One score. Traceable evidence. Versioned rules.

Methodology v0.03 - August 16, 2026

Pop Terminal's research model separates a slower-moving Base Conviction score from a bounded Event Pulse. Every published score identifies its model version, input coverage, as-of time, freshness, and supporting evidence. Conviction is an explainable research measure, not an expected return, probability of profit, or personalized recommendation.

The model-version label on each snapshot identifies the exact executable rules used for that result. Preserved snapshots continue to follow their own immutable model version; they are never silently relabeled during a rollout.

Historical v3.1 Base Conviction

Preserved v3.1 snapshots and current fallback scores use these five factors and original weights:

• Valuation - 25%: price relative to fundamentals and market expectations.
• Growth - 25%: revenue, earnings, income, and forward growth.
• Quality - 20%: margins, returns, cash conversion, and balance-sheet quality.
• Risk - 15%: a higher factor score means lower measured business and financial risk.
• Street - 15%: analyst tilt, target context, and revision momentum.

conviction-v3.1.0 first selects a certified model profile from the instrument, sector, industry, and revenue stage. Banks and brokers do not use generic interest coverage, net-debt-to-EBITDA, debt-to-equity, EV/EBITDA, or gross margin. Insurers exclude those industrial leverage and margin measures as well. Pre-revenue biotechnology companies do not use mature-company valuation or profitability multiples. When certified, class-appropriate observations are unavailable, Pop Terminal withholds the number instead of substituting an industrial template.

Every scored metric records its unit, applicability, fiscal period, period end, reporting currency, exact source, source as-of time, and transformation. Full precision is retained internally; the interface displays an appropriately formatted value.

Missing data and coverage

A missing factor remains null and is never replaced with an artificial 50. Available factor weights are renormalized to calculate the score. Coverage is the sum of the original weights represented by available factors. At least 60% coverage is required to publish a number; otherwise Pop Terminal shows "Insufficient data" and names the missing factors.

Forward valuation context

Forward P/E, PEG, and forward P/S use available consensus estimates and appear as display-only rows inside the existing Valuation panel. They are excluded from Base Conviction and Event Pulse. An unavailable estimate remains unavailable rather than being inferred.

Current Conviction v4 relative score

On covered, actively requested companies, the current ticker score uses the report-pinnedconviction-v4.0.0 model and exposes BASE 68 · RANK 92. It is calculated from immutable nightly peer-cohort snapshots. If an exact current v4 projection is unavailable, the ticker safely falls back to certified v3.1. Historical replay remains versioned, and portfolio and alert aggregates retain their stated model version until those surfaces complete their own cutover.

The v4 factor weights are:

• Momentum: 18%
• Growth: 18%
• Sentiment: 15%
• Street consensus: 12%
• Estimate revisions: 12%
• Valuation: 10%
• Quality: 10%
• Risk: 5%

V4 first converts certified underlying metrics into percentiles inside a peer cohort, combines those metric ranks into factors, then percentiles the weighted factor composite. Growth emphasizes revenue CAGR while retaining smaller earnings and net-income inputs. Quality combines gross margin, returns, and leverage; Risk combines leverage, interest coverage, and dilution. For a positive-growth company, Valuation uses trailing price to sales per point of revenue growth, so an investment-stage company is not assigned an automatic zero merely because P/E or EV/EBITDA is not meaningful. Base is the weighted factor-percentile composite; Rank is the second-layer composite percentile. Rank is relative, not an absolute forecast: in a market-wide drawdown the median company remains near 50 even when absolute conditions deteriorate.

Displayed v4 Confidence is RANK + CERTIFIED EVENT PULSE, clamped to 0-100. Pulse is never applied to the factor composite, because doing so would discard v4's second percentile. The projection is withheld rather than assuming a zero when an exact certified v3.1 Event Pulse snapshot cannot be linked.

Momentum is shown as a named sub-score because a price-driven factor can be reflexive. The selected input is volatility-adjusted six-month relative strength, chosen after it produced the better rank correlation in the historical holdout. Missing factors remain null, available weights are renormalized, and v4 is withheld below 60% coverage or when its report, snapshot, momentum selection, or knot-vector fingerprints cannot be reproduced.

