All revenue is net — it excludes tax and service charge. Day and category money are exact; item-level money is an estimate apportioned within its category. Lunch and dinner are the POS's own category tags, which begin 2026-05-19 when the categories were renamed — no daypart exists before that date, so this window is 15 weeks and roughly fourteen observations of each weekday, not the full 175 days the store holds from 2026-01-24. 1,154,528 LKR — 3.5% of net revenue in the window, being Bar Snack, Happy Hour, Cocktail Club and T20 Worldcup — carries no daypart and is excluded from every lunch-versus-dinner figure. Easter Menu and Father’s Day Edition are counted as lunch. Mondays are a standing closure; 4 further non-Monday days inside the window were recorded closures, which is why 15 weeks hold 85 days.

Lunch % of that day’s tagged net is lunch over lunch-plus-dinner, not over everything the day took. First-shift % of takings is a second, independent measure: the first of two shifts on a day that ran two, from the payments settled in it. Those payments include tax and service charge, so that column is a share only and is never placed beside a net figure. It carries its own day count because a day that ran one continuous service cannot be split — on Sunday it rests on 6 of 13 days. Across 74 days where both measures exist they correlate at 0.928, the clock reading 12.0% against the tag's 14.8%.

Row minus column. A figure is shown only where a two-sided permutation test separates the two days at p < 0.05; · means it does not. 4 of 15 pairs separate, all of them involving Saturday or Sunday; no pair among Tuesday, Wednesday, Thursday and Friday does. These are 15 simultaneous tests — Holm-corrected for that, 2 survive (Fri–Sun and Wed–Sun). Pairs sitting near the threshold rather than clear of it: Wed–Sat at p = 0.034 and Thu–Sun at p = 0.066 and Fri–Sat at p = 0.053. Across all six weekdays taken together, p = 0.001.

All 20 ways to pick three lunch days from six, each the sum of those three weekday means. On that basis a full week of lunch is 346,498 LKR; the mean week actually observed was 321,935, and across the 12 weeks that ran a full six days, 333,495. The proposal ranks 2; Thu / Sat / Sun ranks first, 15,371 LKR a week ahead of it. Tue / Sat / Sun ranks third, 753 LKR behind the proposal. Taken as two blocks rather than six days, the three the proposal keeps average 69,387 LKR of lunch a day against 44,530 for the three it closes — 24,857 LKR, p = 0.018.

Total lunch net LKR per week, all six weekdays combined. Range 183,750 to 516,880 (week of 2026-07-07); mean 321,935. Hollow markers are the 3 weeks that traded fewer than six days because a closure fell in them — the last week shown, 2026-08-25, is one of them at 4 days.

Blank where the day did not trade. 3 of the 15 weeks are short for that reason.

Lunch net LKR for each weekday, week by week, on one shared vertical scale so the six panels are comparable. ρ is the Spearman rank correlation of that weekday’s lunch against the week index, shown with its p only where it reaches p < 0.05. Gaps are weeks where that weekday did not trade.

Spearman rank correlation across the 15 weeks, p by permutation. No early/late split is used: with roughly fourteen observations per weekday the answer depends heavily on where such a split is placed, so none is taken.