Chapter 38: Analyze Performance and Forecast the Future

The city wants one date

It is month twenty-five of the corridor, three weeks before the phased opening of the central and northern segments, and the chair of the city’s transport committee has asked the question every forecasting room eventually hears. She does not ask it as a question. She asks it as a demand, the way committees ask when they have a public announcement to make and a mayor’s month-twenty-four promise that is already behind them.

“When does the eastern segment open?”

Daniel Osei, the consortium’s finance director, has spent the morning building the answer, and the answer is not what the chair wants. He has the earned-value worksheet, the milestone trend chart, the commissioning flow numbers, the risk register’s eastern-segment row, and the forecast page that the monthly control cycle of chapter 31 recomputed the night before. He has learned, in the two years of this corridor, to bring the range and to say it plainly, because the range is the only honest shape a forecast has. The chair is about to discover why: the corridor’s three instruments are telling three different stories about the same future.

“Month twenty-seven in the base case,” he says. “Month twenty-eight if the wet-season resumption slips. The register carries the resumption as the swing factor. We can give you the date, and the date comes with its range.”

The contractor’s project director, who has been quiet, does not like the range. The corridor is ninety-two percent complete by the contractor’s counting rule, the number the mayor’s office has been quoting since month twenty-four, and ninety-two percent sounds like one date, not a range. “Ninety-two percent complete,” he says. “The eastern works are on the contractor’s plan. The opening is a calendar event, not a weather event.”

Lena Voss, the project director, has heard this sentence before, in the month when the seam package was eighty percent physically complete and forty-five percent earned, and she knows the sentence’s shape: the past reported as if it were the future. She does not argue with the contractor’s number, because the contractor’s number is true under its own counting rule, the rule that counts installation rather than accepted evidence, and the argument is not about which number is true. The argument is about which number answers the chair’s question. The chair did not ask what has happened. The chair asked what will happen.

The room has three instruments for that question, and the three instruments are disagreeing, which is why the chair’s demand is the right opening for this chapter. The seam package’s schedule index, the earned value over the planned value, reads about 0.69, which by the arithmetic says the corridor is behind schedule on that package, earning value at about sixty-nine percent of the planned pace. The milestone trend for the eastern segment, the chart that tracks each milestone’s forecast date across the monthly reviews, has held at month twenty-seven for two consecutive reviews, which says the forecast has stabilized, the drift stopped, the recovery held. And the commissioning flow, the team’s own throughput in the weeks before opening, has been clearing the commissioning backlog at nearly twice the rate of two months ago, which says the work is accelerating into the finish. Three instruments, three readings, one question: when will it be done?

All three are true. None of the three agrees. And the discipline of this chapter is the discipline of reading them together, because the single-number answer the chair wants is the answer every instrument will produce if it is read alone, and every single-number answer will be wrong in a different way. The schedule index alone says the corridor is late. The milestone trend alone says the corridor is holding. The flow alone says the corridor is accelerating. The mastery is not choosing the instrument that flatters. The mastery is knowing what each instrument measures, why they disagree, and what the disagreement means before anyone opens the meeting with a date.

Lena closes the item the way she has learned to close every forecasting item this year, with the sentence that separates the instruments into their proper roles. “Ninety-two percent complete is the rear-view mirror,” she says. “It tells the city where we have been. The question from the chair is about the windshield. The windshield is a range, it has assumptions, and it moves as the evidence moves. We are going to spend this meeting learning to read the windshield, because the corridor’s opening does not depend on what has happened. It depends on what will happen, and what will happen is a distribution, not a date.”

The chair does not like the word distribution. The chair will leave with a range, an assumption, and a trigger, which is the honest version of the distribution, and the chapter that follows is the machinery between the chair’s question and that answer.

