chore(0213): complete and archive — deployed and verified on prod

[karczuRF]

Summary

• Task 0213 deployed and verified on prod, status completed, moved to archive/
• Both stacks in order — EventBridge (probe binary) then Observability (alarm text) — so the metric was correct before the text described it. Observability alone would have left the ladder blind, which is 0204's own documented mistake
• Verified by measurement, not deploy exit status. A forced invocation of prices-production-rollup-freshness-probe returned:

field	value	reading
peg_scanned	692	✅ the direction really reads _1m over 48 h — the whole fix
peg_applied	0	matches the pre-deploy prod measurement
stranded_scanned	512	_1h untouched, 7-day window
stranded	9	expected, and climbing — see below

• CleanupRule verified DISABLED before and after the EventBridge deploy

⚠️ The 0141 stale-asset trap was live

The binary sitting in target/lambda/rollup-freshness-probe/ was three hours older than the source. Deploying without rebuilding would have shipped the pre-0213 probe — the one that reads _1h for the peg direction — and every signal would have reported success.

The discriminator, now recorded in the task file:

strings target/lambda/rollup-freshness-probe/bootstrap | grep -c usd-peg-applied
#   old build -> 0   (it says "usd-sanity")
#   new build -> 3

Two corrections recorded for the next reader

• scanned is rows MATCHED, not rows READ. A pre-deploy expectation put stranded_scanned at ~1,000,000 by quoting the EXPLAIN ESTIMATE read figure as if it were the match count — three orders of magnitude out. The existing _1h note already recorded the right relationship ("returns 423 rows but reads ~1.37M").
• stranded = 9 will climb, and that is correct. 0182's repair wrote the coarse tiers to a high-water mark of 2026-08-18 12:00; as the 48 h grace boundary moves past it, more _1h candles fall into the unpriced-past-grace window. It stays latched until 0209. ⛔ Do not widen the grace to quiet it.

Still true after this lands

⚠️ A green usd-peg-applied is not evidence the USDT leg is healthy. All rows in the window are unpriced, so the direction reads 0 for want of input, not want of defects — usd-stranded is what carries the darkness. It stays vacuous until enrichment drains far enough forward to reach a 48 h window, which is gated on 0111.