Across three PIs the org has been getting smaller and more reliable at the same time. Capacity, commitment and velocity have declined in each of the last two PIs, production deployments have fallen a quarter since PI 9, and output per engineer has dropped in both PIs — 7% then a further 4%, about 11% in total. Over the same period sprint-goal achievement went 76.3% → 76.2% → 83.6%, clearing the 80% target for the first time. One number refused to move at all: demand has sat between 192% and 205% of capacity for three PIs, and for four if you include PI 8. Twice as much work surfaces as the org can do, no matter how the plan is drawn.
| Metric | PI 9 | PI 10 | PI 11 | PI 9 → 11 |
|---|---|---|---|---|
| Capacity | 766.8 | 744.9 | 718 | ▼ 48.8 |
| Commitment | 1,212 | 1,011.5 | 984.1 | ▼ 227.9 |
| Committed Done | 554 | 449.4 | 440.9 | ▼ 113.1 |
| Creep | 366.6 | 481.1 | 426.4 | ▲ 59.8 |
| Velocity | 746.4 | 708 | 680.4 | ▼ 66.0 |
| Demand ÷ capacity | 205% | 196% | 192% | ▬ 13 pts |
| Velocity ÷ capacity | 99% | 96% | 92% | ▼ 7% |
| Completion diagnostic | 47.1% | 46.4% | 44.7% | ▬ |
| Creep diagnostic | 30.6% | 52.9% | 42.7% | ▬ |
| Sprint goals hit | 58/76 · 76.3% | 48/63 · 76.2% | 61/73 · 83.6% | ▲ 7.2 pts |
| MRs merged | 1,262 | 1,082 | 1,087 | ▼ 175 |
| Commits | 18,774 | 14,642 | 14,348 | ▼ 4,426 |
| Bugs resolved | 130 | 123 | 113 | ▼ 17 |
| Production deploys | 8 | 6 | 6 | ▼ 2 |
| Staging deploys | 22 | 70 | 36 | ▲ 14 |
| Delivery roster | 30 | 30.7 | 30.7 | ▬ flat |
The roster is the control. Headcount barely moved across the three PIs — 30, 30.7, 30.7 delivery people — while capacity fell 6%, velocity 9% and production deployments 38%. Nothing in the volume decline is explained by people leaving.
This is the one measure that improved, and it improved sharply. Two PIs stuck at 76% then a jump to 83.6%. Underneath it, four of five teams are at or above target in PI 11 and two are perfect: Fusion has now hit 19/19 and 100% in both PI 10 and PI 11, and Web went from 40% in PI 10 to 7/7 in PI 11 — the largest single-team swing in the window.
Mobile is the exception and has been for all three PIs: 48%, 47%, 57%. It is improving, but it has never been within 20 points of target, and in PI 11 it accounted for 8 of the org's 12 missed goals. The org figure clears 80% despite Mobile, not with it.
Teams tick goal checkboxes after a sprint closes, sometimes weeks later. A weekly report run the week of a sprint and the same query run a month later return different numbers — PI 11 S3 read 8/12 on 10 September and 11/12 on 16 October. All three PIs here are historical and settled, so the comparison is sound, but in-flight goal figures are provisional and should carry an as-of date.
Goal quality is not measured at all. A team can raise its hit rate by writing smaller goals, and nothing in the warehouse would show it. Fusion's 100% alongside a 32% fall in per-engineer output is the case to look at.
The team spread tells you more than the org average. Web fell from 154% in PI 10 to 116% — it is now the only team planning inside its own capacity, and it hit every goal it set. Cloud and Mobile both sit at 227%, and between them own almost every missed goal in PI 11. The correlation across the window is not subtle: the teams closest to 100% demand are the teams hitting their goals.
Activity dipped in PI 10 and steadied in PI 11 — 1,262 → 1,082 → 1,087 MRs, and 18,774 → 14,642 → 14,348 commits. What recovery there was is entirely Cloud and Mobile. Data, Fusion and Web all fell again: Data 2,037 → 1,994 → 683 commits, Fusion's MRs 81 → 131 → 74, Web's MRs 93 → 99 → 68 and commits 1,309 → 1,526 → 814.
Read alongside points, the divergence is the interesting part. Mobile's commits are up 19% on PI 10 while its points per engineer fell 24%. Cloud is up on both. Web is the mirror image — code activity down a third, points per head up 12%. For Data and Fusion, points and activity are falling together.
dev only.Production deployments fell a quarter from PI 9 to PI 10 and have been flat since: 8 → 6 → 6. Staging tripled in PI 10 (22 → 70) and then fell back to 36. Work keeps reaching staging; a small and no longer growing share of it reaches production.