The rear-view mirror and the windshield

The vocabulary comes first, because the room’s disagreement is a vocabulary failure wearing a numbers coat, and the five words the chapter needs are the words chapter 31 and chapter 15 built but never assembled into one sentence. The baseline is the authorized model of the work: the scope, the schedule, the budget, the measurement rules, the promise with a process that chapter 31 defended, the thing that moves only through authorized change. The actual is the recorded evidence: the money spent, the work accepted, the milestones hit, the counters read, the fact on the ground that no one can reclassify. The variance is the difference between them, the gap between what the model promised and what the evidence records, favorable or adverse, schedule or cost, and always a statement about the past. The trend is the variance read across time, the index moving from review to review, the slip chart, the flow, the pattern that a single month’s variance cannot show. And the forecast is the evidence-based expectation of the future, the current best statement of where the project will land, built from the trends and the assumptions, revised on a cadence, never a promise wearing a number.

The discipline that separates the five is the discipline of the rear-view mirror and the windshield, the chapter’s primary visual, and it is worth drawing at BlueLine because the drawing is the argument.

Figure 38.1: From the rear-view mirror to the windshield. The
analysis moves from what happened to what will happen, and each
stage carries its own instrument and its own question.

   REAR-VIEW MIRROR (what happened)        WINDSHIELD (what will happen)

   ACTUALS           VARIANCE             TREND              FORECAST
   recorded          baseline             variance over      scenarios,
   evidence:         versus actual,       time: the index    base, optimistic,
   spend,            the monthly          moving across      pessimistic, each
   completions,      report that          reviews, the       with its named
   milestones        explains the         milestone slip     assumptions, its
   hit, the          past, the "why"      chart, the flow,   trigger, and the
   counters read     of the month         the pattern the    decision it would
        |                |                single month       require, and the
        v                v                cannot show        range with its
   recorded on       discussed in         read as the        percentile and
   evidence,         the review,          shape of the       its confidence,
   owned, dated,     then closed,         future, checked    the completion-
   never adjusted    because the          against the        date range and
   after the fact    past cannot be       baseline change    the reserve,
                     changed             record and the     revised every
                                          measurement rule   cycle, decided
                                                             by the people
                                                             who own the risk

The rear-view mirror is not useless. It is how the driver knows where the car has been, whether the road behind was clear, whether the lane was safe to occupy. Variance reporting does the same work: it tells the project what happened, why it happened, what the month’s numbers mean against the baseline, and a project that does not explain its variances is a project that is not learning. The failure is not the mirror. The failure is the report that ends at the mirror: the review that spends the hour explaining February and five minutes on the eastern segment’s forecast, the steering meeting that closes with the variance explained and no decision taken, the dashboard that chapter 37 built so carefully and that this chapter must now point forward.

The windshield is where the disagreement lives, and that is why the chair’s demand matters. A forecast is a statement about the future, which is a statement about uncertainty, which is a statement that can be wrong, and the demand for a single date is usually the demand to hide the uncertainty, not to resolve it. Chapter 15 taught the corridor this lesson in the estimate: the honest number is a range with assumptions, and the precise number is the precise liar. Chapter 38 takes the same discipline and moves it from the estimate to the control cycle, from the number the project planned to the number the project is becoming, and it adds the instruments that turn the rear-view mirror into a windshield: the earned-value bridge, the milestone trend, the flow percentiles, the scenario set, and the simulation that tests them all.

One more discipline belongs in the vocabulary, and it is the discipline that keeps the mirror and the windshield separate on the page. The variance is explained. The trend is read. The forecast is decided. The three verbs are different because the three instruments have different owners and different clocks. The variance belongs to the month that produced it, and its explanation is a learning act, closed when the root cause is named. The trend belongs to the review cadence, and its reading is an interpretive act, closed when the pattern is named and checked against the baseline-change record and the measurement rules of chapter 37. The forecast belongs to the decision, and its revision is an act of judgment, closed when someone decides what to do about it. The project that explains its variances and reads its trends but never decides on its forecast is the project that has built a mirror and no windshield, which is the state the BlueLine review is in when the chair asks for the date.

