[mspinola]

Summary

Two changes, both the same decision: agi's /market is the authority on what the four M-106 exposure reads are; this page decides how they are drawn. Recorded at the top of market_internals, because cot-analyzer cannot import agi (private repo, public consumer) and so a restatement that drifts is a defect here rather than a difference of opinion.

1. The credit leg reads the dividend-adjusted series

JNK distributes about 6.7% a year in monthly instalments of roughly 0.56% of its price, so every ex-dividend date puts a notch in the raw price that has nothing to do with credit, and the 20-day average test reads it as a breakdown. The comment justifying the split tier had the mechanism backwards: the notch is a discontinuity the trailing average smooths over 20 sessions, so it opens a gap rather than cancelling.

raw vs adjusted verdict disagrees, last year	24.6% of sessions
disagreements over two years	98
of those, raw says risk off and adjusted says risk on	98
the opposite direction	0

One-directional because a recurring downward notch can only push a price under its average. JNK has never made a capital-gains distribution, so the total tier is exactly the dividend-adjusted series. Caruso states the adjustment as a condition of the read (M-10a); agi's jnk_leg preserves the worked example where the raw series inverted the July 2026 conclusion.

Up/down volume and the asset table stay on prices as traded, checked rather than assumed: the notch flips the up/down classification on 0 of QQQ's sessions and 1 of SPY's over two years.

2. Two definitions follow agi

• Up/down volume turns at one, not at an invented 0.95–1.05 dead zone. agi's updown_volume calls one "the ratio's own definition" rather than a threshold, since the ratio is up volume over down volume. A session at 1.02 read "Balanced" here and "net buying" there. The heavy tiers stay as this page's own emphasis and move no verdict.
• The FOMO chart is a year, agi's zoom, not a month. A month shows the last wiggle; the zone visits the read exists to locate are only legible against a year of them. Month ticks, a thinner line for the density, caption says "Past year".

Rendering stays ours, per the page's own style: the credit card draws a price and its average where agi draws the gap, and the reads sit in cards rather than a gauge row.

Test plan

• Full suite 880 passed, ruff clean.
• New tests: the tier per read with the reasoning; the single-session regression (a gently rising price plus one distribution-sized drop flips the raw verdict and not the adjusted one); the centre line at one with no neutral band.
• Rendered against the live store and checked at card width.

The TradingView twin gets the credit fix in mspinola/npf#472.

🤖 Generated with Claude Code