The decision needs a review trigger

The decision was still visible, still internally consistent, and no longer appropriate.

In a synthetic D4U7 scenario, a study circle chose a weekly session format because attendance was small, recording was unnecessary, and one shared timezone made the schedule fair enough. The decision record explained the choice clearly.

Two months later, the scenario added participants in another timezone and a need to review material asynchronously. Nothing in the original record was false. Its assumptions had expired.

The room had preserved why the decision was made and provided no reason to look at it again.

I added a review trigger: the observable future condition under which the decision should return to active attention.

A timestamp was the weakest useful trigger

The first implementation offered Review on date. A calendar reminder was better than permanent silence, but many decisions did not become questionable because a date arrived.

The session-format decision depended on group size, timezone spread, and asynchronous access. Reviewing it every month would create ceremony while those conditions stayed stable. Waiting a year could leave it wrong for eleven months after the assumptions changed.

I kept date triggers for decisions with genuinely time-based uncertainty: a temporary experiment, a policy tied to a season, or a dependency whose support window ended. Other triggers named state:

• Participant timezone spread exceeds a chosen bound.
• The evidence source is superseded or withdrawn.
• A previously rare conflict appears in several recent rooms.
• The selected tool no longer supports the required browser range.
• A cost or resource limit crosses the assumption recorded in the rationale.
• Dissent accepted as low risk becomes observable harm.

A date could still act as a backstop when no automatic observation existed. It stopped being the only model for future doubt.

The trigger came from the decision's assumptions

A review trigger invented after the decision was often generic: “revisit later.” The useful trigger emerged while writing the rationale.

The D4U7 decision form asked:

• Which proposal revision is selected?
• What evidence supports it?
• Which constraints make it preferable now?
• What dissent or residual risk is accepted?
• Which change would weaken this reasoning enough to reconsider?

The final question converted an assumption into a future observation.

For the session-format example:

Decision
  Use one live weekly discussion without recording.
 
Current basis
  Five participants share a three-hour timezone range.
  All can attend the selected window for the next six sessions.
  The topics do not require an asynchronous review artifact.
 
Review when
  The timezone range exceeds five hours, two sessions lack required attendance,
  or a topic requires later review by someone absent.

The trigger did not predict the future. It named which future would make the current reasoning incomplete.

A decision could cite a proposal, test result, policy, or external source. Those references might change after the decision.

D4U7 linked the exact immutable evidence revision used. If a current source was superseded, the earlier basis remained visible and the trigger could fire. The system did not silently update the decision to cite newer content it had never considered.

Evidence relationships had states:

• Current and unchanged.
• Superseded by a newer revision.
• Withdrawn or unavailable.
• Contradicted by a new observation.
• Still historical basis but no longer current guidance.

The trigger could not decide whether the newer evidence reversed the conclusion. It opened review with a digest of what changed.

This prevented two opposite errors: pretending the decision had always used current evidence, and invalidating it automatically whenever a linked document received a typo correction.

Materiality belonged to the evidence type. A changed constraint field could fire directly; a revised long-form source might ask the decision author or current room participants to classify the change.

Dissent was a source of future tests

The final record allowed acknowledged dissent. A concern could be answered, accepted as a tradeoff, or left unresolved without granting a veto.

The accepted concern often contained the best review trigger.

Suppose the chosen session format omitted recordings because storage, privacy, and review needs did not justify them. A dissenting participant argued that absence would become exclusion if schedules spread. The decision could proceed while recording:

Accepted risk: participation depends on one live window. Review after two sessions where a required participant cannot attend, or when timezone spread exceeds five hours.

The system did not score whether the dissenter had been “right.” It preserved the observable consequence they had named.

This gave dissent durable value beyond a footnote. A future review began with the concern and the evidence now available.

It also reduced the need to repeat objections indefinitely. The concern had a place and a condition under which it would demand attention again.

A trigger opened review; it did not reverse the decision

Automatic rollback sounded attractive for measurable triggers. If the group grew, switch the session format. If an error rate rose, restore the previous architecture. If an evidence link changed, suspend the decision.

The trigger established that the old reasoning needed review. It did not know the best replacement or whether other constraints had changed.