The earned-value bridge

The earned value system is the bridge between the mirror and the windshield, because it takes the variance, a statement about the past, and turns it into the index, a number that can be extended toward the future, and chapter 31 built the minimum working form while chapter 38 builds the full system. The three primitives were established at the seam package and are worth restating with their full names before the arithmetic, because the full system is the primitives plus the questions they answer. The planned value, PV, is the budgeted cost of the work the plan said should be done by the review date. The earned value, EV, is the budgeted cost of the work the evidence says was accepted, progress earned on evidence rather than claimed on effort, the counting rules of chapter 31 applied. The actual cost, AC, is the money actually spent, from the ledgers of chapter 18. Everything else in the earned-value vocabulary is a combination of these three, and the discipline is knowing what each combination means and, more importantly, what it assumes.

The first combinations are the variances. The schedule variance, SV, is earned value minus planned value, EV minus PV, and it answers the question “has the project earned as much value as the plan scheduled?” A negative schedule variance means the earned progress is behind the planned progress, and it is measured in the baseline’s own currency, which is the source of its honesty and its limitation: it says how far behind, in the plan’s money, not in the calendar. The cost variance, CV, is earned value minus actual cost, EV minus AC, and it answers “has the project spent more or less than the value it earned?” A negative cost variance means the project is spending ahead of its earned value, and it is the number that separates “under budget” from “spending less than the plan assumed,” a distinction chapter 18 taught with the four ledgers and the corridor’s favorable incurred variance that was really the eastern delay.

The second combinations are the indices, and they are the bridge to the forecast. The schedule performance index, SPI, is earned value divided by planned value, EV over PV, and it reads as the pace of earning relative to the plan: an index of 1.00 means the project is earning exactly as planned, 0.90 means it is earning at nine-tenths of the planned pace, and 1.10 means it is ahead. The cost performance index, CPI, is earned value divided by actual cost, EV over AC, and it reads as the efficiency of spending: an index of 1.00 means every unit spent earned a unit of value, 0.90 means ninety percent efficiency, and a CPI below 1.00 is the classic warning that cost is running ahead of value. Both indices are point-in-time readings, snapshots of the earned-value system at one review, and the discipline of chapter 31 applies to both: they are signals to examine, not verdicts to obey, and the examination is the question “what is this index made of, and does its composition make the extrapolation honest?”

The seam package at month twenty-four is the worked example, and its arithmetic is the arithmetic this chapter builds on, because the package’s numbers are already true in the record and they carry the chapter’s whole lesson in one worksheet. The package’s budget at completion, BAC, is 40 million units, the target cost of the shared-savings contract. At the month-twenty-four review, the plan had scheduled the dry run complete, so the planned value is 65 percent of 40 million, 26 million units. The evidence had accepted three of five milestones, so the earned value is 45 percent, 18 million units. And the actual cost is 36 million units, because the package is procurement-heavy, the hardware arrived early, and the failed dry run consumed an expensive week. The worksheet:

The variances and the indices follow from the primitives by arithmetic: SV is 18 minus 26, minus 8 million units; CV is 18 minus 36, minus 18 million; SPI is 18 over 26, about 0.69; CPI is 18 over 36, 0.50. A cost index of 0.50 is normally the smell of serious trouble, and the discipline is reading it correctly. The index is a signal that the package’s cost is running far ahead of its earned value, and the examination shows why: the procurement profile spends early, the hardware arrived before it was earned, and the failed dry run spent an entire expensive week without earning a milestone. The composition of the index matters because the composition decides the forecast. An index depressed by a past, one-time shock is a different instrument from an index depressed by a systemic inefficiency, and the two demand different forecasts, which is where the estimate-at-completion forms come in.

