ops(company-monitoring): run Stage 1B — fourteen-day Exa/X live-operability tracer

## Parent

#6002

## Why this issue exists

The epic names Stage 1B as a promotion stage and #6015/#6016 activation waits on it — but no issue owned *running* the fourteen-day tracer. This issue owns the Stage 1B execution and its evidence.

## Decisions in force (2026-08-18)

- Bring-up portfolio: **50 companies** (#6603 decision log). The 500-company entitlement in #6002 is unchanged; this is promotion posture.
- Shared X spend cap: **$200/month**, console-confirmed 2026-08-18. Track A (#6654) consumes ~$115/mo at 15-min cadence; size the tracer scan cadence to fit the remainder. At 50 companies: hourly ≈ $105/mo (over budget combined with Track A); **2-hourly ≈ $60/mo fits with headroom**. Record the chosen cadence.

## What to run

Fourteen consecutive days of scheduled scans over the 50-company bring-up portfolio with both v1 discovery providers (Exa via #6008, X via #6656) live in tracer mode, publication path exercised, no customer-facing surfaces enabled.

## Acceptance criteria

- [ ] Fourteen consecutive days of scheduled scans complete with both providers live.
- [ ] Zero wrong-company publications — a single occurrence stops promotion (epic stop condition).
- [ ] Zero false quiet: coverage states honestly distinguish no-observed-events from capped, stale, failed, or unresolved for every scan window.
- [ ] Provider failures degrade to `provider_error` / partial gaps; the worker does not crash; lease fencing behaves. 50 companies ≈ 5 recent-search packs — verify single-lease behaviour warm and cold.
- [ ] Attributable Exa and X cost recorded per scan; total inside the caps; the 2M post-read cap is never approached.
- [ ] Evidence recorded as aggregates and opaque IDs only — no portfolio names, company claims, or evidence text.
- [ ] Pass/fail recorded with named product-owner sign-off.

## Blocked by

- #6008 (Exa discovery live in tracer mode)
- #6656 (xProvider promoted)
- #6012 (publication path — wrong-company publication and false quiet are defined against published state)

## Stop condition

Wrong-company publication, false quiet, or an unbounded provider/query path fails Stage 1B: #6015/#6016 stay dark and promotion stops. These are the epic's rollout-killing conditions — this tracer is where they are proven absent.

Review the parent issue #6002 and the blocking issues #6008, #6656, and #6012 before starting. Run the fourteen-day tracer for the 50-company portfolio with both providers live, recording only the specified aggregate evidence; done requires every acceptance criterion, including named product-owner sign-off.

Written by the indexing model from the issue text.