Relationships change over time

You cannot step into the same network twice.

Heraclitus said you cannot step into the same river twice, because the water has moved on and you are no longer the same person standing on the bank. Twenty-five centuries later, the same principle holds for every relationship map you have ever drawn. The org chart from January does not describe your organization in July. The stakeholder map you created at project kickoff is partially fictional by the time you reach the midpoint. The professional network you mapped last year contains connections that have quietly dissolved and is missing connections that have since formed.

This is not a poetic observation. It is a structural fact about how relationships work, and ignoring it is one of the most common sources of silent failure in any system that depends on connections between things.

In the previous lesson (L-0254), you learned that relationship mapping reveals system structure — that when you draw all the connections between elements, the architecture of the system becomes visible. That lesson taught you to see the skeleton. This lesson teaches you something harder: the skeleton is moving. The connections you mapped are not permanent infrastructure. They are dynamic, time-bound, and constantly shifting in strength, direction, and existence. If you treat them as fixed, you will build plans on a foundation that is already changing underneath you.

Every relationship has a lifecycle

Social psychologist Mark Knapp formalized what most people know intuitively: relationships move through stages. His relational development model describes ten stages organized into two sequences — five stages of coming together (initiation, experimentation, intensifying, integration, bonding) and five stages of coming apart (differentiating, circumscribing, stagnation, avoidance, terminating). The model is most commonly applied to interpersonal relationships, but the structural pattern applies universally to any type of connection between entities.

Consider a dependency between two software services. It begins with experimentation — one team discovers the other's API might solve a problem. It intensifies as the integration deepens. It reaches bonding when the dependency becomes load-bearing. And it can come apart through the same stages: differentiating (the teams' needs diverge), circumscribing (interactions narrow to only the minimum required), stagnation (neither team invests in the integration), avoidance (workarounds replace direct engagement), and termination (the dependency is removed or the service is replaced).

The same lifecycle applies to knowledge relationships. A concept that felt tightly connected to your core framework six months ago might have drifted toward the periphery as your understanding evolved. An author whose work felt foundational may now feel tangential. A mental model that linked two domains in your thinking may have been replaced by a more precise connection.

Steven Duck, another social psychologist, added an important nuance with his phase model of relationship dissolution. Duck argued that relationship breakdown is not a single event but a process that moves through distinct phases: an intrapsychic phase (where one party privately reconsiders the relationship), a dyadic phase (where the dissatisfaction surfaces between the parties), a social phase (where the broader network becomes involved), and a grave-dressing phase (where both parties construct narratives about what happened). The critical insight for relationship mapping is this: by the time a relationship visibly dissolves, it has been functionally deteriorating for a long time. Your map showed it as solid while the connection was already hollowing out.

This is why static relationship maps lie. They show you the state of a connection at the moment you recorded it, and they silently assert that this state persists. They don't — they can't — show you the trajectory.

The science of temporal networks

Network science has formalized the study of relationships that change over time under the term temporal networks (also called time-varying networks, dynamic graphs, or evolving graphs). A temporal network is a graph whose topology — which nodes exist, which edges connect them, and what properties those edges carry — changes as a function of time.

The distinction between a static network and a temporal network is not merely academic. Research in temporal network analysis has demonstrated that many properties of networks — their ability to transmit information, their resilience to disruption, their community structure — change fundamentally when the time dimension is incorporated. A path that exists in a static aggregation of a network (where you collapse all connections across time into a single snapshot) may never have actually been traversable, because the edges that compose it were never active simultaneously.

Consider a simple example. In your professional network, you know Alice, and Alice knows Bob, so in a static network there is a path from you to Bob through Alice. But if your relationship with Alice was active in 2022 and Alice's relationship with Bob didn't form until 2024, there was no point in time where that path was actually usable for information transfer or introduction. The static map shows a connection. The temporal reality shows a gap.

Researchers distinguish two primary representations for temporal networks. Discrete-time dynamic graphs capture the network as a sequence of snapshots at fixed intervals — like photographing your org chart once per quarter. Continuous-time dynamic graphs record each change as a timestamped event — edge created, edge deleted, edge weight changed. The continuous representation is richer but harder to manage. The discrete representation is simpler but can miss changes that occur between snapshots.

Both representations converge on the same principle: a relationship map without a time dimension is a lossy compression of reality. It discards exactly the information you need to understand whether the connections you're relying on still exist, still function, and still carry the weight you're putting on them.

