Cycle-Time and Lag Retrospective¶
Review — instantiates Temporal Orchestration Design
Reviews finished work by comparing predicted against actual durations, waits, and feedback lags, then feeds the gaps back to recalibrate the next cycle's temporal estimates and rules.
Most timing errors repeat because nobody measures the last cycle's timing against what was assumed. Cycle-Time and Lag Retrospective is the backward-looking closure step that takes a completed run — a sprint, a patient episode, a shipment, a campaign — and lays the predicted timeline beside the actual: how long each stage took, how much of the elapsed time was waiting rather than work, where feedback arrived too late to act on. Its defining move is that the point of the comparison is to change the model, not to grade the people: each systematic gap between forecast and outcome becomes an adjustment to the estimates and rules the next cycle will run on. It looks only at what already happened; it does not set the recurring rhythm or open the next piece of work.
Example¶
An emergency department suspects its door-to-doctor and door-to-disposition times are worse than its own targets claim, but each shift blames a different bottleneck. The retrospective pulls a month of completed visits and reconstructs each one's real timeline from timestamps: triage, room assignment, physician first contact, labs ordered → resulted, consult requested → consult arrived, disposition decision, bed request → transfer. Two patterns surface that no single shift could see. First, "physician time" was estimated at ~45 minutes but the elapsed time from room to disposition averaged nearly three hours — most of it waiting on lab turnaround and specialist consults, not clinician work. Second, the consult lag (request to bedside) ran so long that decisions were routinely made against stale information or delayed until it arrived.
The output is a set of recalibrations, not blame: the elapsed-time model for an admitted patient is rebuilt to separate active care from waiting; the consult-lag figure is written into the escalation rule so a consult not answered within the measured typical window is auto-escalated. The next month's plan assumes the durations that actually occur, not the ones the team wished for.
How it works¶
- Reconstruct predicted vs. actual. For a set of completed cycles, place the planned timeline beside the measured one, stage by stage, from real timestamps rather than recollection.
- Decompose elapsed time. Split each interval into active work versus waiting, setup, transfer, and recovery, so hidden queues become visible instead of hiding inside "it took a while."
- Isolate the lags. Measure sensing, decision, and feedback delays separately; a decision that was correct but arrived against a changed state is a lag finding, not a judgment error.
- Recalibrate the rules. Convert each systematic gap into a concrete change — a revised duration estimate, a new escalation threshold, a corrected buffer size — and hand it to the next cycle.
Tuning parameters¶
- Lookback window — how many completed cycles feed the review. More cycles give a stabler estimate but blur recent shifts; fewer react faster but sample noise.
- Decomposition depth — how finely elapsed time is split. Finer splits locate the true waiting but cost data-collection effort and can over-fit.
- Attribution stance — how strictly findings are kept systemic rather than individual. A blame-free stance surfaces honest data; a punitive one drives the timestamps underground.
- Recalibration aggressiveness — how much a single retrospective is allowed to move the standing estimates. Large swings track change quickly but risk chasing one unusual cycle.
- Cadence coupling — whether the retrospective fires per cycle, per milestone, or on demand; too frequent burns attention, too rare lets bad estimates persist.
When it helps, and when it misleads¶
Its strength is closing the loop that most temporal systems leave open: it converts lived timing into better forecasts, so the chronic optimism of plans that count effort while ignoring waiting gets corrected by evidence — the disciplined "outside view" of estimating from a reference class of past actuals rather than a fresh optimistic guess.[n1]
Its failure mode is the ritual retrospective that produces observations and no recalibration — findings are discussed, nothing updates the model, and the same lag reappears next cycle. The classic misuse is turning it into a performance tribunal, which corrupts the very timestamps it depends on as people learn to make the numbers look good. A subtler trap is over-fitting to one dramatic cycle and swinging the estimates too far. The guarding discipline is to require every retrospective to end in a specific, written change to an estimate or rule, keep attribution systemic so the data stays honest, and weight findings across enough cycles that the recalibration tracks the trend rather than the last surprise.
How it implements the components¶
closure_stop_and_temporal_learning_loop— it is the learning loop: retrospective evidence from a closed cycle that changes future estimates and rules.duration_effort_and_elapsed_time_model— it rebuilds the elapsed-time model from actuals, separating active work from waiting, setup, transfer, and recovery.latency_delay_and_feedback_lag_register— it measures sensing, decision, and feedback lags after the fact and writes the observed figures back into escalation windows.
It does not implement cadence_recurrence_and_review_cycle — setting the forward recurring interval (and when to adapt it) is Cadence Calendar with Adaptation Rule, its nearest twin; this retrospective looks backward to recalibrate, whereas the calendar looks forward to schedule.
Related¶
- Instantiates: Temporal Orchestration Design — supplies the closure-and-learning step that recalibrates the next cycle.
- Consumes: Cadence Calendar with Adaptation Rule — its recalibrated estimates and lags feed the calendar's adaptation signal and the buffer sizes downstream.
- Sibling mechanisms: Cadence Calendar with Adaptation Rule · Dependency Network and Critical-Path Map · Buffer and Float Allocation Model · Rolling-Wave Schedule · Event-Triggered Rescheduling Controller · Phase-Alignment and Staggering Plan · Time-Window and Readiness Gate · Timebox and Timeout Rule · Temporal Scenario and Stress Test
Editorial Notes¶
Form Classification¶
Form family: Assessment, Review & Assurance
Rationale: Cycle-Time and Lag Retrospective operates as a bounded evaluation of existing evidence or work that produces a finding or disposition because it reviews finished work by comparing predicted against actual durations, waits, and feedback lags, then feeds the gaps back to recalibrate the next cycle's temporal estimates and rules.
Independent corroboration: The frozen evidence defines Cycle-Time and Lag Retrospective as 'Reviews finished work by comparing predicted against actual durations, waits, and feedback lags, then feeds the gaps back to recalibrate the next cycle's temporal estimates and rules', so its operative form is Assessment, Review & Assurance.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Organizational & Management Science
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Multi-domain
Rationale: Lean, project, and agile management cohered post-cycle reviews comparing planned and actual stage durations, waiting time, and delayed feedback to revise future process estimates.
Related originating lineages:
- Operations Research — Queueing and flow analysis supplied separation of service time from waiting and formal lag measurement.
Review resolution: Lean, project, and agile management cohered post-cycle reviews comparing planned and actual stage durations, waiting time, and delayed feedback to revise future process estimates.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] Reference class forecasting predicts a task's duration or cost from the distribution of outcomes in a class of comparable past efforts — the "outside view" — rather than from an inside-out estimate of the specific case, which is the standard corrective for the chronic optimism of plan-time estimates. ↩