The retro is late by design.

A team notices a painful handoff on Tuesday. By Friday, the details have softened. By the retrospective, the event has become a story: memorable enough to survive, clean enough to tell, and detached from much of the context that made it useful.

The common response is to improve the meeting. Add a new board, a warmer prompt, a tighter format, or a better facilitator. Those changes can help the conversation. They do not change the evidence schedule. The team is still asking memory to reconstruct work after the moment has passed.

The meeting is not the problem

Retrospectives create a protected moment for collective learning. That function matters. Amy Edmondson’s field research connected team psychological safety with learning behavior: people need a climate where interpersonal risk does not automatically become punishment. A thoughtful retrospective can help create that climate.

But a meeting also compresses several jobs into one hour. People must remember events, interpret them, negotiate a shared account, decide what is controllable, and invent an action. The format asks for memory, safety, analysis, and commitment at the same time.

A longitudinal case study of 37 team retrospectives over almost three years found that discussions often centered on topics close to the team, while some issues recurred and some accounts were shaped by strong opinions unsupported by hard evidence. The authors did not conclude that retrospectives were useless. They showed that the path from discussion to organizational learning is uneven.

The opportunity is not to abolish reflection. It is to stop making reflection recover its evidence from scratch.

Delay changes the evidence

Research on ecological momentary assessment begins from a related measurement problem: global retrospective self-reports are vulnerable to recall bias and are poorly suited to showing how experience changes over time and context. That literature comes from clinical research, not software delivery, so it cannot prove that an engineering signal captured on Tuesday will be more useful than one remembered on Friday.

It does offer a disciplined design principle: if context matters, reduce the distance between experience and capture. Preserve the observation first. Interpret it later, with other people and better evidence.

This separation is important. Immediate capture is not immediate truth. A frustrated note is one piece of evidence, not a verdict on a person, a process, or a team. Solv calls it a signal precisely because it has not yet earned a stronger status.

What “continuous” should mean

A continuous retrospective is not a permanent meeting and not a feed of complaints. It is a four-stage evidence rhythm:

The live retrospective can remain, but its role changes. Instead of beginning with “What happened?”, it can begin with preserved signals, unresolved tensions, and the history of prior commitments. The scarce meeting time moves from recollection toward judgment.

Guardrails against surveillance

Continuous capture becomes dangerous when the system treats people as the thing being measured. A stream of workplace observations can easily become a performance dossier, an emotion profile, or a quiet channel for management surveillance.

Three boundaries are therefore structural, not optional:

• Pseudonymity: the improvement ledger does not carry real identity.
• Content-level affect: emotion describes a signal’s content; it never becomes a per-person profile.
• Human promotion: neither the original signal nor an AI interpretation can become team truth without resolution.

Continuous should describe the availability of reflection, not the intensity of observation. People must be able to understand what is recorded, where it travels, who can see it, and how a conclusion can be challenged.

What we would test

Solv’s thesis is not that continuous capture automatically improves a team. It is that the model produces testable advantages over meeting-only recall. The first field studies should attempt to disprove three hypotheses:

1. Shorter capture latency preserves more actionable context without increasing perceived surveillance.
2. Separating proposal from human resolution produces clearer ownership and fewer unowned action items.
3. Returning prior commitments to the next cycle reduces the recurrence of accepted-but-unfinished work.

The measures should include capture latency, resolution rate, completion rate, observed outcome, recurrence, and team-reported safety. No single number is sufficient. If context improves while safety declines, the system has failed.