Follow the same eight pull requests carrying 29 code points. Speed up their arrival, then give the reviewer more capacity. Watch work collect at the review gate and return from CI. Only a unique merge contributes to the feature.
Same work, one intervention.Baseline: eight PRs arrive 1.45 model seconds apart. PR sizes, initial risks, physics seed, model geometry, and the 20-second observation stay fixed.
Read the bodies: circles are PRs, attached squares are risks
PR coreCircle area grows with code points; arrivals are staggered by the scenario.
Negative traitSquare satellites remain attached until AI or human review transforms or removes them.
Shared serviceHuman review is one shared queue—PRs wait their turn.
FeatureOnly merged PR work fills the feature. Reviewing alone does not.
Coding throttles the system: 8 PRs carry 29 code points. Incoming work is 2.9 points per second against 5 review points per second. 0 of 18 Feature points have merged.
Event tape
What actually changed
Stream startingPR compounds will arrive on their own project-local clocks.
Reusable mechanics
What the model assumes
This is a hypothetical workflow, not measured developer productivity. Every code point consumes one review-work unit; each review has a separate remediation budget. CI checks for missing tests, not their quality. Physics makes waiting, repeated visits, and attached risk visible. The event ledger decides what merged; a collision cannot approve a PR.
compound entityOne semantic PR is a core plus repeated negative-property occurrences.
shared capacityA gate processes root entities by work units and re-arms only after a real exit.
weighted groupMerged PR values accumulate once toward a generic threshold outcome.