V4 Rank bands are:

• High Conviction: 90-100
• Constructive: 75-89
• Balanced: 50-74
• Cautious: 25-49
• Low Conviction: 0-24

Historical v3.1 and fallback score bands

• High Conviction: 85-100
• Constructive: 70-84
• Balanced: 55-69
• Cautious: 40-54
• Low Conviction: 0-39

Bands describe model output only. They are not Buy, Sell, Hold, or suitability labels.

Event Pulse and displayed Confidence

Under v4, displayed Confidence equals Rank plus the exact certified v3.1 Event Pulse, clamped to 0-100. Under v3.1, displayed Confidence equals Base Conviction plus Event Pulse. Event Pulse is bounded to -10 through +10. A single direct event can contribute no more than -3 through +3, and each event category is capped at -4 through +4, except the dedicated price-action category, which is capped at -1.5 through +1.5. Syndicated copies count once.

Numeric impacts come from deterministic rules. AI may structure, classify, and explain evidence but cannot choose polarity or a numeric delta. A scored event must be tied to a verified issuer or regulator source, or to a publisher in the versioned established-media registry at its certified reliability tier. Multiple reports of the same economic fact count once. A provider label, social post, rumor, or generic documentation link is not sufficient scoring proof. Ambiguous, uncorroborated, and material-but-neutral evidence can remain visible with zero impact.

A release-certified relative-price observation may contribute through the Street/price-action lane when its exact ticker and SPY inputs, sessions, calculation, source timestamps, and payload fingerprint reproduce the displayed claim. Same-day excess return must differ by at least 1%; otherwise the stronger normalized 5- or 20-session signal is used. The raw impact is capped at -1.5 through +1.5, uses provider reliability 0.9, and decays over seven days. It describes observed price action and never claims that a headline caused it.

Direct and related-company evidence

v0.02 scores direct events about the issuer only. Customer, supplier, peer, and sector relationships may still appear as research context, but related-company read-throughs do not contribute to Event Pulse.

Decay and double-counting controls

The active event-decay-v0.02 rules decay impacts linearly over these windows:

• Earnings, guidance, company filings, insider events, and institutional filings: 30 days.
• Analyst revisions, rating changes, and price-target changes: 14 days.
• Ordinary direct-issuer news and price-action observations: 7 days.

When a base-factor refresh incorporates the same information, its remaining pulse impact is marked absorbed and removed. The post-decay contributions must sum exactly to the displayed Event Pulse and remain replayable.

Creator and social context

YouTube Insights and X Insights provide source-linked external commentary. That content is context-only and cannot change Base Conviction, Event Pulse, or displayed Confidence. An empty insight panel means no relevant monitored content was matched; it is not a negative view of the company.

What changed today

Today's change compares the current snapshot with the previous US trading-day close snapshot. Seven-day movement is shown separately. Every event should identify the affected symbol, direct or related status, factor or thesis pillar, signed impact, source, relevant event date, ingestion time, and freshness. For 13F evidence, the reporting period and filing date are distinct; the filing does not imply a trade occurred today.

Historical replay and model changes

A historical date restores the score, factors, active evidence, and explanation stored at that time. One immutable snapshot owns its base inputs, event evaluations, evidence cutoff, effective time, and source watermarks. No later event can be merged into an earlier snapshot. Material formula or threshold changes create a new model version rather than silently rewriting earlier results.

Portfolio methodology

Portfolio Confidence, Base Conviction, Event Pulse, and factor values are weighted aggregates of covered holdings. Explicit weights are used only when every holding supplies a valid weight; otherwise complete market values are used, then shares multiplied by the latest available price, and finally equal weights. Accepted weights are normalized to 100%, and the selected method is disclosed.

Concentration, sector concentration, beta, correlation, and largest risk contributors are construction diagnostics shown separately. Construction quality is never blended into conviction. Missing or stale holding data lowers coverage and can make the portfolio output incomplete.

Model governance

Pop Terminal tests exact reconciliation, coverage behavior, caps, decay, deduplication, direct-issuer scope, absorption, conflicting evidence, provider failures, and Free/Pro score equality before activating a model version. Source or model corrections can change future outputs; they do not create a guarantee of completeness or performance.

See Data Limitations and the Financial Disclaimer for the boundaries of these outputs.