Chapter 18 introduced the three estimate-at-completion forms on the systems package, the rate-based, the schedule-inclusive, and the bottom-up, and chapter 38 deepens them with the question that makes them usable: every form is an assumption about what persists, and the choice of form is the choice of which assumption the evidence supports. The rate-based form, EAC equal to BAC divided by CPI, assumes the current cost efficiency continues into the remaining work, and on the seam package it produces 40 divided by 0.50, 80 million units, the naive estimate that chapter 31 examined and rejected. The rest-at-budget form, EAC equal to AC plus the remaining value at budget rates, AC plus (BAC minus EV), assumes the remaining work is done at the planned efficiency, and it produces 36 plus 22, 58 million units. The schedule-inclusive form, EAC equal to AC plus the remaining value over the product of the two indices, assumes both the cost performance and the schedule performance persist, and it produces 36 plus 22 over about 0.346, about 100 million units, the harshest of the four because it prices the delay as well as the inefficiency, the compression, the overtime, the rework of chapter 17. And the bottom-up form, EAC equal to AC plus the re-estimated cost of the remaining work, produces 36 plus 6, 42 million units, inside the contract’s ceiling of 48.

The four forms disagree by nearly 60 million units on a 40 million-unit package, and the disagreement is the teaching, exactly as it was in chapter 18. The estimate at completion is not a number. It is a set of assumptions about what persists, and the question to ask of any forecast is not “what is the number” but “what does this number assume persists, and do we believe that?” On the seam package the answer is that the rate-based forms assume the failed dry run is the new normal, that the package will keep earning half a unit of value per unit spent and keep slipping at 69 percent of its planned pace. The evidence says the opposite: the shock was a past event, the repair is estimable at 6 million units, and the remaining work, the re-run and the end-to-end acceptance, is a different shape from the work that produced the index. The bottom-up forecast of 42 is the honest forecast, not because it is the smallest, but because its assumption, that the remaining work is estimable and the shock is past, is the assumption the evidence supports, and because the price flows through the shared-savings formula that chapter 31 priced at 43.6 million against the at-target 43.2, an increase of 0.4 million, the project’s 20 percent share of the overrun, with the consortium absorbing the remaining 1.6 million in fee. The forecast at completion, in the vocabulary of chapter 15, is the current evidence-based expectation, and the evidence says the shock is priced and the package is contained.

The to-complete performance index, TCPI, is the last instrument of the earned-value bridge, and it answers the question the forecast opens rather than the question the forecast answers: can the remaining work still be done within the remaining budget? Its form is the remaining value over the remaining funds, (BAC minus EV) over (BAC minus AC), and on the seam package at month twenty-four it is 22 million units of remaining value over 4 million units of remaining budget, 5.5 units of value to be earned for every unit spent. A TCPI above 1.00 says the remaining work must be done more efficiently than the project’s plan assumed; a TCPI of 5.5 is not a number the team can simply work harder to meet, and the discipline is reading it as what it is: a statement that the target cost cannot be met within the remaining funds, which is a decision about funding and scope and commercial structure, not a math error and not a call for effort. At the seam package the decision is already built: the contract has a ceiling of 48 million units, the shared-savings formula allocates the overrun, and the honest forecast of 42 sits inside the ceiling, so the TCPI against the ceiling is the number that matters, 22 over 12, about 1.8, demanding efficiency but not impossibility. The corridor-level numbers carry the same discipline one level up: the authorized envelope of 2,540 million units, the capital 2,400 plus the mitigation 140, against the forecast at completion of about 2,551 million, a variance at completion, VAC, of about minus 11 million, inside the reserve band that chapter 15 built and chapter 18 described, with the 80-percent range on the remaining work running to about 2,630, still inside the total authorized limit of 2,720. The corridor does not compute its EAC from its CPI, because its CPI would carry the lumpy procurement and the seam shock of the past; it computes its EAC bottom-up, package by package, on the evidence of the remaining work, and the reserve is the arithmetic of the range.

The earned-value bridge, in one sentence, is the discipline of reading the index as a question about the composition of the past and the persistence of its causes, and building the forecast from the assumption the evidence supports rather than the assumption the formula makes. The failure is the formula worship that chapter 31 warned against, the index extrapolated without examination, the EAC printed because the spreadsheet computed it, the report that says “CPI 0.50, therefore EAC 80 million,” with no one asking whether the failed dry run is the new normal. The field signal of the formula worshipper is the forecast that changes only when the index changes, and the field signal of the honest bridge is the forecast that changes when the assumptions change, which is the difference between arithmetic and analysis.

