At noon on May 12, 2026, IONQ had fallen, bounced, and rolled over again. We asked Chart Library for 50 historical moments resembling its last ten bars, then opened the preceding hour for every match.
The first finding: a close match still needs inspection. Nine of the 50 matches used ten stored bars spanning more than ten actual minutes. One of the closest complete examples had been rising over the preceding hour while IONQ had been falling. These are differences we can identify before asking what happened next.
This is the first entry in the Chart Casebook: start with a real chart, retrieve examples, inspect the differences, and decide whether there is a coherent idea worth testing. This entry examines the comparisons. It does not establish a trading edge.
The chart we started with
The clock is 12:00 p.m. Eastern, taken from a saved approximate midday marker. The final bar begins at 11:59 and completes at noon. In the stored data, the move from the 11:00 open to that final close is -4.57%; the corresponding move over the final ten minutes is -1.41%. Both numbers describe information before the cutoff.
The question is whether the preceding decline and attempted recovery recur in the matches. A red candle at the end of a chart is only one part of that sequence.
Three comparisons that change the reading
| Example | What matches or differs before the cutoff | What we can conclude |
|---|---|---|
| #1 HYFM, March 25, 2021, 14:04 ET | The nearest returned example ends in a sharp decline. Its ten stored bars span 13 minutes, and only 34 of 60 expected minute rows are present in the displayed hour. | The time basis and data coverage differ. We cannot silently draw a continuous ten-minute path or label the gaps as halts. |
| #3 PENN, March 31, 2020, 11:04 ET | Its final ten-minute move is -3.32%, but its full hour is +3.47%. The hour begins with a strong advance, followed by a retreat. | A decline after an advance differs from IONQ's decline after an earlier selloff and bounce. The short window misses that distinction. |
| #28 VRT, June 10, 2024, 11:39 ET | Its hour is +1.16% and its final ten minutes are approximately flat (+0.02%). The final bar has a large upper wick. | A place in the nearest 50 is not a certificate that the complete setup matches. This is a useful counterexample to inspect. |
These examples were chosen by a disclosed rule: the first-ranked match, the first match with a complete rising hour, and the first anchor whose detector was not a volume surge. They were not chosen using later returns. Vertical scales differ between panels; the printed percentages preserve the difference in move size.
What was in the full set?
The search returned 50 moments across 49 symbols and 49 dates. Two members share March 8, 2022; 49 distinct dates is not proof of 49 independent market events.
- 41 of 50 have a complete ten-minute window. The other nine use ten observations spread over a longer interval; KOD's ten bars span 22 minutes.
- 18 of 50 have all 60 rows in the preceding hour. Nine occur less than an hour after the regular-session open. Separately, 25 have at least one missing regular-session minute in the displayed interval. The last two groups overlap, so they should not be added together.
- 48 of 50 end on the corpus's volume-surge detector, one on a high-break detector, and one on a large one-bar-move detector. Those labels describe how the library selected moments, not verified news catalysts.
- The stored input features for all 50 could be reconstructed from the captured bars within normal floating-point rounding. The timing differences come from how the corpus represents bars, rather than a failure to reproduce this retrieval.
The sample is heavily composed of abrupt declines at the end of the window. The earlier paths vary: a steady selloff, a failed recovery, a fall from an earlier rally, or an opening move with little preceding session history. We have recorded a short reading of every panel in the downloadable review notes.
What was actually known about IONQ?
IonQ released its first-quarter results on May 6, 2026, before this chart date. That establishes a recent public earnings event; it does not establish why a particular noon candle moved. IonQ's results announcement.
The company's May 12 Needham appearance was scheduled for 3:00 p.m. Eastern. Statements from that appearance cannot explain what traders already knew at noon. The fact that an event was scheduled and the later content of that event must be kept separate. We did not use the later transcript. IonQ's event record.
We have not established a shared catalyst, business situation, or participant flow across the 50 matches. A similar decline is insufficient to fill that gap.
The useful lesson
For this seed, the next comparison needs to retain three things: elapsed time, the preceding price sequence, and what information was available at that moment. The ten-bar ranking is a way to surface charts for inspection. We would not pool this set into a claimed win rate for the noon setup.
The hypothesis worth writing down is narrower: a renewed decline after a failed recovery may have different implications from a first decline after a strong advance. That remains a question. This inspection has not tested the difference, and it does not revive or confirm an earlier failed-push strategy.
Before an outcome study, we would need an explicit recurring process, a counterexample, consistent data coverage, and enough comparable cases under rules fixed in advance. A separate, previously unexamined sample would then be required for confirmation. If those conditions fail, the useful publication is the failure and what it rules out.
Inspect all 50 charts
The gallery preserves the original retrieval order. Every chart stops at its own cutoff. The shaded region is the ten stored bars used by the matcher; the dotted line marks ten actual minutes. Missing minute rows remain blank. An opening window shorter than an hour is labeled with only the regular-session rows that could exist.
The price charts are rebased to the first displayed open and use separate vertical scales. Ranking also incorporates volume and other input features; these panels show price. No post-cutoff prices or matched forward-return columns were queried for this case study.
Match 1 of 50: HYFM / 2021-03-25, 14:04 ET
Observed prices weaken late, but large gaps in the hour prevent a continuous-path reading; ten stored bars span 13 minutes.
Original retrieval order. Notes describe the price before each cutoff. They do not classify later winners or losers.
For an overview, open the printable sheets: matches 1-10, 11-20, 21-30, 31-40, and 41-50.
Method and limits
The retrieval protocol was committed as 720fbfd4 before retrieving the identities. We used the existing intraday corpus and its frozen 44-feature representation, requesting 50 approximate nearest neighbors strictly before the seed's New York session date and excluding IONQ. We did not refill or rerank the set after inspection. This is a study of that intraday substrate, not a test of the website's daily situation-first search.
Two limitations prevent interpreting it as a historical trading backtest. The corpus's construction uses full-session eligibility, including dollar volume and range, which would not all have been known at an intraday cutoff. Its normalization was fitted on July 22, 2026, after the seed date. The search is therefore a retrospective comparison, with future price columns deliberately excluded, rather than a simulation of what this exact model could have done live on May 12.
Historical issuer identity, corporate-action semantics, event context, execution costs, and causal explanations are not certified by the distance. Within-day percentage paths reproduce the stored features, but that does not establish complete coverage or instrument comparability. Approximate retrieval also does not guarantee that these are the globally nearest 50 windows.
This is a descriptive case note, not a registered outcome experiment, independent replication, or directional recommendation. No setup-intake admission or outcome test was conducted. The archived seed was already selected as interesting; it is not fresh confirmation evidence.
Evidence and downloads
- All 50 members and pre-cutoff measurements (CSV).
- A visual reading of every returned chart (CSV).
- Retrieval and review protocol (Markdown).
- Reconstruction checks, source hashes, and export receipt (JSON).
- Comparison figure for reuse (SVG).
The source snapshot contains 150 pre-cutoff IONQ bars and 6,890 pre-cutoff bars for the returned members. Raw vendor OHLCV is retained with the research record rather than redistributed here. Its SHA-256 is f6c1b05d94656906c23484c33be9d06e0093940756a8099fecebdd860b25178c. Published measurements and figures are derived from that fixed snapshot.
