Running spec conflict detection

Spec conflict detection is one of the highest-leverage features Buildra offers. It runs across your spec sections, finds places where two parts of the document contradict each other, and surfaces the conflict before it costs you a re-do. Here is how to use it.

What it finds

The detector looks for three kinds of conflict:

• Direct contradictions.Section 09 5113 calls for 5/8″ Type X gypsum board; Section 06 1000 calls for 1/2″.
• Schedule mismatches. The door schedule says one hardware set; the door spec narrative says another.
• Plan-vs-spec conflicts. The plan annotation says one product; the spec section says another.

Running the detector

1. Go to /projects/<id>/spec-conflicts.
2. Click Run detection. The job starts in the background.
3. Detection takes 3–10 minutes depending on spec size. You will get an email when it is done.
4. When complete, the page populates with a sorted list — highest severity first.

You can also run detection automatically on every new spec upload by toggling Auto-detect on upload at the top of the page.

Reviewing results

Each detected conflict shows:

• A short description of the conflict in plain English.
• The two source locations (section number + page, or plan sheet + area).
• A severity rating (High, Medium, Low) based on whether the conflict is likely to cause field rework.
• Suggested next action (open an RFI, flag the architect, mark as intentional, etc.).

Tip:Start with High-severity conflicts and work down. A typical 200-page spec returns 8–20 conflicts; usually only 2–4 are High.

Triaging conflicts

For each conflict, click the row to expand the detail panel. From there you have three actions:

1. Open RFI. Pre-populates a new RFI with the conflict text and links to both source locations. One click and you are 80% to a written RFI.
2. Mark as resolved. Use this when you have confirmed the right answer (with a CCD or RFI response). Captures the resolution rationale for audit.
3. Mark as intentional. Sometimes the architect intends two specs to coexist (e.g., performance vs. prescriptive spec). Flag it as intentional with a note and it stops surfacing.

False positives

The detector aims for high recall — it would rather flag a non-conflict than miss a real one. Expect roughly 10–20% false positive rate. Marking false positives as intentional teaches the model your project's conventions and reduces noise over time.

When to re-run

Re-run detection after any of these events:

• A new spec section is uploaded.
• A revised plan set lands.
• A major addendum drops.

Detection results are versioned, so you can compare across runs to see what is new, what was resolved, and what is still pending.