Verdict: Options Picks structure selector (M-005) is rated “NOT VALIDATED (structure, direction, strikes) — FIT FOR LIMITED USE (vol-regime legs only)” in its validation review dated 3 Sep 2026.
Key result: Structures tested vs forward P&L: 0.
Key result: Live scoreboard hit rate: 33.3% (n=6 matured).
How it was tested: the record does not state separate training and test periods; out-of-sample design: none.
Horizon: 14-70 DTE; the scoreboard marks each structure at expiry intrinsic. Coverage: US optionable names passing OI ≥5,000, ATM spread ≤12%, ≥$10M/day.
Forward record: 4 of 21 published lists have reached their 30-session horizon. Average pick return minus SPY's over the same dates: -6.74 percentage points (standard error 1.51). 33% of the matured picks beat SPY over the horizon.
Stage: Production. Version v1; the hub record is as of 6 Nov 2026.
The verdict
NOT VALIDATED for structure choice, direction and strikes — nothing here has been tested against forward P&L. FIT FOR LIMITED USE for the vol-regime legs, which carry audited edges and decide only whether to buy or sell premium.
Development summary: The vol-regime step is the only part with measured support; the direction lean, the structure map, the strike choice and the wing/gamma adjustments are reasoned, not measured. No structure has ever been tested against forward P&L.
What the review allows
The buy-premium / sell-premium call implied by the vol regime (iv_pctile and vrp), which is audited.
The gates (liquidity, spread, DTE) as a hard exclusion list.
The published structure as a starting proposal for hand review, read together with px_src and the caveats string.
What the review rules out
Rely on the structure map, the direction lean, the strike choice or the wing/gamma adjustments as measured signals — none has been tested.
Read R:R as a ranking or as an expectancy; it is a consequence of the structure and the engine never sorts on it.
Treat theo-priced rows (px_src = theo) as fillable prices.
Read the scoreboard as evidence in either direction at n=6.
Assume the hub's forward tracking scores the trade: it follows the UNDERLYING, not the structure.
What it predicts
Item
Detail
Question it answers
Not which stock — which OPTION. Given a name whose vol is cheap or rich, which structure, expiry and strikes express the view, and what does it cost?
Method
Gated structure cascade
How it works
Gates → vol regime (iv_pctile, vrp) → direction lean (SMA50/200) → structure table → delta-targeted strikes snapped to the gamma peak → quote-or-theo pricing.
Inputs
IV percentile (cheap / rich), Volatility risk premium (VRP), Trend lean: SMA 50/200, OI gamma-peak strike, Wing edge (skew)
14-70 DTE; the scoreboard marks each structure at expiry intrinsic
Coverage
US optionable names passing OI ≥5,000, ATM spread ≤12%, ≥$10M/day
Markets
US
Model type
Model · Rule
Stage
Production
Version
v1, first recorded 14 Aug 2026
Validation evidence
Training data: not stated. Test data: not stated. Out-of-sample design: none.
Headline results from the model's development record, as the hub publishes them.
Measure
Result
Compared with
Note
iv_pctile ≤10 buy premium
+1.91 vol pts
+1.65 always-sell baseline
audit harness T-006, run as_of 2026-09-24, 175,722 symbol-days; CI [+0.16, +4.03] (was [+1.57, +4.53] as_of 2026-07-31)
vrp > 8 sell premium
+1.28 vol pts
same baseline
CI [+0.54, +1.98] — the leg the rich-vol structures rest on
vrp < 2 buy premium
+0.55 vol pts
same baseline
CI [+0.10, +0.92] — the cheap-vol leg
Direction lean (close vs SMA50/200)
not measured
edge-lab re-validation
momentum and MA-stack inputs were flat or sign-flipping out of sample; the engine labels the lean as a lean
Structures vs forward P&L
0 tested
7 templates shipped
the tab tracks itself forward instead; scoreboard n=6 matured, hit 33.3%, median P&L -62
Forward record: 4 of 21 published lists have reached their 30-session horizon. Average pick return minus SPY's over the same dates: -6.74 percentage points (standard error 1.51). 33% of the matured picks beat SPY over the horizon.
What the validation review found
Is the design sound? The decision order is the right one and is honest about its own evidence: the engine puts the measured step first and labels every later step as reasoning. Refusing to rank on R:R is the single best design choice here. The gap between 'the regime is measured' and 'therefore this structure' is where the model's risk lives, and the code says so in its own docstring.
Data quality. Three upstream feeds with known holes: sampled OI (T-005) undercounts the book, wings are a diagnostic (M-007), dealer gamma rests on a sign assumption (M-010). straddle_edge is null on many rows, which silently blocks long_straddle; atm_spread_pct is null on rows that still pass the spread gate.
Backtest. None at structure level. The inherited vol-regime evidence is one regime, ~11 months, overlapping windows, survivorship-filtered IV history. The forward scoreboard has 6 matured picks — 33.3% hit, median -62, median -77.2% of risk — which is far too small to read.
Against simple benchmarks. No benchmark structure is priced alongside the chosen one (e.g. the same view as an outright vs a spread), so the structure decision has nothing to beat. The obvious benchmark — always express the regime as the simplest structure — is cheap to add and would make the map testable.
Independent replication. 12 live picks re-ranked from the feed with an independent implementation of conviction; order and values reproduce. Scoreboard figures were read from the feed, not recomputed.
Sensitivity. Untested. The thresholds sit on audited lines (10th/90th percentile, VRP 2/8), but nothing measures how the pick set changes as they move, nor how often a wing/gamma adjustment flips the structure.
Monitoring. The append-only log plus score_forward/scoreboard is the monitoring, and it is the right shape; what is missing is a per-structure breakdown with a minimum-n gate before any number is displayed as a result.
Limitations
No structure has been tested against forward P&L — the mapping from regime × direction to template is judgement.
Direction is explicitly unmeasured, and it selects half the table (bull vs neutral vs bear).
Wing and gamma adjustments are Tier B: sound reasoning, no measurement.
Pricing falls back to Black-Scholes theo when no book exists, so cost/breakeven/R:R can be theoretical rather than fillable, and slippage on a multi-leg order is not modelled.
The scoreboard is tiny (n=6 matured, hit 33.3%, median -62) and mixes structures; it cannot separate structure choice from vol regime yet.
Sampled OI (T-005) undercounts the real book, so the liquidity gate is conservative in an unquantified way.
Questions
Is the Options Picks structure selector validated?
Its validation review dated 3 Sep 2026 rated it NOT VALIDATED for structure choice, direction and strikes — nothing here has been tested against forward P&L. FIT FOR LIMITED USE for the vol-regime legs, which carry audited edges and decide only whether to buy or sell premium.
What does the Options Picks structure selector predict?
It answers: Not which stock — which OPTION. Given a name whose vol is cheap or rich, which structure, expiry and strikes express the view, and what does it cost? Output: ≤6 structures/session: legs, strikes, expiry, cost, breakeven, R:R, px_src. Horizon: 14-70 DTE; the scoreboard marks each structure at expiry intrinsic. Coverage: US optionable names passing OI ≥5,000, ATM spread ≤12%, ≥$10M/day.
How was the Options Picks structure selector tested?
Training data not stated, test data not stated, out-of-sample design: none. Structures tested vs forward P&L: 0; Live scoreboard hit rate: 33.3% (n=6 matured); Vol-regime legs (audited): +0.55 to +1.91 vol pts. Forward record: 4 of 21 published lists have reached their 30-session horizon. Average pick return minus SPY's over the same dates: -6.74 percentage points (standard error 1.51). 33% of the matured picks beat SPY over the horizon.