Relationship decay is the default

Perhaps the most important insight from the research on temporal relationships is that decay is the default state. Relationships do not maintain themselves. They require active investment, and in the absence of that investment, they weaken.

Robin Dunbar's work on social network size provides the quantitative framework. Dunbar proposed — and has defended for over thirty years — that humans can maintain approximately 150 stable social relationships at any given time. But the number 150 is less important than the structure underneath it. Dunbar's research reveals that social networks are organized in concentric layers: an innermost circle of about 5 people (to whom you devote roughly 40 percent of your available social time), a sympathy group of about 15, an affinity group of about 50, and the full active network of about 150. Beyond 150, relationships exist in name only — you may recognize someone, but the connection carries no functional weight.

The layered structure exists because time is an inelastic resource. You have a fixed number of hours available for relationship maintenance, and maintaining a relationship requires recurring investment. Dunbar's research shows that the primary maintenance mechanism is communication — direct, personal interaction. Without it, relationships decay. They don't decay instantly; they drift through the layers, moving from the inner circle outward until they fall off the edge of your active network entirely.

A 2015 study on managing relationship decay found that the decline in friendship quality was mitigated by increased effort invested in the relationship, but the key finding was that effort had to be ongoing. A single burst of reconnection might temporarily restore a relationship's position in your network, but without sustained follow-through, the decay resumed. The researchers also found striking contextual effects: relationships that depended on a shared context (same workplace, same neighborhood, same project) deteriorated faster when that context was removed.

This has direct implications for any relationship map you maintain. Every connection on your map is subject to decay unless actively maintained. And the connections most at risk are the ones that depend on a shared context that may itself be changing — the colleague on the project that's winding down, the collaborator at the company you're about to leave, the mentor whose field is diverging from yours.

Bitemporal truth: what was real and when you knew it

Database engineers have been thinking about temporal relationships longer than most disciplines, and they've developed a concept that is directly applicable to personal knowledge infrastructure: bitemporal modeling.

A bitemporal model tracks two distinct timelines for every relationship. Valid time is when the relationship was actually true in reality. Transaction time is when the relationship was recorded in your system. These two timelines can diverge significantly.

Consider a practical example. You discover in March that two of your team members have been collaborating on a side project since January. The valid time of that relationship is January. The transaction time — when it entered your map — is March. For two months, your relationship map was wrong. It showed no connection where one existed. Any decisions you made during that period based on your understanding of team dynamics were based on an incomplete picture.

The reverse happens too. Your map still shows a partnership between two departments, but the partnership effectively ended last quarter when the executive sponsor left. The valid time of that relationship's termination was months ago. The transaction time — when you update your map — hasn't happened yet. You're operating on stale data.

Martin Fowler, writing about bitemporal modeling in software systems, emphasizes that the distinction matters precisely because it allows you to answer two different questions: "What was actually true at time X?" and "What did we believe was true at time X?" For financial reporting, regulatory compliance, and audit trails, these are critically different questions. For your personal knowledge infrastructure, they are too. When a plan goes wrong because a relationship you depended on had already dissolved, the postmortem requires distinguishing between what was real and what your map said.

You don't need to implement a formal bitemporal database for your personal relationship maps. But you do need the mental model. Every connection on your map has two timestamps: when it actually formed or changed (which you may not know precisely) and when you became aware of it (which you can track). The gap between those two timestamps is the latency in your situational awareness. Reducing that gap — through more frequent verification, through closer attention to weak signals of change — is one of the highest-leverage improvements you can make to any relationship mapping practice.

Types of relationship change

Not all relationship changes are the same. Understanding the taxonomy of change helps you watch for the right signals and update your maps with precision rather than panic.

Formation is the creation of a new edge. Two entities that had no connection now have one. This is the most visible type of change — new connections tend to announce themselves. A new hire joins the team. A new vendor is contracted. A new concept enters your knowledge graph. Formation events are usually easy to detect but often slow to be recorded. The relationship exists before you put it on the map.

Dissolution is the removal of an edge. A connection that existed ceases to exist. Dissolution is harder to detect than formation because absent things generate no signals. You notice a new colleague immediately. You may not notice for months that a former collaborator has stopped responding to emails. In temporal network research, this asymmetry — formation is visible, dissolution is silent — is a well-documented source of map staleness.

Strengthening is an increase in edge weight. The connection becomes more important, more active, or more load-bearing. A casual acquaintance becomes a close collaborator. A loose dependency becomes a critical one. Strengthening is often gradual and may not cross any threshold that triggers conscious recognition. By the time you notice the relationship is strong, it has been strengthening for a while.

