How it works · v2.1 Phase 1

Methodology

A standing analytical framework, a falsifiable conviction formula, and an explicit operational rulebook. Built around one belief: every trade competes for scarce capital.

§01The seven-stage framework

applied to every ticker, every event

Every event passes through seven sequential stages. Skipping a stage is not allowed; missing data is flagged as a DATA GAP. The output is a desk call, not a hint.

A
Stage A · Executive Verdict

Conviction / 10 · primary track · risk-reward summary

A single-paragraph desk call before any analysis loads. This locks the prediction before motivated reasoning sets in. Thresholds: Strong Buy ≥ 8 · Buy 6–7 · below 6 = Watchlist / Avoid.

B
Stage B · Earnings Edge

Quarter-beat probability AND guidance risk · scored separately

Beat-rate (last 8 quarters), surprise std-dev, estimate revision velocity, Polymarket pricing where available. Guide risk is scored independently — "beat-and-light-guide" is the single most common fade pattern in premium-multiple names (ZS, MDB confirmed n=3).

C
Stage C · Market Positioning

Priced-in score X/10 · standalone metric

Run-up exhaustion (5d/20d), IV rank, options crowding, analyst PT vs current price, insider distribution (90d). When the stock is already trading 25% above the average analyst PT, the bar to beat is the price itself, not the consensus EPS.

D
Stage D · Technical Structure

Levels, regime, dealer gamma context

52-week structure, ATR-anchored stops, SPY GEX context (positive gamma = vol-suppressive = rallies fade-prone). Sector-relative strength. Daily mark vs trigger reference for divergence flag.

E
Stage E · Trade Construction

Mandatory catalyst chain · ≥3 independent links

If the desk can't articulate three independent catalyst links (fundamental + technical + flow), conviction drops automatically. The catalyst chain is the brief. Stops are implied-move-anchored or ATR-anchored, never round-number-based.

F
Stage F · Risk Review

Concentration · correlation · regime dampener

Effective conviction = max + 0.5·(sum − max), correlation-adjusted (Phase 8). Concentration flag triggers below 80% raw-to-effective ratio. Regime context (STAGFLATION_HOT, RISK_OFF) dampens long-duration growth into rising-yield tape.

G
Stage G · Final Desk Call

Dual-track conviction · Fade-Risk referee · explicit action

Two scores published: hold-through (carry to event) and pop-sell (exit into spike). A Fade-Risk referee picks the primary track. Action is explicit — size, stop, trigger, and rule that would change the call.

§02Conviction formulas

falsifiable, cap-bounded
Hold-through conviction
conv_hold = round1( E[R] · 2 + P(beat) · 4 + base_floor )
— capped at 5.6 for premium-multiple BEAT_AND_GUIDE_FADE pattern (ZS/MDB rule)
— capped at 7.9 if P(beat) > 75% AND E[R] < 1.8
— MDB hard rule: never hold-through
Pop-sell conviction
conv_pop = P(initial-spike) · 4 + min(spike_mag% / 3, 3) + liquidity_floor
— liquidity_floor: 3 for high ADV / options liquidity
— dampened by Fade-Risk for already-elevated AH
Composite alpha (engine layer)
alpha_norm = 0.50 · base + 0.15 · fund + 0.20 · short + 0.15 · implied_hist
— L1 implied_hist_alpha bound ±0.20
— L4 catalyst_chain veto: cap alpha at 5.0 if links < 3
— L5 structural_penalty: −0.15
— emit at alpha ≥ 0.5 · fill gate at alpha ≥ 5.0

§03The Fade-Risk referee

10-point scale · directional both ways
High Fade-Risk (≥ 7)

Drift +10% into print · IVrk > 90 · stagflation regime · gamma-positive tape · close ABOVE avg PT · premium-multiple class. Action: primary track is POP-SELL. Exit into spike, do not hold through unless guide-raise is uncontested.

Confirmed pattern (n=3): DE 5/21 round-trip, MDB 5/28, ZS 5/26 (−20.8% AH).

Mid Fade-Risk (4–6)

Mixed signals · run-up modest · IVrk normal · sector-supportive context. Action: SPLIT call — both tracks have merit. Trim into strength, hold residual.

Confirmed (n=2): ROST 5/21 mild-fade-then-hold (both tracks WIN); ORCL hold-bias confirmed.

Low Fade-Risk (≤ 3)

Flat-to-mild run-up · narrow surprise band · clean beat streak · room to PT · sector tailwind. Action: primary track is HOLD-THROUGH. Carry to event, no stop tighten on drift.

Active example: AVGO 6/03 carry (8/8 beats, narrow band, 17% below avg PT).

The discipline

Every binary event is scored against BOTH tracks after print, regardless of outcome. Calibration adjustments are explicit, sourced, and logged. The 20-event backtest is the gate to v3 of the system.

§04Technology stack

data, compute, delivery
Data · Fundamentals + EPS

FMP Premium

Point-in-time fundamentals, analyst PT distribution, earnings calendar, EPS history. Cross-validated against Yahoo PIT collector to catch the FMP annual÷4 EPS artifact (affects ~82/109 tickers).

$69 / month
Data · Options + Flow

Unusual Whales Basic API

IV rank, implied move, GEX, dark-pool prints, short interest, institutional delta, options chains, net premium flow. Six engine functions integrated into build_composite.

$150 / month
Data · Point-in-Time

Yahoo PIT collector

Task-scheduled (6h cadence) point-in-time fundamentals as Zacks replacement. Quarterly periods, revision diffs after 2nd daily run. 351 success / 320 empty / 0 errors at last collection.

$0 (in-house)
Compute · Language Model

Qwen3-32B-FP8 on RunPod

Self-hosted on demand. Runs the prompt routing and composite scoring pipeline. Ephemeral URL per session.

on-demand
Storage · Registry

SQLite PIT database

Single-file SQLite at E:\PIT data\pit.sqlite. Universe of ~99–109 large-cap tickers. Signal registry with completion flags, 150-min staleness timeout.

$0 (in-house)
Delivery · Dashboard

Flask + Cloudflare Tunnel

Flask dashboard on port 5057, fronted by Cloudflare Tunnel at signals.pinnacledominion.com. Discord + Telegram webhooks for fill alerts and regime-change events.

$0 (in-house)

§05Standing rules

non-negotiable · enforced by code
Rule 01 · Capital scarcity

Every position must outperform the next-best alternative on a falsifiable score. Conviction below 6 is watchlist, not capital deployment. No portfolio slot exists by default.

Rule 02 · Catalyst chain

If the desk can't articulate three independent catalyst links (fundamental + technical + flow), conviction is capped. No exceptions, no narrative-only entries.

Rule 03 · Implied vs realized

No ticker is graded without comparing the option-implied move to its trailing realized move. The delta is the edge.

Rule 04 · Journal authoritative

The journal's OPEN POSITIONS header is the authoritative source on conflicts. Daily report or trigger-prompt divergence is flagged; the journal's position is booked.

Rule 05 · Post-mortem

Every binary event is scored against both tracks after print, regardless of outcome. Calibration adjustments are explicit and sourced.

Rule 06 · MDB hard rule

Never hold-through on MDB. The pattern is too clean (post-print fade observed across multiple events). Hold-through is policy-banned for this ticker.