Jodie Analytics

Methodology

Built on a walk-forward validation, not a backfit

Every Jodie signal is generated by the same engine in production today. The performance numbers below come from running that engine across 12 months of historical data with no look-ahead — the same gates, the same filters, the same quality thresholds. No parameters were tuned after seeing the results.

316
Signals validated
US equities · 12 months
56%
Hit rate
Above 7% random expectation
+6.4%
Avg return per signal
At 2% fixed position sizing
3.65
Sharpe ratio
Max drawdown 3.3% · CAGR 52%

Scorecard by activation stage

Walk-forward run 6a9d09710236e6e7 · n=316 · universe 1,928 US equities · all stages included

No look-ahead · same engine as production
StageSignalsHit rateAvg returnSQ gateWhat it means
Broadening7056%+10.3%≥ 7Strongest alpha — confirmed propagation across the cohort
Building7656%+6.0%≥ 7Leader plus at least one confirmed follower
Mature12055%+4.7%≥ 7Broad participation established
Igniting5048%+1.2%≥ 8Early — no confirmed followers yet, requires stronger structure
All stages31656%+6.4%Sharpe 3.65 · max DD 3.3% · CAGR 52.1%

Returns measured from signal timestamp to lifecycle exit (theme goes inactive or leader loses structure). Position sizing: 2% fixed per signal, no leverage, cash otherwise. Hit rate defined as positive return at exit. Signals with SQ ≤ 6 excluded — that cluster is the only one with negative expected value (48% hit, −1.1% avg).

Signal quality score (SQ)

Every signal gets an SQ from 0 – 20. It is the sum of four model sub-scores, each 0 – 5: structural validity, propagation evidence, timing advantage, and narrative coherence. The score is transparent — you can see the components on each signal — and it is the primary quality gate used in the walk-forward.

Structure validity
Theme probability, cluster cohesion, reaction breadth. Penalised 25% if the theme class is noise or local move.
Propagation evidence
Reaction breadth, confirmed follower count, simultaneous impulses across the cluster.
Timing advantage
Correlation jump, breadth expansion pace. Capped when breadth is already very high — late entry is scored down.
Narrative coherence
Theme probability, cohesion, news alignment, and structural edge score. Penalised for low confidence or generic theme labels.

SQ bands and what they mean

14 – 20High conviction

Strong structure, breadth evidence, narrative alignment

10 – 13Tradeable

Passes all gates — core of the signal population

7 – 9Marginal

Valid but expect lower average returns

≤ 6Rejected

48% hit rate, −1.1% avg — negative expected value. Filtered.

How the engine works

Three layers run in sequence. The structural layer runs nightly. The live signal engine runs every 15 minutes during market hours. The signal quality score is computed at the point of emission and does not change after the fact.

1

Structural graph

A nightly graph encodes durable relationships across the universe — which names travel together, which clusters have real semantic cohesion, where sector bridges sit. This is the baseline that lets Jodie distinguish a genuine propagation move from correlated noise.

2

Live signal engine

During the session the engine runs every 15 minutes. It tracks relative move quality, intraday breadth, propagation evidence, and whether confirmers are arriving in sequence. A signal is raised when the structure meets the quality gate — not before.

3

Signal quality score

Every signal gets an SQ score from 0 – 20 built from four sub-scores: structure validity, propagation evidence, timing advantage, and narrative coherence. The gate is SQ ≥ 7 (≥ 8 for early-stage igniting signals). SQ ≤ 6 is rejected — it is the only validated negative-expectancy cluster in the dataset.

Limitations and scope

The validation is real but it has boundaries. These are not caveats added for legal cover — they are things that materially affect how you should use the product.

1

The scorecard covers US equities only, 1,928-ticker universe, May 2025 – May 2026. International and crypto signals are not yet walk-forward validated.

2

~316 signals per year across the full universe. This is intentional — the engine only raises signals that clear the quality gate. Frequency is not a goal.

3

Intraday theme signals (Intraday type) are scoped to the high-quality sub-population (member ≥ 5, rank score ≥ 100) with 6 weeks of live data — directional only, not paper-traded yet.

4

The signal quality score underestimates propagation edge contribution in historical replay because intraday lagged edges are not available for past dates. Live SQ values are higher than replay equivalents.

5

Jodie does not predict price targets, magnitude of moves, or entry/exit timing beyond lifecycle stage. It identifies propagation setups and whether structure is strengthening or failing.

Go deeper

The full walk-forward record — per-stage scorecard, signal funnel, SQ analysis, and exit mechanics — is on the track record page.