How we would judge a price

60.4% was the three-month failure rate for Wait if flexible: 29 of 48 testable signals were followed by a price increase. The pre-declared limit was 40%. The rule failed, so recommendations are disabled. The site describes only the price state already observed and gives no buy-or-wait advice.

This is the result from the thresholds specified before the test. We did not change the thresholds after seeing it.

Test data: 11 monthly files from the US Bureau of Labor Statistics Public Data API, retrieved 2026-07-25T16:11:53.656Z and tested again on 2026-07-26. No retailer, marketplace or Amazon-derived price history entered the test.

The backtest read each of the 11 BLS series as if every observed month were a new decision point. The files contain 1,252 observed series-months from January 2017 through June 2026. The candidate rule produced 51 Wait if flexible signals, 120 Buy if needed signals and 1,081 decisions with no recommendation.

For Wait if flexible, a contradiction means the observed index was higher at the named horizon. Waiting would have cost more according to that index.

For Buy if needed, a contradiction means the observed index was lower at the named horizon. Buying then would have missed a lower later index.

The sample sizes differ by horizon because a signal is scored only when the exact future month exists in the source file. Missing months are not interpolated, and observations beyond the end of a series are not invented.

We separately scored decision months from January 2021 through December 2023 for BLS series PCU334413334413, Semiconductor and related device manufacturing. There were 36 decision points. The rule produced five wait signals, two buy signals and 29 decisions with no recommendation. A late-2023 decision may use its exact horizon observation in 2024; the date restriction applies to decision points, not to their later outcome.

The buy sample is only two signals. It is reported in full, but it is too small to support a broad claim.

Price state is retrospective. It compares the current trustworthy offer with the entity's own observed history, using at most 365 days.

The site reports not enough data to judge when the window is shorter than 14 days, when there are fewer than 10 observations, or when coverage is below 60%. In that state it does not calculate or show a percentile and it cannot recommend.

With sufficient data, the position is the percentile of the current offer within the observed window. Tied prices use their midpoint rank. This keeps a flat series at the 50th percentile instead of falsely calling it cheap. A position at or below 25 is labelled cheap compared with its own recorded history. A position at or above 75 is labelled expensive compared with its own recorded history.

Trends use an ordinary least squares slope over the observed 7, 30 and 90 day windows. The slope is divided by the mean price in that window and expressed as percent per 30 days. Missing dates remain missing. A trend window with fewer than two dated observations has no trend value and cannot produce a recommendation. The 30-day trend is rising above 3%, falling below negative 3%, and stable otherwise.

BLS data is monthly. For this backtest only, one real calendar month is represented as one 30-day trend step. A missing calendar month remains a 60-day or longer gap and lowers coverage. No price value is filled in.

The candidate rule says Buy if needed only when position is at or below 25, the 30-day trend is zero or positive, and the substitute check confirms that no better substitute makes the low price an obsolescence signal.

It says Wait if flexible only when position is at or above 75 and the 30-day trend is negative. Every other case gives no recommendation.

The rule correctly refuses to tell a reader to wait merely because a price is high while still rising. The historical test shows that this protection is not enough. The candidate was still contradicted too often, so the published site gives no recommendation. A replacement would need its own recorded rule and a new pre-declared test.

Market outlook is a separate forecast and defaults to abstention. It can ship only with all six conditions: a named horizon, at least two named and dated causal sources less than 30 days old, a stated confidence, a recorded forecast version, a future date scheduled for outcome scoring, and an explicit abstention state.

If any condition is missing, the answer is:

> We do not forecast prices. Here is the current retail state, and the condition that would change the recommendation.

This method cannot predict the next price. A percentile is not evidence of mean reversion, and a falling 30-day trend is not proof that the fall will continue.

The BLS series are monthly category and manufacturing indexes, not current retail offers for an individual RAM kit, CPU, GPU or other part. The supplied history therefore does not reproduce the specific retail RAM increase from October 2025 through June 2026. That episode is protected by the rising-trend property test, but it cannot be claimed as a successful real-data backtest from these files.

The files currently cover January 2017 through June 2026, not decades. Two consumer CPI series have previously produced identical values, and several indexes are broad or weak category proxies. Treating every series-month equally does not make the 11 series independent evidence.

The backtest does not validate substitute quality, causal market explanations, daily retail behavior or forecast accuracy. Those need first-party retail observations, defined substitute sets and separately recorded forecast outcome scores. Until that evidence exists and a pre-declared recommendation test is cleared, the honest answer is price state with no recommendation.