Most “AI for disputes” promises answers. That’s the wrong job. In a serious dispute, you don’t need a chatbot’s opinion—you need evidence that stands up.

The real leverage is triage: turning a 10-gigabyte dump of emails, daily reports, schedules, and contract versions into clause-linked, searchable evidence in hours, not months. Then a human—your counsel, commercial manager, or arbitrator—applies judgment on a clean record.

This post shows how to run AI triage the right way, using Hevi to surface facts, link them to the clause, and keep the human in control.

The Dispute Reality (and Why “AI Answers” Don’t Help)

When contracts go sideways, the data swamp shows up fast:

• Years of emails and reply-alls
• Daily reports and attachments
• Meeting minutes with action items that never made it to a system
• Multiple contract versions and unlabeled “final-final” PDFs
• Schedules and rate tables that don’t match the variation trail

A black-box “answer” from AI won’t be accepted by the other side—or an arbitrator. What wins is a tight, clause-linked chain:

1. What happened
2. When it happened
3. Where it lives in the record (email/report/file path)
4. Why it matters (the clause and the time bar)
5. What you did next (or didn’t)

AI is best at steps 1–3. Humans must own steps 4–5.

The Goal: Clause-Linked, Searchable Evidence

Think of triage outputs as a dispute-ready index you can trust:

• Every finding cites the exact clause/sub-clause and provides a clickable link to the source page.
• Each event (delay, instruction, access issue, KPI miss) includes date/time, who said what, and attachments if relevant.
• All material changes between base vs executed contracts are highlighted and summarized.
• The entire set is searchable by clause, party, date, and topic, and exportable for your legal team’s preferred format.

No opinions. Just fast facts with sources.

The Hevi Dispute Triage Playbook

Step 1 — Pick the Scope and Freeze It

Create a dedicated Dispute Project in Hevi. Add:

• Executed contract + annexures
• Relevant versions (base, V1–V4, executed)
• Agreed schedule/rate tables
• The inbox export (MBOX/EML directory) and daily report PDFs
• Any third-party logs you’ll need (weather, access records, machine downtime)

Freeze this set. No new files until triage completes round one.

Step 2 — Map the Questions to Clauses

Before touching emails, outline what the dispute will turn on:

• Delay triggers and notice requirements
• Variation procedure and pricing rules
• LD triggers/cure periods
• Site access and client-cause obstruction
• KPI definitions and measurement method

Drop these into a CSV template in Hevi. This is your “question spine” to anchor all findings.

Step 3 — Run Base vs Executed (Material Changes)

Hevi compares the executed contract to the base:

• Highlights material changes
• Summarizes impacts in plain English
• Links every change to its clause/page

Why now? Because many disputes hinge on “we thought X” vs “we actually signed Y.”

Step 4 — Triage the Inbox and Reports

Point Hevi at the email archive and daily reports. It will:

• Detect events relevant to your questions (delay, instruction, scope creep, access, weather, KPI shortfalls)
• Extract facts (date/time, parties, subject, text snippets)
• Attach the clause link that governs each event
• Cluster by topic (e.g., “Haul road closure,” “Crusher downtime,” “Unplanned blast delay”)

You’ll get a sortable list of events with clause links—not AI “opinions.”

Step 5 — Build Chronologies That Cite Themselves

Ask Hevi to assemble timelines by topic:

• “Build a chronology for all access delays impacting production between 1 Mar and 30 Apr; show timestamps, parties, and clause references.”

Each row is fact-checked back to source with a click. That’s what arbitrators care about.

Step 6 — Convert Findings into Issues (Humans Decide)

For each cluster/chronology, a human makes the call:

• Issue statement: “Client-caused access delay affected production 12–19 Apr.”
• Evidence list: email IDs, report pages, photos, attachments.
• Contract basis: clause/sub-clause links and required notice timing.
• Compliance check: did we meet the time bar? If missed, note mitigation.
• Quantum path: use schedule/rates and logged hours for rough order impacts.

Hevi prepares the bundle; you set the position.

