# FIN-1 — nothing owns sizing when half-Kelly and the vol targeter disagree, and the edge estimate Kelly is sizing on is statistically indistinguishable from zero

status: open · raised 2026-09-10 (external financial review F2, re-derived — see `research/2026-09-09_financial_review_verification.md`) · class: risk-model coherence · judgement: **YES, on the estimator argument only** · effort ~0.25 pd (a decision memo, not code) · horizon: before any real-money deployment

---

**Plain-language summary.** Two risk systems answer different questions and
currently disagree. Half-Kelly says *the estimated edge is tiny, so size
tiny* — it sizes the book to roughly 11% gross. The VIX-scaled vol targeter
says *realized volatility is below the 12% target, so lever up* — it
multiplies by ~2.7 to reach 29%. Nobody has written down which one wins when
they pull against each other, or why.

**The argument that holds.** Half-Kelly requires an edge estimate you
believe. At Deflated Sharpe 0.611 the edge is statistically indistinguishable
from zero, and Kelly applied to a zero edge sizes to zero. The 11% gross is
therefore not information about how much edge exists — it is the noise in the
edge estimator. Sizing on that and then levering the result 2.7x compounds an
estimation error rather than managing a risk.

**The argument that does NOT hold, and must not be repeated.** The review
claimed the `max_leverage: 3.0` ceiling is "the binding constraint in the
current regime" — i.e. that an arbitrary constant is setting position size.
Measured over the last 20 sessions of `data/processed/portfolio_snapshots.jsonl`:
`vol_scalar` min 2.33, max 2.688, **at-cap count 0**. The cap has never bound.
The scalar sits at ~90% of ceiling, which makes this a forecast worth watching,
not an observation. Any build that opens by asserting the cap binds is building
on a false premise.

## Fix shape (propose-only)

A decision memo, pinned: which system owns sizing when they disagree, and why.
Explicitly NOT a change to `max_leverage` — moving the ceiling tunes a symptom
of a decision nobody has made. Candidate resolutions worth costing: vol-target
scaling floors at the Kelly gross when the edge estimate is indistinguishable
from zero; or the scalar is capped as a function of DSR rather than by a
constant; or the disagreement is declared acceptable with the reason written
down. Any of the three is better than the status quo, which is "whichever hits
its limit first."

## Kill criterion (pre-registered)

If the memo concludes the current behaviour is correct as-is, that is a
SUCCESS and the row closes with the reasoning recorded — the deliverable is a
written decision, not a change.

## Prior art

`leverage_cap_basis` (C1, 2026-08-01) identified the mechanism; this row is
the decision it deferred. Does not relitigate half-Kelly itself
(memory/strategic-posture: a variance regularizer, kept).