Replay of 09-27 16:30-19:30 ET (baseline run "W3 baseline", 224 calls) reproduced the 40-call Ch1 rollover sink. The bad links were almost all consensus tiebreak links (rules said new/orphan):
"Sam 1 ... flagged down for report of another four-car accident" -> linked (unit overlap)
idle resets on every link, so the 2.5h incident read as "2min ago"
Changes (llm_correlator.py only — affects both the cheap vote and the tiebreak):
_inc_summary adds open: (via _incident_span_minutes, sweep-safe) and calls: (substantive count)
_call_block shows reassignment
_RULES: dropped "a wrong link is cheap"; unit overlap alone no longer a link reason; explicit EXCEPT for calls that report a new event
_fmt_since clamps negative minutes
Risk: pushes toward splitting; follow-ups phrased as reports may split. Must re-score both answer keys (09-22, 09-21) in reuse mode before trusting it.
Refs #189.
Replay of 09-27 16:30-19:30 ET (baseline run "W3 baseline", 224 calls) reproduced the 40-call Ch1 rollover sink. The bad links were almost all consensus **tiebreak** links (rules said new/orphan):
- "Sam 1 ... flagged down for report of another four-car accident" -> linked (unit overlap)
- idle resets on every link, so the 2.5h incident read as "2min ago"
Changes (llm_correlator.py only — affects both the cheap vote and the tiebreak):
- `_inc_summary` adds `open:` (via `_incident_span_minutes`, sweep-safe) and `calls:` (substantive count)
- `_call_block` shows reassignment
- `_RULES`: dropped "a wrong link is cheap"; unit overlap alone no longer a link reason; explicit EXCEPT for calls that report a new event
- `_fmt_since` clamps negative minutes
Risk: pushes toward splitting; follow-ups phrased as reports may split. Must re-score both answer keys (09-22, 09-21) in reuse mode before trusting it.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
The tiebreaker forced ~all of the 09-27 rollover sink's bad links: idle resets
on every link so a 2.5h, 40-call incident read as fresh, and the rules treated
unit overlap as enough on a county channel where units work jobs back to back.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
You are not authorized to merge this pull request.
This pull request can be merged automatically.
View command line instructions
Checkout
From your project repository, check out a new branch and test the changes.
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Refs #189.
Replay of 09-27 16:30-19:30 ET (baseline run "W3 baseline", 224 calls) reproduced the 40-call Ch1 rollover sink. The bad links were almost all consensus tiebreak links (rules said new/orphan):
Changes (llm_correlator.py only — affects both the cheap vote and the tiebreak):
_inc_summaryaddsopen:(via_incident_span_minutes, sweep-safe) andcalls:(substantive count)_call_blockshows reassignment_RULES: dropped "a wrong link is cheap"; unit overlap alone no longer a link reason; explicit EXCEPT for calls that report a new event_fmt_sinceclamps negative minutesRisk: pushes toward splitting; follow-ups phrased as reports may split. Must re-score both answer keys (09-22, 09-21) in reuse mode before trusting it.
🤖 Generated with Claude Code
View command line instructions
Checkout
From your project repository, check out a new branch and test the changes.