Step 7 — Package the Record for Counsel or the Board

Export the index, chronologies, and evidence sets:

• Clause-linked index (CSV/XLSX)
• Chronology PDFs per topic with source links
• Material changes summary
• Issue briefs (one-pagers) for fast review
• Optional: Data room structure mirrored from Hevi folders

Result: a clean, reviewable record that survives first contact with the other side.

What “Good” Looks Like (and How to Avoid Noise)

Do

• Keep the question spine tight. If it won’t matter in front of an adjudicator, it doesn’t matter now.
• Maintain one source of truth for versions and indexes.
• Use consistent labels: topic, clause, date range, party.
• Review the top 50 events first; 80/20 applies.

Don’t

• Search across the whole corporate archive “just in case.” Scope creep kills speed.
• Let AI brainstorm arguments. That’s for counsel.
• Mix drafts and finals. Freeze evidence sets per triage round.

Where the 5–6× Productivity Comes From

• Zero spelunking: Clause links and page anchors end the “hunt the paragraph” game.
• Auto-clustering: Related emails and reports group themselves by topic.
• Reusable spines: Your CSV question set is evergreen across disputes.
• Faster lawyer time: Counsel starts at a clean, clause-linked record—not a shoebox.

You’re not replacing legal expertise. You’re amplifying it.

Example: Turning “Email Noise” into an Admissible Story

Problem: Three months of intermittent crusher downtime. Contractor claims client-caused access issues; client says contractor maintenance failures.

Triage flow:

1. Map clauses: access, instructions, production relief, notice timing.
2. Inbox triage: cluster all “crusher,” “access,” “clearance,” “permit” subjects.
3. Chronology: build a timeline of each stoppage with dates, durations, parties.
4. Evidence links: cross-link to daily reports and photos.
5. Compliance check: were notices issued on time with required contents?
6. Issue brief: “Client gate access delayed maintenance crews on 6/12, 12/12, 18/12; notices sent on time for two, late for one; production relief calculated per SoR item 12.4.”

No theatrics. Just facts with clauses.

Security and Confidentiality (Plain English)

• No model training on your data. Your documents aren’t used to train public models.
• Project-level isolation. Evidence and indexes are scoped to the Dispute Project with permissions and logs.
• Audit trail. Every finding has a timestamp, source path, and approver.
• Controlled exports. Hand counsel clean bundles without sharing the entire system.

You speed up without compromising privilege.

Operating Norms That Keep You Safe

1. Clause or it didn’t happen. No answer goes forward without a clause link.
2. Humans own positions. AI prepares bundles; people write arguments.
3. Freeze evidence per round. No moving targets while counsel reviews.
4. Label everything. Topic, clause, party, date range—every time.
5. Weekly 30-min review. Triage owner and counsel sync on top issues and gaps.

These rules keep speed and defensibility in balance.

What to Pilot This Month (Low Lift, High Return)

• Choose one live dispute or early warning. Limit scope to 60–90 days of data.
• Load the executed contract + inbox + reports into a Dispute Project.
• Run the question spine (10–15 rows in CSV).
• Produce three outputs:
  1. Clause-linked material changes summary
  2. One topic chronology with evidence links
  3. One issue brief approved by Commercial/Legal

Share the three outputs with your executive sponsor. If they get through them in 10 minutes and ask for more, you’re on the right track.

Why This Matters for Mining & Construction

• Disputes are data-heavy and time-poor. Triage converts bloat into signal.
• Frontline truth lives in emails and reports. AI surfaces it; humans decide.
• Arbitrators need sources, not stories. Clause links win trust fast.
• Counsel is expensive. Give them a clean record and they’ll give you better odds.

AI’s job isn’t to argue. It’s to find and link the facts—so your people can argue well.

Want to See It on Your Case?

Bring a real dispute (or a live early warning). In a single working session we’ll:

• Map the question spine
• Compare base vs executed
• Triage the inbox + reports into clause-linked events
• Build one chronology and one issue brief you can send to counsel

Clause. Link. Evidence. Human judgment. That’s how you turn 10GB of chaos into a case you can win.

‍