Weakening is the inverse — a decrease in edge weight. The connection becomes less active, less important, less functional. Granovetter's definition of tie strength — time invested, emotional intensity, intimacy, and reciprocal services — gives you four dimensions along which weakening can occur. A relationship can weaken along one dimension while remaining strong along another, which makes it harder to assess the overall trajectory.

Transformation is a change in the type or nature of the connection. A mentorship becomes a peer relationship. A collaboration becomes a competition. A dependency becomes a provider relationship. Transformation doesn't add or remove an edge — it changes what the edge means. This is perhaps the most dangerous type of change to miss, because your map still shows a connection, and the connection still exists, but its function has fundamentally shifted. You're routing trust through a channel that now carries rivalry.

Reversal is a change in the direction of a directed edge. Information that used to flow from A to B now flows from B to A. Authority that used to run downstream now runs upstream. Reversals are common in mentoring relationships (the student eventually teaches the teacher), in organizational hierarchies (reporting lines change), and in knowledge domains (the field you used to draw from now draws from your work).

Granovetter's bridge and the temporal paradox

Mark Granovetter's 1973 paper "The Strength of Weak Ties" remains one of the most cited works in social science, and its central insight is deeply temporal even though it is rarely discussed that way.

Granovetter argued that weak ties — relationships with low time investment, low emotional intensity, and low intimacy — are paradoxically more valuable for certain functions than strong ties. Specifically, weak ties act as bridges between densely connected clusters. Your close friends all know each other and share the same information. Your acquaintances connect you to entirely different clusters, giving you access to novel information, opportunities, and perspectives.

But here is the temporal paradox: weak ties are also the most vulnerable to decay. They require less maintenance to exist, but they also provide less intrinsic motivation for maintenance. You will always find time to talk to your closest friends. You will not always find time to email that person you met at a conference two years ago. The connections that provide the most structural value to your network are the ones most likely to dissolve through simple neglect.

A 2022 study published in Science provided causal experimental evidence for the strength of weak ties using data from LinkedIn, but it also revealed an important nonlinearity. The relationship between tie weakness and job transmission followed an inverted U-shape: weaker ties increased job transmission, but only to a point, after which there were diminishing returns to weakness. Ties that were too weak — essentially dormant — provided no benefit. The sweet spot was what the researchers called "moderately weak ties."

This has a direct implication for temporal relationship mapping: you need to track not just which weak ties exist, but which ones are at risk of falling below the threshold of usefulness. A weak tie that you engage with once a quarter may be your most structurally valuable connection. A weak tie that you haven't engaged with in two years is functionally dissolved, even if both parties would still recognize each other.

Your Third Brain: temporal knowledge graphs and versioned memory

The challenge of relationships changing over time is not unique to human cognition. It is one of the central problems in artificial intelligence and knowledge engineering, and the solutions being developed there illuminate what's possible for your personal epistemic infrastructure.

Temporal knowledge graphs extend traditional knowledge graphs by adding a time dimension to every relationship. In a standard knowledge graph, a triple like (Jay, works_at, Cosm) is stored as a timeless fact. In a temporal knowledge graph, it becomes (Jay, works_at, Cosm, [2024-01-15, present]) — the relationship carries its valid time interval. This means the graph can answer questions that static graphs cannot: "Who did Jay work for in 2022?" "When did the dependency between Service A and Service B form?" "Which relationships existed at the time of the system failure?"

A 2024 survey on temporal knowledge graph representation learning identified this as an active frontier in AI research. The challenge is not just storing temporal information but learning from it — recognizing patterns in how relationships evolve, predicting which relationships are likely to form or dissolve, and reasoning about the state of the network at arbitrary points in time.

For your personal knowledge infrastructure, the principle is more important than the implementation. Every relationship map you maintain is implicitly temporal. The question is whether you make that temporality explicit or leave it hidden. Adding timestamps to your connections — even informally, even approximately — transforms your map from a snapshot into a history. And a history is dramatically more useful than a snapshot, because it lets you see trajectories, not just positions.

Knowledge graph versioning systems provide another model. Rather than timestamping individual edges, you version the entire graph — saving snapshots at regular intervals so you can compare the network at time A with the network at time B. This approach trades precision for simplicity. You lose the ability to see exactly when each edge changed, but you gain the ability to see the overall structural evolution of your network. For most personal knowledge management purposes, periodic snapshots (monthly, quarterly) provide sufficient temporal resolution to catch the changes that matter.

