understanding conflict patterns before the session — escalation markers, stonewalling frequency, response-time drift. Structural context without the advocate framing.

Status

Where the platform is today.

Done

Ingest prototype

Working prototype for iMessage/SMS and Gmail thread ingestion with provenance capture. Proven against a real case corpus.

Done

Multi-model consensus prototype

Claude + additional model runs against normalized events. Structured output comparison working.

In progress

Dashboard & exhibit builder

Web UI for timeline, dashboard, exhibit packet export. Design and data-layer scoping.

Next

Attorney pilot

Identify 1–3 family-law attorneys willing to run a live case through the platform. Priority: Austin/Central Texas.

Eventually

Productization

Own subdomain, SaaS pricing, HIPAA-adjacent data handling posture, expanded channel support.

Interested in the attorney pilot?

If you practice family law in Central Texas and would consider running one case through the platform as a pilot, I'd like to hear from you.

Contact about pilot Back to projects