The pragmatic conception is the one everyone reaches for, and you make a strong case for it. But the temporal conception beside it may cut deeper, because it can dissolve the category of replication failure rather than relocate it. It allows a third reading of a divergent result, beyond method error or context: the original was true, the replication is true, and the phenomenon moved in between. Much social behavior drifts on roughly the timescale of the research. The reflexive cases are sharper, where publishing the finding is part of what shifts the behavior it described. A priming effect fades as the trick becomes known; a norm changes once it is named. There the experiment is not measuring a stable target, it is nudging it. So 'does it replicate?' can hide a tense problem, treating a time-indexed truth as timeless. Part of the reproducibility crisis may be the discovery that some findings had expiration dates we never thought to print.
"The temporal conception involves the relative temporality of the experimental procedure and the phenomenon of interest; that is, the ratio between these two temporalities. More “contained” experiments can be implemented and evaluated more quickly than the substantive phenomenon is expected to change. More “uncontained” experiments are (relatively) temporally slow. One example involves research on phenomena that can be expected to change quickly: experimental audits of the recommendation algorithm on YouTube are temporally uncontained because YouTube could dramatically change the algorithm from one minute to the next. Another example involves experimental processes which take a long time to unfold: evaluations of the impact of kindergarten on lifetime earnings take decades to unfold, and are thus temporally uncontained even if the phenomenon of kindergarten itself is temporally stable."
That is the exact frame, thank you. Though notice both of Munger's examples treat the phenomenon as moving on its own clock, independent of the study: YouTube changes its algorithm whatever the auditor does, kindergarten effects unfold whether or not anyone measures them. The reflexive case is a different animal his ratio doesn't quite catch. There the phenomenon does not drift alongside the procedure; it drifts because of it. Publishing the priming effect is part of what retires it. At that point there are no longer two independent temporalities to take a ratio of, since the measurement has entered the causal history of the thing measured. So 'temporally uncontained' may hide two species: phenomena fast enough to outrun the study, and phenomena the study itself helps move. Only the first is really about speed. The second is about a loop, and a loop has no containment value at all, because each clean measurement spends a little of what it measured.
Maybe Gergen (1973) is a better fit to your point....
"If subjects possess preliminary knowledge as to theoretical premises, we can no longer adequately test our hypotheses. In the same way, if the society is psychologically informed, theories about which it is informed become difficult to test in an uncontaminated way."
"Knowledge about nonverbal signals of stress or relief (Eckman, 1965) enables us to avoid giving off these signals whenever it is useful to do so; knowing that persons in trouble are less likely to be helped when there are large numbers of bystanders (Latane & Barley, 1970) may increase one's desire to offer his services under such conditions; knowing that motivational arousal can influence one's interpretation of events (cf. Jones & Gerard, 1967) may engender caution when arousal is high. In each instance, knowledge increases alternatives to action, and previous patterns of behavior are modified or dissolved."
Gergen, K. J. (1973). Social psychology as history. Journal of Personality and Social Psychology, 26(2), 309–320. https://doi.org/10.1037/h0034436
Yes, Gergen is the better ancestor here, I should have reached for him. Though the gloom he drew from it feels optional to me. A finding that only holds while the subject can't see it isn't a failed law, it's a true report about a particular kind of pattern, the kind that lives on the subject's not-knowing. Those are real. They just have a different ontology than gravity. The disappointment comes from expecting them to behave like the stable kind. Gergen wanted timeless laws and found dated ones; maybe the dated ones were the actual object all along, and the timeless framing was the contamination. Which loops back to your containedness, since whether a phenomenon can survive being known is itself a containedness property, and not one of the three he lists.
The "contained vs. uncontained" distinction lands hard.
There's a parallel in Buddhist epistemology that might be worth a footnote. The pramana tradition (roughly: the study of valid cognition, developed by Dignaga and Dharmakirti in 5th-7th century India) was obsessed with exactly this question: what makes knowledge genuinely reliable vs. context-bound?
Their answer was unsettling. All knowledge involves universals that the mind constructs. And those constructions are always partly context-dependent. Which puts most social science in the "uncontained" camp by design, not by failure.
The "effect seeker vs. context monger" framing maps onto this well. Effect seekers want a universal that survives replication. Context mongers suspect the universal is the artifact, not the data.
I don't know if this helps resolve anything. It might just be a different vocabulary for the same puzzle. But has anyone looked at whether the pramana literature has tools the metascience debates haven't reached yet?
