What a Stock Score Means: An Audit of 6,213 Records

A stock score summarizes a particular model, dataset and point in time. It does not, by itself, establish an investment's probability of profit or loss. Our audit of an archived Stock Expert AI dataset shows why checking the model version, the captured universe and the date is essential before interpreting a number.

Snapshot: August 29, 2026, 20:20 UTC. Analysis published: September 12, 2026. This is a historical data-quality case study, not a current stock screen.

What we actually measured

The stored snapshot contains 6,213 distinct instrument records. We selected only finite, positive numeric values from the field named moonshotScore. That leaves 4,675 records, with a median of 47 and a mean of 47.96. The observed range is 34 to 74.

Measure	Result
Distinct instrument records captured	6,213
Positive historical moonshotScore values	4,675
Records outside that filter	1,538
Median / mean	47 / 47.96
Minimum / maximum	34 / 74
Records with a numeric moonshotScoreV2 field	786

The separate V2 field is not mixed into the statistics above. Numeric presence is not evidence that a score is current or that the instrument is eligible today. The 1,538 records outside the positive-score filter are not automatically poor investments; missing values, zero values and instrument eligibility are different questions.

A closer look at the distribution

The grouped frequencies below describe the 4,675 selected historical values. These are descriptive numeric intervals, not the current V2 model's score bands.

Historical score interval	Records	Share of selected records
30 to below 40	164	3.51%
40 to below 45	1,408	30.12%
45 to below 50	1,643	35.14%
50 to below 55	858	18.35%
55 to below 60	345	7.38%
60 to below 65	203	4.34%
65 to below 70	50	1.07%
70 to below 80	4	0.09%

One score value, 44, appears 939 times: 20.09% of the selected records. This concentration is a feature of the captured values that deserves investigation. It does not demonstrate that a fifth of companies have identical underlying businesses, identical risks or identical future returns. A final composite can conceal different inputs, missing-data rules and rounding effects.

Three boundaries that change the meaning of a score

1. The captured dataset is not the whole market

The collector recorded 66 page requests and deduplicated records by ticker. A completed collection does not prove complete market coverage: the endpoint's ordering may change while pages are being retrieved, and removing duplicates cannot recover a record skipped between requests. We describe this as 6,213 captured records, not a census of US equities.

2. Historical and current model versions are different

Today's MoonshotScore methodology describes a five-pillar V2 model for eligible common stocks and ADRs. The older stored field used for this case study must not be presented as today's V2 score. Nor should an old value fill a gap when a current eligible score is unavailable. Instrument coverage, score availability and page availability are separate concepts.

3. A fresh price does not make an analysis fresh

A quote, a score calculation and an AI-written company dossier can carry different dates. A page can receive a new price while its explanatory text remains dated. Compare those timestamps before reading a sentence as a statement about the present. The site's methodology explains the current update policies and the distinction between quote snapshots and dated analysis.

Reproduce the summary from the aggregate data

Each pair below is [historical score, record count]. The complete frequency list is included so that readers can reconstruct the count, mean and median. This aggregate does not reproduce instrument-level selection or verify the original provider data; those are separate checks. No ticker-level prices or company financial statements are redistributed here.

const frequencies = [[34,4],[35,5],[36,21],[36.5,1],[37,13],[37.5,5],[38,55],[38.5,5],[39,52],[39.5,3],[40,37],[40.5,5],[41,68],[41.5,2],[42,191],[42.5,6],[43,134],[43.5,19],[44,939],[44.5,7],[45,201],[45.5,12],[46,445],[46.5,6],[47,472],[47.5,9],[48,238],[48.5,8],[49,242],[49.5,10],[50,279],[50.5,6],[51,177],[51.5,7],[52,174],[52.5,5],[53,93],[53.5,2],[54,112],[54.5,3],[55,92],[55.5,4],[56,81],[56.5,2],[57,65],[57.5,4],[58,61],[58.5,1],[59,32],[59.5,3],[60,54],[60.5,4],[61,47],[62,39],[63,32],[63.5,1],[64,25],[64.5,1],[65,12],[65.5,2],[66,12],[66.5,1],[67,12],[67.5,1],[68,4],[68.5,2],[69,4],[71,1],[72,2],[74,1]];
const count = frequencies.reduce((n, [, frequency]) => n + frequency, 0);
const sum = frequencies.reduce((n, [score, frequency]) => n + score * frequency, 0);
let cumulative = 0;
const middleRank = (count + 1) / 2;
const median = frequencies.find(([, frequency]) => {
cumulative += frequency;
return cumulative >= middleRank;
})[0];
console.log({ count, mean: sum / count, median });
// count: 4675; mean: 47.96096256684492; median: 47

The frequency total was checked against the selected-record count. Its weighted mean and weighted median were also checked against direct calculations on the captured records. These checks validate the arithmetic of this summary, not the predictive quality of a stock-scoring model.

A practical reading checklist

• Version: Which model and field produced this value?
• Date: When were the inputs, score and written explanation last updated?
• Coverage: Which instruments are eligible, and which captured records were excluded?
• Missing data: Is an unavailable value shown as unavailable, or silently replaced?
• Interpretation: What has the model actually been validated to measure?
• Reproduction: Can the published summary be recalculated from the disclosed data?

This checklist is for evaluating the information in a research tool. It is not a method for selecting a stock or predicting returns.

Provenance and correction record

Source: Stock Expert AI's archived moonshot-snapshot-2026-08-29, captured at 2026-08-29T20:20:02.852Z. The snapshot is retained internally; only the aggregate score frequencies are published here. The archive hash identifies the exact local source used, but is not independent verification of its contents.

SHA256: ca1fabb38a2fb1ce651a2c504baf599110edaf08869260f565e874cb4b434498

This case study extends our corrected August snapshot note. The earlier note's unsupported interpretation of scores as a safety/risk ranking and of an extreme raw percentage as a market event was withdrawn. This article adds the aggregate frequencies needed to reproduce the descriptive summary.

Disclosure: Produced by Stock Expert AI with AI assistance and deterministic calculations on its own archive. Editorial responsibility rests with Stock Expert AI. The statistics describe a historical dataset; they are not independent validation, investment advice or a forecast.