The emerging tool Graphiti, designed for building real-time knowledge graphs for AI agents, takes yet another approach: it treats the knowledge graph as a continuously evolving structure where every change is logged as an event. New edges, deleted edges, and modified edges are all recorded with timestamps, creating a full audit trail of the graph's evolution. This is essentially bitemporal modeling applied to knowledge infrastructure — and it represents the direction that personal knowledge tools are likely to move as AI-augmented thinking systems mature.

Protocol: The monthly relationship audit

Here is the operational protocol for keeping your relationship maps honest about time. This takes fifteen to twenty minutes once per month, applied to whichever relationship map is most critical to your current work.

1. Select one map. Your stakeholder map, your professional network, your project dependency diagram, your knowledge graph, your team collaboration map. Don't try to audit all of them at once. Rotate monthly.

2. Scan for ghost edges. Go through each relationship on the map and ask: have I directly verified this connection in the last 60 days? Not assumed it still exists — actually interacted with it, observed it, or received evidence of it. Mark any connection you have not verified as "unconfirmed."

3. Check for missing edges. Ask: are there connections that exist right now that are not on this map? New collaborations, new dependencies, new information flows, new conflicts. Add them, with today's date as the formation date.

4. Assess trajectory. For your five most important connections, assign a direction: strengthening, stable, or weakening. Base this not on how you feel about the relationship, but on observable indicators — frequency of interaction, depth of engagement, reciprocity of investment. If a critical connection is weakening, decide now whether to reinvest or accept the decline.

5. Retire dissolved connections. If a connection has been "unconfirmed" for three consecutive audits (roughly 90 days), move it to a historical layer of your map. Don't delete it — you may want to understand your network's history later. But remove it from the active view so it stops influencing your current decisions.

6. Record one transformation. Identify at least one relationship whose nature has changed — not formed or dissolved, but transformed. A mentorship that became a partnership. A collaboration that became a dependency. A peer relationship where the power dynamic shifted. Update the edge label to reflect its current function.

7. Date the map. Write today's date on the map itself. This sounds trivial, but it is the single most important habit for temporal awareness. A map without a date implicitly claims to be timeless. A map with a date honestly declares: this is what I believed was true on this day.

The goal is not to maintain a perfect, real-time model of every relationship in your life. That is neither possible nor desirable. The goal is to make the temporal nature of your relationships visible to yourself — to prevent the specific failure mode where you build plans, route information, and make decisions based on a network structure that has already changed.

---

In the next lesson (L-0256), you will encounter a phenomenon that makes temporal awareness even more critical: transitive relationships propagate effects. If A relates to B and B relates to C, there may be an implied relationship between A and C. But transitivity only works if both edges are active at the same time. A change in one relationship doesn't just affect that connection — it can ripple through chains of dependency, amplifying or severing pathways you didn't even know existed. When relationships change over time, the transitive effects change too — and tracking those cascades is where relationship mapping becomes genuinely powerful.

---

Sources

• Holme, P. & Saramaki, J. (2012). "Temporal networks." Physics Reports, 519(3), 97-125. Temporal network — Wikipedia
• Granovetter, M. (1973). "The Strength of Weak Ties." American Journal of Sociology, 78(6), 1360-1380. Original paper (Stanford)
• Rajkumar, K. et al. (2022). "A causal test of the strength of weak ties." Science, 377(6612), 1304-1310. Science
• Dunbar, R. I. M. (2021). "Dunbar's number: why my theory that humans can only maintain 150 friendships has withstood 30 years of scrutiny." The Conversation. The Conversation
• Roberts, S. & Bhatt, S. (2015). "Managing Relationship Decay: Network, Gender, and Contextual Effects." Human Communication Research, 42(2), 265-287. PubMed
• Knapp, M. L. & Vangelisti, A. L. (2008). Interpersonal Communication and Human Relationships. Knapp's relational development model — Wikipedia
• Duck, S. (1982). "A topography of relationship disengagement and dissolution." In S. Duck (Ed.), Personal Relationships 4: Dissolving Personal Relationships. Duck's Phase Model — tutor2u
• Fowler, M. (2005). "Bitemporal History." martinfowler.com. Martin Fowler
• Cai, L. et al. (2024). "A Survey on Temporal Knowledge Graph: Representation Learning and Applications." arXiv:2403.04782
• Senzing (2024). "Temporal Knowledge Graphs: Uncovering Hidden Patterns." Senzing GPH