The milestone trend is the trajectory

The earned-value indices are snapshots, and a snapshot cannot show a trajectory, which is why the second instrument of the windshield is the milestone trend, the chart that tracks each milestone’s forecast date across the reviews, and why the corridor’s eastern segment is its worked example. The milestone dictionary of chapter 16 defined the milestones as evidence, named events with weights and acceptance criteria; the milestone trend takes the dictionary and draws the forecast date of each milestone at every review, and the drawing is the schedule’s trajectory. Each review plots the milestone’s planned date, its baseline position, and its current forecast date, and the line between the two is the drift, the number of months the milestone has slipped, held, or recovered. The chart is the schedule index’s antidote, because the index says “the project is behind” while the trend says “the project is no longer slipping,” and both are true, the index about the accumulated position and the trend about the direction.

The eastern segment’s commissioning-complete milestone is the example, and its numbers are teaching constructions built from the facts chapter 31 established, the resequenced plan, the wet-season break, the variation of 38 million units sitting in the baseline, the resumption with the contractors’ two-week mobilization, and the forecast holding at month twenty-seven with the risk of month twenty-eight. The milestone’s baseline date, after the variation was authorized, is month twenty-seven. At the month-twenty-two review, the forecast date was month twenty-nine, two months of drift, the wet season and the flood-plain conditions pushing the resumption out. At month twenty-three, the forecast date had recovered to month twenty-eight, one month of drift, the resumption dates firmed on the network’s logic. At month twenty-four, the forecast date reached month twenty-seven, on baseline, the re-sequencing working, the look-ahead carrying the resumption with its owners and its hold points. And at month twenty-five, the review this chapter lives in, the forecast date holds at month twenty-seven for the second consecutive review, on baseline, with the register carrying the risk that a slip in the resumption moves the date to month twenty-eight.

Figure 38.2: The eastern segment's commissioning-complete milestone
trend across four monthly reviews. The forecast date drifts, then
recovers, then holds; the width of the range is the schedule
confidence, and the trigger is the resumption's two-week slip.

   review     baseline     forecast     drift
   month      date         date
   22         month 27     month 29     +2 months
   23         month 27     month 28     +1
   24         month 27     month 27     on baseline
   25         month 27     month 27     on baseline, holding

The reading of the trend is the chapter’s lesson in one chart. The drift from month 22 to month 23, the two-month and then one-month slips, was the wet season doing what wet seasons do, and the forecast moved because the evidence moved, which is the discipline of chapter 31’s reforecast, the forecast recomputed every cycle while the baseline holds. The recovery at month 24 was the re-sequencing proving itself, the variation absorbed into the baseline through the change process, the look-ahead converting the network into owners and dates. And the hold at month 25 is the signal that matters most for the chair’s question: the forecast date has been stable for two reviews, which is what schedule confidence looks like in a trend. Schedule confidence is not the absence of a range. It is the range’s width and its stability: a range that tightens review after review is a forecast converging on evidence; a range that stays wide is an honest statement that the uncertainty is real; and a range that is a single date, printed without width, is not confidence, it is a claim. The eastern segment’s honest shape is the sentence Daniel Osei brought to the room: month twenty-seven in the base case, month twenty-eight if the resumption slips, with the trigger named, the resumption’s two-week slip, and the register row carrying the owner and the decision date.

The milestone trend has its gaming risks, and they are the risks chapter 31 named for milestones as evidence, wearing a chart’s clothes. The milestone that is renamed when it slips, the dictionary entry redefined to match the achieved date, the “substantially complete” that absorbs the final hold point, the trend that looks flat because the milestones were moved rather than the work finished, these are the gaming moves that make the slip chart lie, and the field signal is the trend that never shows a slip while the index and the flow tell other stories. The discipline that keeps the trend honest is the one the measurement chapter built: the milestone dictionary is fixed by the baseline, the acceptance evidence is independent of the performer, the counting rule is written before the milestone is claimed, and the baseline-change record says when the dictionary moved, by whom, and what it was before and after. A trend chart without the dictionary behind it is a drawing; with the dictionary, it is evidence.

