1Demonstration 1 of 4
The review window picks the winner
When does a team that reserves capacity for repayment overtake one that pushes hard?
Repayment spends capacity now and drains the debt that later creates defects and drag. Early on only the spending shows; later the saved capacity shows, and the paths cross.
Features delivered is a cumulative count. The repayment share is the fraction of usable capacity reserved for paying debt. A period is one time step; capacity is 10 engineer-weeks per period.
Predict first. At a review in period 12, which policy has delivered more, push hard or repay 20 percent?
Choose an example
Constructed example: the chapter's two policies, recomputed with the pack's run function.
Calculated values
- Review at period
- 12
- Push hard, features
- 100.1
- Repay 20%, features
- 87.2
- Repaying minus push hard
- (-12.9)
- Debt under push hard
- 50.1
At period 12: 87.2 - 100.1 = (-12.9) features for the repaying policy. The larger total over the smaller is 100.1 / 87.2 = 1.148. At this review push hard looks better, and a short window would confirm it.
Use the idea
Ask what horizon a policy comparison was scored on before accepting it, and what the two policies look like one review later.
Where the conclusion applies
Constant pressure, the pack's drag and defect rates, and cumulative features as the score. A short-lived codebase is the case where the short window gives the right answer.
Check your understanding: At period 60, how many times more features has repay 20 percent delivered than push hard?
Chapter 33 source: section "Why the first review favors the wrong policy". Demonstration C33-D01.
2Demonstration 2 of 4
Capacity is not what the plan says
How much of the nominal capacity exists once debt drag and morale are subtracted?
Drag takes the drag coefficient times the debt off nominal capacity, then morale scales the remainder. Debt only grows under push hard, so the gap widens with every period.
Nominal capacity is 10 engineer-weeks per period. Debt is shortcuts taken and not repaid. The drag coefficient is capacity lost per unit of debt. Morale scales what is left.
Predict first. After a year of pushing hard, is available capacity closer to 10 or to 6?
Choose an example
Constructed example: states from the chapter's push-hard run, recomputed with the pack's available_capacity function.
Calculated values
- Debt carried
- 115.0
- Morale
- 0.82
- Drag coefficient
- 0.020
- Available capacity
- 6.31
- Plan over reality
- 1.58
Drag = 0.020 x 115.0 = 2.30. Available = max(0, (10.0 - 2.30) x 0.82) = 6.31. Plan over reality = 10.0 / 6.31 = 1.58. A plan built on nominal capacity assumes 1.58 times what exists.
Use the idea
Compare the capacity a delivery plan assumes with the capacity measured over the last quarter.
Where the conclusion applies
The states are those of the push-hard run. Debt and morale are proxies in practice, so the exact available figure is an estimate; the direction of the gap is the claim.
Check your understanding: With debt 50 and drag 0.02, what is the capacity after drag, before morale?
Chapter 33 source: section "Capacity is not what the plan says". Demonstration C33-D02.
3Demonstration 3 of 4
Rework share moves before delivery does
Which number warns first, features delivered or the share of capacity spent on rework?
Debt surfaces defects in proportion to itself. Each defect fixed takes 0.6 engineer-weeks from delivery, so the rework share climbs from the first periods, and a standing repayment share slows the climb.
Rework share is capacity spent fixing defects divided by usable capacity. Each defect fixed costs 0.6 engineer-weeks. Repayment share is held back before delivery or rework.
Predict first. With no repayment, is the rework share in period 6 above or below 10 percent?
Choose an example
Constructed example: the chapter's push-hard and repaying runs, rework share rebuilt from the pack's step rule.
Calculated values
- Repayment share
- 0.00
- Period read
- 12
- Usable capacity
- 8.65
- Rework share
- 24.0 percent
- Features delivered that period
- 6.57
In period 12 the team starts with 3.47 open defects and 8.65 usable capacity. After repaying 0.00 x 8.65 = 0.00, it fixes min(3.47, 8.65 / 0.6) = 3.47 defects, spending 3.47 x 0.6 = 2.08 on rework. Rework share = 2.08 / 8.65 = 0.240, which is 24.0 percent.
Use the idea
Add a rework category to how work is tracked and plot its share beside delivery volume.
Where the conclusion applies
Rework is booked honestly. In many teams it is booked against the feature it belongs to, which hides this signal.
Check your understanding: With 4 open defects and 9 units to deliver, what is the rework share of 9 usable units?
Chapter 33 source: section "The three-act shape". Demonstration C33-D03.
4Demonstration 4 of 4
Test the conclusion against the proxy's range
Does the repayment conclusion survive not knowing the drag coefficient?
A larger drag makes debt more costly sooner, so the repaying policy overtakes earlier. At a short horizon and a small drag the lead can stay with push hard.
The drag coefficient converts debt into lost capacity and cannot be read from a register. Debt is shown as a direction, not a level. The horizon is the period at which features are compared.
Predict first. At period 60, does the repaying policy lead for every drag value from 0.005 to 0.03?
Choose an example
Constructed example: drag values defined for this reader around the pack's 0.02, run with the pack's run function.
Calculated values
- Drag coefficient
- 0.020
- Horizon (periods)
- 60
- Push hard, features
- 230.1
- Repay 20%, features
- 319.0
- Repaying minus push hard
- 88.9
With drag 0.020 at period 60: 319.0 - 230.1 = 88.9 features. Debt under push hard is 115.0, so the drag is 0.020 x 115.0 = 2.30 engineer-weeks. Repaying leads at this horizon.
Use the idea
State the recommendation with the range of the unmeasured value, and say whether the range changes the decision.
Where the conclusion applies
Drag is held constant over time and across the codebase, which real systems do not do. The conclusion also depends on the shortcut and defect rates being as set.
Check your understanding: If push hard has 152.7 features and repaying has 154.9, which leads and by how much?
Chapter 33 source: section "Making the invisible reviewable". Demonstration C33-D04.