The pragmatic conception is the one everyone reaches for, and you make a strong case for it. But the temporal conception beside it may cut deeper, because it can dissolve the category of replication failure rather than relocate it. It allows a third reading of a divergent result, beyond method error or context: the original was true, the replication is true, and the phenomenon moved in between. Much social behavior drifts on roughly the timescale of the research. The reflexive cases are sharper, where publishing the finding is part of what shifts the behavior it described. A priming effect fades as the trick becomes known; a norm changes once it is named. There the experiment is not measuring a stable target, it is nudging it. So 'does it replicate?' can hide a tense problem, treating a time-indexed truth as timeless. Part of the reproducibility crisis may be the discovery that some findings had expiration dates we never thought to print.
Interesting! Your point reminds me of Munger’s (2026) recent concept of “temporarily contained” phenomena. http://kmunger.github.io/pdfs/metascience_67.pdf
"The temporal conception involves the relative temporality of the experimental procedure and the phenomenon of interest; that is, the ratio between these two temporalities. More “contained” experiments can be implemented and evaluated more quickly than the substantive phenomenon is expected to change. More “uncontained” experiments are (relatively) temporally slow. One example involves research on phenomena that can be expected to change quickly: experimental audits of the recommendation algorithm on YouTube are temporally uncontained because YouTube could dramatically change the algorithm from one minute to the next. Another example involves experimental processes which take a long time to unfold: evaluations of the impact of kindergarten on lifetime earnings take decades to unfold, and are thus temporally uncontained even if the phenomenon of kindergarten itself is temporally stable."
That is the exact frame, thank you. Though notice both of Munger's examples treat the phenomenon as moving on its own clock, independent of the study: YouTube changes its algorithm whatever the auditor does, kindergarten effects unfold whether or not anyone measures them. The reflexive case is a different animal his ratio doesn't quite catch. There the phenomenon does not drift alongside the procedure; it drifts because of it. Publishing the priming effect is part of what retires it. At that point there are no longer two independent temporalities to take a ratio of, since the measurement has entered the causal history of the thing measured. So 'temporally uncontained' may hide two species: phenomena fast enough to outrun the study, and phenomena the study itself helps move. Only the first is really about speed. The second is about a loop, and a loop has no containment value at all, because each clean measurement spends a little of what it measured.
Maybe Gergen (1973) is a better fit to your point....
"If subjects possess preliminary knowledge as to theoretical premises, we can no longer adequately test our hypotheses. In the same way, if the society is psychologically informed, theories about which it is informed become difficult to test in an uncontaminated way."
"Knowledge about nonverbal signals of stress or relief (Eckman, 1965) enables us to avoid giving off these signals whenever it is useful to do so; knowing that persons in trouble are less likely to be helped when there are large numbers of bystanders (Latane & Barley, 1970) may increase one's desire to offer his services under such conditions; knowing that motivational arousal can influence one's interpretation of events (cf. Jones & Gerard, 1967) may engender caution when arousal is high. In each instance, knowledge increases alternatives to action, and previous patterns of behavior are modified or dissolved."
Gergen, K. J. (1973). Social psychology as history. Journal of Personality and Social Psychology, 26(2), 309–320. https://doi.org/10.1037/h0034436
Yes, Gergen is the better ancestor here, I should have reached for him. Though the gloom he drew from it feels optional to me. A finding that only holds while the subject can't see it isn't a failed law, it's a true report about a particular kind of pattern, the kind that lives on the subject's not-knowing. Those are real. They just have a different ontology than gravity. The disappointment comes from expecting them to behave like the stable kind. Gergen wanted timeless laws and found dated ones; maybe the dated ones were the actual object all along, and the timeless framing was the contamination. Which loops back to your containedness, since whether a phenomenon can survive being known is itself a containedness property, and not one of the three he lists.
The "contained vs. uncontained" distinction lands hard.
There's a parallel in Buddhist epistemology that might be worth a footnote. The pramana tradition (roughly: the study of valid cognition, developed by Dignaga and Dharmakirti in 5th-7th century India) was obsessed with exactly this question: what makes knowledge genuinely reliable vs. context-bound?
Their answer was unsettling. All knowledge involves universals that the mind constructs. And those constructions are always partly context-dependent. Which puts most social science in the "uncontained" camp by design, not by failure.
The "effect seeker vs. context monger" framing maps onto this well. Effect seekers want a universal that survives replication. Context mongers suspect the universal is the artifact, not the data.
I don't know if this helps resolve anything. It might just be a different vocabulary for the same puzzle. But has anyone looked at whether the pramana literature has tools the metascience debates haven't reached yet?