D4U7 created a review event and a bounded room state:

trigger observed
  → review requested
  → current decision remains effective | temporary containment applied
  → evidence and proposals updated
  → reaffirm | amend | replace | retire

For low-consequence decisions, the current choice remained effective while review proceeded. For a safety or permission decision, the trigger could also name a pre-agreed containment action, but that was an explicit policy rather than generic behavior.

The review displayed which condition fired, the evidence supporting it, and the original assumption it challenged.

Automation brought the question back. People retained responsibility for the answer.

A review request with no responsible reader could remain open forever. D4U7 used an owner label tied to the room: decision author, current action owner, or a named participant role for the study scenario.

The personal project did not simulate a corporate governance chain. In many tests I was the only operator and used labels to make the state recoverable after interruption. The invited study circle used actual room participants only for its low-consequence topics.

Ownership meant someone or some current room role would receive the review digest and could reassign or close it. It did not imply exclusive authority to decide.

If the original participant left the room, assignment followed an explicit fallback chosen at decision time. The system did not retain notification authority for a removed account merely because old metadata named it.

An unowned trigger surfaced as a room maintenance condition. Archive could not hide it silently.

The design kept accountability concrete without manufacturing a management structure around a small experiment.

It was easy to write a condition the system could not detect: “review if this becomes confusing” or “review if the architecture gets too complex.”

Such statements were still useful as human guidance but could not promise automatic notification. D4U7 distinguished:

• Observed trigger: derived from a typed room event or trusted external input.
• Scheduled prompt: asks a person to evaluate a qualitative condition at a date or interval.
• Manual trigger: any participant with appropriate room authority can open review and cite evidence.

The session-attendance trigger could observe explicit attendance acknowledgements only if the study chose to record them. I avoided passive presence telemetry. If attendance was not a justified retained event, the trigger became a short scheduled prompt instead of a hidden tracking feature.

This linked governance to data minimization. A desire for automatic review did not authorize collection of every behavioral signal that might someday support it.

The trigger's implementation field said how the condition would be known and what blind spots remained.

Thresholds required hysteresis and context

A numeric trigger at an exact boundary can flap. If the timezone range alternated around five hours as one participant's schedule changed, the room could repeatedly open and close review.

D4U7 treated firing as an event that stayed open until addressed. It did not withdraw the review merely because the measurement dipped back below the line. The review could conclude that the change was temporary, adjust the threshold, or reaffirm the decision with a new trigger.

For continuously observed signals, I used a duration or count where the domain justified it: two affected sessions rather than one momentary schedule change. The exact policy remained visible beside the trigger.

I avoided false precision. “Timezone range exceeds five hours” made sense because the original decision named a scheduling assumption. “Engagement score falls below 72” would introduce a model whose relationship to decision quality was unclear.

Thresholds encoded patience and consequence. They needed the same justification as alerts in W93H.

Review produced a new decision record

When a trigger fired, D4U7 did not unlock the old decision for editing. The review ended with a new immutable record:

• Reaffirmed: current evidence still supports the choice, with updated rationale and next trigger.
• Amended: the choice changed within the same direction, preserving the prior record.
• Replaced: a different proposal became effective.
• Retired: the decision no longer governed anything.
• Deferred: evidence remained insufficient, with temporary state and another trigger.

The new record linked the triggering observation, original decision, evidence revisions considered, dissent, effective time, and follow-up actions.

Reaffirmation was not a no-op. It said the decision had survived a named challenge under newer evidence. It also prevented a trigger from firing forever against an already reviewed condition.

The overview showed the current decision and its lineage. A reader could understand that the current state resulted from revision rather than an in-place edit that erased earlier assumptions.

Rooms could accumulate decisions whose dates passed, evidence changed, or owners disappeared. If that work lived only in notifications, muting a channel would hide it.

The room overview included a maintenance section:

• Review requested by observed trigger.
• Scheduled review due.
• Evidence changed; materiality not classified.
• Trigger has no current owner.
• Decision has no review trigger by explicit choice.

The last state mattered. Some decisions were intentionally final within the room's lifetime or too trivial to justify review machinery. The absence was recorded as a choice, not inferred from a blank field.

Archived rooms displayed unresolved review debt before closure. A room could still archive with a waived item if the reason and residual effect were documented.

This borrowed W93H's recovery conditions. Important obligations needed durable state beyond the alert that first announced them.

