# Case: estimates versus actuals > Use three earlier Klasna launches and the three-point estimates of the chain to tell Leo whether physics can open on 1 February, and how sure you are. - Course: [Project Management Without the Bureaucracy: Run One Project from Brief to Closure](https://productovi.com/courses/project-management-without-bureaucracy.md) - Module: Estimates and the budget - Lesson 16/32, ~36 min - Source: https://productovi.com/courses/project-management-without-bureaucracy/estimates-vs-actuals-case ## Can Leo promise the date? **Leo:** Marketing wants to put "Physics opens 1 February" on the landing page next week. Once families see it, I can't take it back. Can I promise it, and how sure are you? **Alex Tutor:** You have two datasets: the five tasks of the chain that sets the opening date, as of 30 November, and 17 tasks from three earlier launches. Answer from both, not from hope. **Dataset (CSV):** [The chain to opening, as of 30 November 2026](https://productovi.com/datasets/project-management-without-bureaucracy/estimates-vs-actuals-case/klasna-physics-estimates.csv) Five tasks run one after another, in working days. From 30 November to 29 January there are 43 working days, after two days off at New Year. - `task`: Task - `owner`: Who answers for it - `owner_estimate_days`: The single number the owner first gave - `optimistic_days`: Optimistic - `likely_days`: Most likely - `pessimistic_days`: Pessimistic **Dataset (CSV):** [Three earlier Klasna launches: estimates and actuals](https://productovi.com/datasets/project-management-without-bureaucracy/estimates-vs-actuals-case/klasna-past-launches.csv) One row per task, in working days. The exam math and programming launches were subject launches; the simulator was mostly a platform build. - `launch`: Launch - `finished`: Month it finished - `task`: Task - `kind`: Kind of task - `estimate_days`: Estimate given in the plan - `padded`: yes: the owner said in the retrospective the estimate held a private margin - `actual_days`: Actual duration ## What the history says **Alex Tutor:** Start with the reference class. Physics has a curriculum, a review, demo lessons and onboarding, like the two subject launches. Take their tasks without padding: those estimates were meant as most likely values, just like the ones in the chain. **Sofia:** So I add up their estimates and their actuals and divide. **Vira:** Divide the totals, please, not an average of ratios. A long task should weigh more than a short one. **Alex Tutor:** Exactly. Then look at the padded rows separately. > **Exercise:** In the two **subject launches** (exam math and programming), take only the tasks **without padding**. What is total actual ÷ total estimate? Round to two decimals. **Sofia:** 1.34. So most likely estimates in launches like ours ran about a third over. **Alex Tutor:** Vira insisted on dividing the totals. Was she right? > **Exercise:** **True or false:** to get the overrun of past launches, dividing total actual days by total estimated days is better than averaging each task's actual ÷ estimate. **Danylo:** And the padded rows? Two of them are mine. **Alex Tutor:** Count how many finished below their estimate, then compare their totals. **Vira:** If padding worked, at least some of them should have finished early. > **Exercise:** What do the six padded rows show? ## The chain from the inside **Danylo:** Our own owners' numbers add up to 54 days. That would put opening in mid-February. **Alex Tutor:** And the most likely values add up to 32, leaving an 11-day buffer. Now the honest middle: the expected value of each task, then the chain. Remember to add the variances. **Sofia:** (O + 4M + P) ÷ 6 for each row, then add them. **Alex Tutor:** Yes. Then compare with the 43 days you have. > **Exercise:** What is the **expected value** of the whole chain, in working days? **Sofia:** So in expectation three days of the buffer are gone before anything goes wrong. **Alex Tutor:** That is normal: most likely values leave out the long bad side. The question is how far it can stretch. **Danylo:** The demo lessons have the widest range, 8 to 20 days. That is where I expect trouble. **Alex Tutor:** Now the spread of the chain. > **Exercise:** What is the **standard deviation** of the whole chain, in working days? Round to two decimals. ## The chain from the outside **Sofia:** From the inside it looks comfortable: 35 expected, 41 at the high end, 43 available. **Alex Tutor:** Apply the history to the chain yourself first. > **Exercise:** The chain's most likely values add up to 32 working days. Launches like ours ran 1.34 times their most likely estimates. What does the outside view give for the chain, in working days? Two decimals. **Alex Tutor:** Now apply the history. The most likely values were 32 days; similar launches ran 1.34 times their most likely estimates. 32 × 1.34 ≈ 42.9 days. **Vira:** And my review: reviews ran 1.69 times their estimates. Six days becomes about ten. **Danylo:** So the outside view eats almost the whole buffer. **Alex Tutor:** The two views disagree, and the honest answer holds both. The date is achievable, but there is no room for anything new in the chain, and the review is the place to watch first. > **Exercise:** Leo, the sponsor, asks whether he can promise a 1 February opening; 43 working days remain. The inside view (three-point estimates) gives the chain 35 days expected, about 41 at two standard deviations. The outside view (32 most likely days × 1.34, the overrun of similar launches) gives about 42.9 days, and the review, at the 1.69 that past reviews ran, may take about 10 days. Given both views, what is the best answer to Leo? **Alex Tutor:** Last: what should make you go back to Leo early? > **Exercise:** You tell Leo, the sponsor, that the 1 February date is achievable but tight. Which signals should you commit to watching, so he hears early if it changes? Choose all that apply. ## Your turn: the answer to Leo **Leo:** Give me the answer in writing. If I put the date on the landing page, I want to know what could break it. **Hanna:** And if the buffer runs out, say who decides what happens then. **Alex Tutor:** One short message: the date and its confidence, the numbers behind it from both views, the buffer and who holds it, and the signals you will watch with dates. **Sofia:** Numbers from both views first, then the conditions and the checkpoint. > **Exercise:** Write Leo a message on whether he can promise 1 February, and how sure you are. > **Key points** > > - Start from a reference class that looks like your project, and divide total actuals by total estimates. > - Padded tasks in the history never finished early: padding moves the date without protecting it. > - Inside view: expected values add up, variances add up; the chain is 35 ± 3.07 days. > - Outside view: the subject launches' overrun (×1.34) takes the chain to about 42.9 of 43 days. > - Say the date with its confidence, the buffer and its owner, and the signal that would change your answer. **Alex Tutor:** The plan now has dates and money people can trust. Next module: the people around it, from Leo's shifting priorities to the stakeholder nobody has talked to yet. - Previous lesson: [A budget over time with a reserve](https://productovi.com/courses/project-management-without-bureaucracy/budget-over-time.md) - Next lesson: [Roles: sponsor, lead, expert, one owner per task](https://productovi.com/courses/project-management-without-bureaucracy/roles-and-one-owner.md) --- Productovi Learning. Free, no sign-up needed to read.