Colombo's 17th Over: Where Bangladesh's Baseline Cracked

Colombo's 17th Over: Where Bangladesh's Baseline Cracked

R. Premadasa Stadium, Colombo — 4 March 2026. Third ball of the 17th over. Bangladesh needed 62 from 24. Three rows below the press box I had already written the number in my notebook before the over began: baseline 11.4 runs. The ball climbed towards long-on, the deep midwicket fielder raised his hands, the over finished on four. A shortfall of 7.4 runs. The final margin was six. The over the crowd will talk about on the way home was, in numerical terms, the single largest loss of the night.

Fourteen years of ball-by-ball watching build a habit. Where the crowd is loudest, the data is quietest. At Kazan in 2026, Germany's 26 shots and 2.7 expected goals produced nothing at all; I learned then that possession and control are not the same thing. In Colombo the lesson arrived from the opposite direction.

Context and method

In this group-stage fixture Sri Lanka batted first and made 178 for six. The last three World Cup innings at this venue averaged 165 — so the winning score sat thirteen runs above the venue baseline, pushed there by evening dew and breeze.

My ER-20 model is not complicated. It is trained on 12,400 legal deliveries from the last three T20 World Cups and two Asia Cups. Inputs for every ball: phase (1-6, 7-15, 16-20), wicket-loss state, batter quality index, bowler type, venue-pitch index, dew flag. Output is an expected run value with 90 percent confidence intervals. On 2026 World Cup data, out-of-sample innings-level mean error was 7.1 runs. The code and the raw data are public, because a model you cannot re-run is not a model — it is an opinion.

Bangladesh chased 178 and finished on 172 for eight. A margin of six runs.

The phase-by-phase account

| Phase | Actual | Baseline | Deviation | |------|--------|----------|-----------| | Powerplay (1-6) | 46/1 | 48/1 | −2 | | Middle (7-15) | 78/3 | 89/2 | −11 | | Death (16-20) | 48/4 | 41/3 | +7 | | Total | 172/8 | 178/3 | −6 |

Read the table and the first surprise is not the death overs. It is the middle. The match was not lost between overs 16 and 20; it was lost in the four overs from 12 to 15. That block produced 22 runs from 24 balls, eleven dot balls and a wicket. The model's baseline for those four overs was 33. An eleven-run shortfall that ended up five runs larger than the margin.

In the death overs Bangladesh scored seven above baseline but lost four wickets where the baseline said three. Runs came, wickets came faster, and the innings never held its tempo for a full over. Bangladesh's death-over strike rate in the 2026 World Cup was 129.4 (source: my compiled ER-20 dataset, February 2026 update). In Colombo that figure was broadly held — the problem was elsewhere.

The real powerplay number

Bangladesh made 46 in the first six overs, two below baseline. By runs alone it looks precise. The dot-ball rate tells a different story. Between overs 4 and 6, 58 percent of deliveries were dots, against a tournament baseline of 44 percent at this venue. Losing time without losing wickets in the first six overs was the innings' first silent loss. The singles-per-ball rate in the powerplay was 0.38; the team's tournament average was 0.59.

The matchup: yorker against length

Across the last five overs, Sri Lankan bowlers hit yorker length on 14 of 24 deliveries, six of them wide yorkers. In my pitch map, Bangladesh's strike rate in that postcode was 92; on length deliveries across the rest of the innings it was 138. That is the forty-eight-run gap.

A caution is still required. Yorkers executed and a match lost — wiring cause to effect is easy, but the model does not support it. Bangladesh finished above baseline in the death overs. Sri Lanka's plan did not win them the game; it saved them. The winning work was done much earlier.

Overs 12 to 15: three small causes

The Sri Lankan leg-spinner bowled three of those four overs himself, taking one for 24. The baseline economy for spinners in that phase at this venue is 8.3, so roughly two runs per over were saved. Two batters together made 14 from 19 balls in that block, and the singles-per-ball rate dropped to 0.42.

The second cause is small but measurable. A DRS review in the 13th over ran three minutes and 42 seconds. The next eight balls produced five runs. Whether a long review breaks an innings' rhythm — I have twelve matches of data for that question, which is not proof, only a suspicion. That tempo matters more than theatre, though, is plain enough.

The third cause sits with squad management. The Sri Lankan seamer listed as week-to-week before the match bowled two powerplay overs, left the field, and returned only in the 16th over. In the exact window where the game turned — overs 12 to 15 — he was unavailable. We have not seen the scan; we have seen a workload chart.

Fielding residual

My xRuns Saved metric returned +7.2 runs for Sri Lanka — two diving stops, one run-out, four singles saved in the ring. The margin and the fielding residual were nearly the same number, and that is the most uncomfortable coincidence of the night.

Contrarian: did momentum actually exist?

So I ran a placebo test. I shuffled the phase labels of all twelve Asia group matches, re-ran the model a thousand times, and asked whether "the side that finds rhythm in the last five overs wins" has any predictive value. Output came back close to zero. Partial correlation between death-over strike rate and result was 0.09; between middle-over dot-ball rate and result, 0.41.

Momentum is a name, not a variable. What can be measured is the ability to rotate strike between overs 7 and 15.

The second caution points at the baseline itself. The 165 venue baseline covers three seasons, but squads have changed, balls have changed, batter intent has changed. The baseline is ageing fast on its own. Treating it as scripture is a mistake; it needs an audit every season, or we will keep giving precise answers to the wrong question.

Signal for the next match

Pakistan await in Kandy. If Bangladesh's dot-ball rate in the 7-15 block drops below 35 percent, the same eleven can chase 180 without changing a single death-over net session. If strike rotation stays stuck under 0.45 in that block, the scoreboard may look prettier and the result will not change.

The table is already built. Until Colombo's six runs and the eleven-run shortfall in overs 12 to 15 reconcile with each other, I am not prepared to treat the eye test as final evidence. A witness testifies; the data cross-examines.