Temporal Graph RAG must distinguish when a fact was true in the modeled world from when the system recorded or believed it. These are valid time and transaction time. Keeping both can answer different historical questions; freshness ranking is separate from whether a fact applies at the time being asked about.
Valid time and transaction time answer different questions
Valid time describes when a fact applied in the modeled reality. For a graph relationship, it can mark the period during which that relationship actually held.
Transaction time describes when the database recorded or treated a fact as current. It records the system’s own history, which may lag behind events in the world or change when an assertion is corrected.
A system that tracks both is called bitemporal. The distinction matters because “What was true at that time?” and “What did we believe on March 1?” are not necessarily the same question. The first asks about valid time; the second asks about transaction time. One timestamp cannot always answer both. The distinction is described in the temporal graph RAG paper by Zhang and in work on temporal data models (Zhang, 2026; Rost et al., VLDB Journal).
#1 Best Overall
How the two timelines handle change, late facts, and corrections
Consider a graph that records which company employs a person. Its history can differ depending on whether the question is about the world or the database’s knowledge.
Ordinary change
If the person worked for Company A until a known date and then moved to Company B, the valid-time interval for the first relationship ends when that relationship stops being true. A current-state graph may show only Company B; a temporal graph can retain the earlier relationship for historical questions.
Late-arriving information
Suppose the database learns today that the person joined Company A last month. The relationship’s valid time begins last month, while its transaction time begins today, when the system records it. A query asking what was true last month may include the relationship; a query asking what the database knew last month should not treat it as already recorded.
A correction to an earlier assertion
If the system later discovers that an earlier assertion was wrong, the correction does not necessarily mean the real-world relationship changed at that point. A bitemporal history can preserve what the database previously believed while recording the corrected assertion and the period to which it applies. The exact way a database represents corrections depends on its temporal model.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →What temporal graph data represents
In the temporal property graph model discussed by Rost and colleagues, vertices and edges carry time intervals. The paper defines a temporal property graph as “a property graph with additional time information on its vertices and edges, which primarily describes the historical development of the structure and attributes of the graph, i.e., when a graph element was available and when it was superseded.” This is the authors’ definition, not a universal standard.
That model uses closed-open intervals: the start is included and the end is excluded. For example, an interval from 1 March to 1 April includes 1 March but not 1 April. Adjacent intervals can meet at 1 April without both claiming that instant. Other implementations may represent intervals differently, so check the database’s boundary and open-ended interval conventions before relying on them.
Rank #3
How temporal history changes Graph RAG retrieval
A static or latest-state graph can answer from the edges it currently contains, but it may not preserve enough information to answer what was true earlier or what the system believed before a correction. A temporal graph can make those histories queryable; it does not make the RAG application historically accurate by itself.
For a time-specific answer, retrieval needs to select evidence for the requested time. The answer should retain enough provenance to show which assertion and interval support it. Otherwise, a system can retrieve a plausible current edge and present it as historical evidence.
Zhang’s July 11, 2026 arXiv preprint presents one research design using typed temporal operators and trace-grounded answer verification. It is an example of an approach, not evidence that every Graph RAG system requires that exact design or that a graph database automatically verifies historical answers (paper and benchmark description).
What the reported benchmark does—and does not—show
On its development benchmark, the paper reports an exact-match score of 0.409 for TGMS with a 14B open-source model, compared with 0.045–0.182 for the Vector-RAG, static-graph RAG, and text-to-Cypher baselines in the same reported setup. On correction probes, it reports 0.67 exact match for TGMS and zero for the three 14B baselines. The paper also says its verifier detected all 500 injected count and entity errors and reported no false positives on its clean answers.
These are results reported by that paper on its benchmark, not an industry-wide comparison or independently replicated performance guarantee. They do not establish how another application will perform on its own update, correction, and historical-query patterns.
Freshness is not the same as validity
Validity filtering asks whether a fact applies at the time in the user’s question. If a relationship expired before that time, it should not be returned as true for that period.
Best Value
Freshness or recency ranking helps order otherwise relevant evidence, such as choosing among document versions. A newer document is not automatically valid for an earlier date, and an older record may still be the right evidence for a historical question.
A temporal RAG README illustrates one project’s approach of classifying validity separately from document kind and applying expiry and time-decay handling (project README). That is an implementation example, not a standard or a general ranking formula.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How to assess temporal support in a graph RAG stack
Feature labels such as “temporal” or “bitemporal” are not enough to establish that a system can answer the questions an application needs. Check the data model, query language, and retrieval path separately.
- Time dimensions: Does it support valid time, transaction time, or both?
- What receives timestamps: Do intervals apply to vertices, edges, properties, documents, or only selected records?
- Interval semantics: Are starts and ends inclusive or exclusive? How are open-ended intervals represented?
- History and corrections: Can the system preserve late-arriving facts and earlier assertions for the audit or replay questions the application needs?
- Query expressiveness: Can queries ask both “valid at time T” and “known as of transaction time T”?
- Retrieval integration: Do temporal filters apply consistently through vector retrieval, graph traversal, ranking, and evidence provenance?
- Evaluation fit: Do benchmark tests reflect the application’s own update timing, corrections, and historical questions?
Temporal graph models vary in which time dimensions and graph changes they support, and in whether history is represented through snapshots or time properties. A model’s ability to store time and its query language’s ability to use that time are distinct questions.
Free tools Windows power users keep installed
One-click scans. No signup required.
Check the product documentation for the exact version
Database behavior is implementation-specific and can change between versions. For example, XTDB version 1 documentation says that if a write has no explicit valid-time value, valid time and transaction time take the same value. It also notes a limitation on using valid time in Datalog queries unless a temporal component is present in the documents. Those are version 1 details, not a statement about current XTDB behavior; consult the documentation for the version being deployed (XTDB 1.24.3 Datalog query documentation).
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
