Software Engineering · Cluster
Engineering Process
Planning, estimation, and incident learning.
- 01 Cycle Time, Work-in-Progress Limits, and Little's Law in Delivery Queues How the identity L = lambda W ties items in flight to delivery time, why operating near full capacity inflates waiting, and how to size a work-in-progress limit from a target cycle time.
- 02 Aggregating Task Estimates: Why Summed Ranges Come Out Too Narrow Why adding per-task ranges produces an interval that is too tight for the whole job, how skew and correlation break the diversification argument, and what to compute instead.
- 03 Critical Path, Float, and Why the Longest Chain Sets the Date How the forward and backward passes compute earliest and latest dates, what total float measures, and why a zero-float path is not necessarily the riskiest one.
- 04 Calibrating Interval Estimates Against Recorded Outcomes What an eighty percent interval promises, how coverage is measured on small samples, why width must be scored alongside the hit rate, and how reference classes supply the correction.
- 05 Forecasting Completion Dates from Historical Throughput Resampling the observed rate of finished work to produce a distribution of completion dates, and the assumptions about scope, unit, and independence that decide whether the answer means anything.
- 06 Severity Tiers, Escalation Timers, and the Cost of a Page Separating impact from urgency so a tier is decidable in minutes, converting tiers into acknowledgment and update timers, and the attention economics that decide when paging is the wrong response.
- 07 Postmortem Causation: Triggers, Latent Conditions, and Failed Defences Why a single root cause discards the information needed to prevent recurrence, how counterfactual reasoning behaves in stochastic systems, and how to rebuild a sequence from telemetry with its uncertainty intact.
- 08 Error Budgets as a Release Gate: The Arithmetic Behind the Freeze Converting a service level objective into a measured allowance of failure, normalising it with burn rate, alerting on two windows at once, and the places where the arithmetic stops matching user experience.
- 09 Decomposing Recovery Time into Detection, Diagnosis, and Mitigation Why a single mean time to recovery mixes intervals with different floors and different distributions, how concurrent incidents queue behind one responder, and how to measure each phase without corrupting the comparison.
- 10 Sequencing Work by Cost of Delay Under Precedence Constraints Why the weight-over-duration ratio is the exact ordering rule for one machine, what changes when dependencies and deadlines enter the model, and where the ratio becomes numerically fragile.