Lead time is the clearest regression in the window. Cloud went from a 6–15 day median in PI 9 to 39–40 days in PI 10, recovering to 24–25 in PI 11 — better, but still three and a half times the 7-day target and nowhere near where it started. Web moved the wrong way: 34 days to 53 on prolaio-workbench. Neither team has been at target in any PI in this window, and three of five teams cannot be measured at all.
| PI 9 | PI 10 | PI 11 | Points/eng | |||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Team | Vel | Dem | Goals | Vel | Dem | Goals | Vel | Dem | Goals | index |
| Cloud | 166.3 | 191% | 83% | 164.3 | 167% | 94% | 199.6 | 227% | 89% | 121 |
| Data | 113 | 281% | 86% | 117.2 | 223% | 50% | 83.2 | 195% | 86% | 94 |
| Fusion | 119 | 214% | 100% | 129.5 | 198% | 100% | 120 | 197% | 100% | 79 |
| Mobile | 230.1 | 221% | 48% | 180.5 | 247% | 47% | 147.6 | 227% | 57% | 76 |
| Web | 118 | 145% | 88% | 116.5 | 154% | 40% | 130 | 116% | 100% | 112 |
Four PIs of total scope against capacity: 211%, 206%, 200%, 196%. Four PIs of completed work against capacity: 97%, 97%, 95%, 95%. Teams have changed how they plan, how much they commit and how many goals they set; neither of these numbers moved more than a few points. This is a prioritisation and staffing question for leadership, not a discipline question for teams, and it is the single most stable finding in the dataset.
Both are shown here as diagnostics with no pass/fail colouring. They share a denominator the team sets at planning, and the 80% / 20% targets pull it in opposite directions. Across 93 sprints, no sprint has hit both. The volume of unplanned work arriving is statistically unrelated to the size of the plan (r = −0.11, p = 0.30), so creep ratio moving is denominator arithmetic, not less interruption. Restricting the measure to honestly sized plans — commitment within 80–120% of capacity — was tested on the 20 sprints that qualify: completion was still 50%, creep still 52%. The predictability problem is real; this instrument cannot find it.
The clearest pattern in the team table. Web cut demand from 154% to 116% and went from 40% to 100% on goals. Cloud and Mobile are the two teams at 227% demand and own almost every missed goal in PI 11. Fusion, the only other team at 100% on goals, runs at 197%, which is the counter-example — but Fusion's goal count fell alongside its output, which is the other way to raise a hit rate.
Production deployments fell from 8 to 6 between PI 9 and PI 10 and have stayed there, while staging swung from 22 to 70 and back to 36. Lead time to production has not been inside the 7-day target for any team in any PI in this window, and Cloud's best PI in the window — PI 9, at a 6-day median — is now four times further away. Three of five teams have no production environment mapped at all, so most of the org's flow is simply not measured.
No open-bug backlog: issues carries only completed bugs, so inflow and ageing remain
invisible across all three PIs. No cycle time, no WIP. No measure of goal quality, which is now the
metric carrying the most weight. These were the gaps named at the end of PI 10 and they are the same
gaps at the end of PI 11.
Source. prolaio-engineering-metrics.metrics in BigQuery, pulled
16 October 2026. PI 9 ran 4 Feb – 29 Apr 2026, PI 10 28 Apr – 4 Aug 2026, PI 11 22 Jul – 14 Oct 2026.
Each team has its own sprint boundaries and every window-based figure uses the team's own dates.
IP sprint. The last sprint of each PI (S6) is the Innovation & Planning
sprint. Rate metrics — demand, commitment and velocity against capacity, completion and creep — are
computed over S1–S5. Totals and counts include all six sprints. sprint_pi_summary pools
all six for its ratios, so its completion and creep figures differ from these.
Ratios are point-weighted (SUM(x)/SUM(y)), never averages of
per-sprint ratios. The PI 8 figures quoted in the demand section come from the preceding
PI 8–10 trend analysis and use the same all-sprint basis.
Per-engineer index. Each team against its own prior-PI per-engineer average; 100 means unchanged. Delivery roster = engineers, QE and interns, excluding Product Owners and PMO. The index is used rather than raw per-head levels because capacity per engineer ranges from 2.1 to 3.9 points across teams — that is estimation scale, not availability — and because the index moves by at most 1–2 points across every denominator definition tested, while raw levels do not. Three PI 11 joiners have unconfirmed roles (Domenic LaVitola, Mobile; Grant Smith and Tyler Lofgren, Fusion) and are counted as delivery.
Sprint goals are parsed from markdown checkboxes in sprints.sprint_goal.
Teams edit them after sprint close, so figures are as of 16 October 2026.
DORA coverage is partial. Only Cloud and Web have environments with a
production or staging tier in dim_environment; the view is deduplicated by
(team, environment) before aggregating. Lead time is reported per project because the median of a
pooled population is not the pooled median.