Product guide · historical evidence
Use price history to understand what SaaS Sentinel actually observed over time without turning a short retained record into a forecast. Read snapshot ranges, median movement, price dispersion, distinct rate changes, and evidence-only refreshes as separate signals, then return to current pricing before making a purchase decision.
Reading order
Start with evidence depth and scope. Only then interpret movement. This prevents a visually obvious change on a chart from being treated as a market conclusion when the underlying provider set is sparse or has changed between dates.
1 · Depth
The history hub shows how many timestamped observations and distinct retained days exist for each activated GPU. More retained depth gives you more evidence to inspect; it does not by itself make the series representative of the whole market.
2 · Latest snapshot
The model page summarizes the observed low-to-high range and median for each retained date. The low is not automatically the deployable market price, the high is not a ceiling, and the median is descriptive of the rows retained for that date.
3 · Median movement
The chart reports the percentage movement between retained endpoint medians and may translate that rate change into 730 GPU-hours. Treat it as a snapshot comparison, not a matched-cohort index or a prediction of the next month.
4 · Dispersion
Absolute range width shows the dollar gap between low and high. Relative range width expresses that gap as a percentage of the snapshot median, helping separate a wider dollar range caused by a higher overall price level from wider relative dispersion.
5 · Changes
A new source check, timestamp, or evidence hash is not automatically a new price. The change log labels a transition as a rate move only when the normalized USD/GPU-hour value actually changed for the tracked offer identity.
6 · Current decision
History explains the retained evidence trail. The current GPU page and comparator answer what the latest qualified catalog contains now. Re-check official sources and availability evidence at decision time.
Metric reference
The historical surface intentionally exposes several different measurements instead of compressing them into one trend score. Keep their meanings separate when citing the page in a FinOps review.
| Signal | What it describes | Do not infer |
|---|---|---|
| Low | The lowest normalized rate retained in that model/date snapshot. | Guaranteed capacity, universal market floor, or best workload fit. |
| Median | The middle observed rate across the retained snapshot rows. | A transaction-weighted or provider-size-weighted market price. |
| High | The highest normalized rate retained in that model/date snapshot. | A provider ceiling or the maximum price available anywhere. |
| Median movement | The change between retained endpoint medians. | A matched-cohort market index or future direction. |
| Range width | The observed high minus low within a snapshot. | Volatility, because provider mix and configuration mix can also change. |
| Relative spread | Range width divided by the same snapshot median. | Probability, risk, or a forecast confidence interval. |
| Rate move | A normalized price change for a tracked offer identity. | That every provider or configuration for the GPU moved with it. |
| Evidence refresh | A changed evidence record with the same normalized rate. | A price change merely because the observation timestamp is newer. |
Provider-mix boundary
The model history page summarizes whichever qualifying observations are retained for that model on each date. That is useful for an audit trail, but the set of providers or offers can differ between dates. The GPU Cloud Price Index uses a different fixed-cohort design when the question is broad market movement.
Model history
Inspect observed ranges, tracked identities, individual rate transitions, source evidence, and how the retained model slice changed between dates.
Price Index
The index separates current provider-balanced model summaries from a fixed matched-cohort historical series so catalog growth cannot masquerade as a historical price move.
Movement versus freshness
SaaS Sentinel retains evidence so that a provider source can be rechecked without manufacturing a new rate event. This keeps the audit trail honest: a newer check is evidence that the source was observed again, not evidence that the price moved.
Actual rate move
The history compares a tracked offer identity with its previous retained observation. A higher or lower normalized rate becomes a distinct rate transition and can feed the price-change alert product.
Evidence-only refresh
A new timestamp or source hash with the same normalized rate remains visible in the audit trail but is suppressed from the price-alert stream so monitoring does not manufacture activity.
Decision workflow
Historical evidence is most useful when it changes what you verify next. Use it to identify whether a current quote sits near a previously observed part of the range, whether a tracked identity recently moved, and whether the history is deep enough to deserve weight in the decision.
Current price
Use current semantic GPU pages or the global comparator for the latest qualified price, provider, purchase-model, region and availability evidence.
Future monitoring
Use the all-GPU or model-specific Atom feed when the decision can wait and you want a signal only when a normalized current-identity rate actually changes.
Workload economics
A lower historical GPU-hour price does not prove a lower completed-workload cost. Use benchmark-backed workload economics or calculators when throughput, runtime, scaling or utilization matters.
Need the raw audit trail?
Use the public history JSONL for reproducible downstream analysis, and preserve the observation dates and evidence boundaries when deriving your own metrics.