Not every decision deserved the full structure

The first decision form required rationale, evidence, dissent, owner, effective time, and review trigger for choosing a discussion title. Participants wrote filler to satisfy fields.

I scaled structure by consequence and lifetime.

A lightweight choice could record the selected option and expire with the room. A decision that governed future sessions, changed access, or accepted a material tradeoff needed fuller evidence and review. Templates suggested fields without imposing one ceremony on every question.

The interface asked for a trigger when the rationale contained explicit temporary assumptions or accepted risks. A participant could choose No review trigger needed and give a concise reason.

This kept the useful discipline from becoming bureaucratic text generation. Structure should reduce future ambiguity, not increase present compliance theater.

The study scenarios intentionally included both levels so the product did not optimize only for its most elaborate demonstration.

Some triggers depended on state outside D4U7: a dependency end-of-support date, a project cost threshold, or a published evidence revision.

I did not build a universal polling system. Each external trigger used a narrow adapter with a source identity, observation time, data shape, and failure state. A missing source produced unknown, not “condition false.”

The adapter could propose that a trigger had fired. The review record preserved the source observation and any transformation. If the source was unauthenticated or approximate, that limitation appeared in the evidence.

Manual confirmation remained appropriate for many external conditions. A person could attach a source and open review without teaching D4U7 to scrape every Web page it cited.

The architecture stayed smaller by matching automation to maintained sources. A one-off decision did not justify permanent integration.

The adapter-shaped boundary left me with a broader rule: any integration that could alter a maintained record needed explicit provenance and failure states.

A generic Decision needs review message would send the reader back into archaeology.

The notification included a safe summary:

Review requested: the participant timezone range now exceeds the five-hour assumption in the current session-format decision.

The authenticated room opened at the trigger evidence and original rationale. It showed the current decision remained effective unless a containment policy said otherwise.

Repeated observations grouped under the same open review. They did not produce a new interruption every time the threshold was sampled. A material new change could update the review digest.

Delivery followed the participant's decision-notification preference and quiet hours unless the decision itself named justified urgency.

The message carried a reason to reconsider, not an abstract maintenance count.

Evaluation looked for changed decisions, not created triggers

It would have been easy to count how many decisions had review triggers and call higher coverage better.

The study evaluation asked:

• Did the trigger follow from a real assumption or accepted risk?
• Could the condition be observed as described?
• When it fired, could a participant understand why without rereading the whole room?
• Did the review preserve the old decision and produce a new explicit outcome?
• Were triggers quiet while conditions remained stable?
• Did any trigger require disproportionate data collection?
• Could a participant deliberately decide that no trigger was warranted?

I included false-positive and missing-evidence scenarios. A trustworthy trigger needed a visible unknown state and a way to revise its logic.

Success was not trigger volume. It was a decision becoming revisable at the moment its basis changed.

The first trigger logic could itself prove wrong. A threshold might fire too often, an external adapter might change meaning, or a qualitative prompt might ask the wrong question. Editing the trigger in place would make an old review appear to have fired under new logic.

D4U7 versioned the trigger definition. Every observation named the version evaluated, the inputs available, and the result. Revising a trigger required a short rationale and stated whether the new definition should evaluate current state immediately or only future changes.

An immediate evaluation could open review as soon as the correction made an existing condition visible. It did not fabricate a historical firing time. Earlier observations remained tied to the rule that had processed them.

Disabling a trigger was also a decision. The record named whether the underlying decision now had no review path, a replacement trigger, or a scheduled manual check.

This mirrored W93H's alert-version lesson: changing the rule is not the same as changing the world. A governance tool should preserve which question it was actually asking when it demanded attention.

A durable decision includes its future doubt

The original session-format decision was reasonable for its small, colocated scenario. Growth did not make the earlier reasoning foolish. It made the constraints different.

The revised record said when those constraints would matter. When the timezone range crossed the bound, D4U7 opened review with the original dissent and new schedule evidence. The current choice remained visible until the room selected an asynchronous alternative and recorded its own review conditions.

Nothing was edited to make history appear prescient. The system showed an opinion responding to a changed world.

A decision record usually explains why a choice should be followed now. A review trigger explains why it should not be followed forever without thought.

The strongest decisions in D4U7 did not claim permanence. They named the evidence that supported them, the risk they accepted, and the future observation that would earn another look.