The milestone trend is the schedule’s windshield in predictive work, and it answers the question the SPI cannot: not “how far behind is the project” but “is the project still falling behind?” The SPI says the seam package is at 0.69, a snapshot of the accumulated position. The trend says the corridor’s milestones have stopped slipping, a statement about the direction. The two are not in conflict; they are in sequence, and the forecast is built from the sequence, the position corrected by the direction, the accumulated variance examined and the trajectory read, which is the reading the room must do before the chair’s date can be answered, and the reading that the third instrument, the flow, will complicate and complete.

The flow speaks in percentiles

The third instrument of the windshield belongs to the adaptive clock, and it is the instrument that answers the same question, when will it be done, from a different model of time, the flow system of chapter 17 and the empirical loop of chapter 32. KijaniPay’s delivery system in July is the worked example, and its numbers are the numbers chapter 32 established and this chapter extends, because the platform’s forecast is the flow forecast, and the flow forecast is a probability statement, not a date.

The flow instruments were built in chapter 17 and run in chapters 30 and 32: the work in progress, the WIP limit of six holding; the cycle time, the calendar time from start to done, running under five days, about 4.7, the Little’s Law arithmetic that chapter 30 worked and this chapter inherits; and the throughput, the completions per week, running about nine with the observed range seven to eleven. The backlog, after the June rewrite that parked twenty items in discovery, holds about twenty-seven actionable items, and the forecast follows from the flow: at the middle rate, twenty-seven divided by nine, about three weeks to clear the actionable queue; across the observed range, twenty-seven divided by eleven to twenty-seven divided by seven, about two and a half to four weeks, a range of about a week and a half. The sentence to the board is the chapter 15 discipline in one breath: at the current flow, the actionable backlog clears in about three weeks, with the range two and a half to four, and the number moves as the flow moves, because the forecast is a model, not a promise.

The flow forecast has a shape, and the shape is what the percentiles carry. The observed completions across the recent weeks, read as a distribution rather than an average, give the forecast its probability: the middle of the distribution, the median, says three weeks; the 85th percentile, the number the project wants to be wrong about only once in seven times, says about four; and the pessimistic tail, the sequence of bad weeks that the system’s variability can produce, runs to five or six. The percentile is a deliberate choice, and the choice belongs to the decision: a private pilot can live with the median, a public launch with the 85th, a regulatory deadline with the tail, the discipline chapter 17 taught and chapter 32 ran. The forecast that reports a single week number without a percentile is the forecast that has chosen a percentile silently, usually the median, and a silent median is a quiet bet that the project will be average, which is a bet the project should make aloud or not make at all.

The flow’s visual instrument is the burnup chart, and it is the windshield on the adaptive clock because it draws the two lines the date depends on: the scope line, the total amount of work the backlog represents, and the completion line, the amount finished, with the gap between them the remaining horizon and the forecast the rate at which the completion line crosses the scope line. The burnup is the honest sibling of the burndown: the burndown draws one line, the remaining work, and hides the scope growth, the new items that arrive and stretch the top of the chart, the ready column that grew to forty-seven in June before the rewrite parked twenty; the burnup draws the scope line as well, so that scope growth is visible as the top line rising, and a slipping date is explained by the top line moving, not by the team slowing. The field signal of the burndown-only report is the release date that slips while the burndown line looks healthy; the repair is the burnup, the two lines drawn from the same data, the scope changes logged through the same change discipline that protects the baseline in predictive work.

The second forecast on the adaptive clock is the one the product goal needs, and it is the benefit forecast, the estimate-to-target distance on the benefit clock that chapter 32 introduced, the number that says whether the work is working rather than when the work will be done. The June reading was established in chapter 32: 155 applications against the linear path of about 233 at month six, the target 350 by September, about 25 percent of the eligible cohort of about 1,400 merchants, the gap about 78, the required run-rate about 65 applications a month against the current 43. July is the next reading, and its numbers are teaching constructions that extend the June facts. The cumulative applications reach about 210, about 15 percent of the cohort, against the linear path at month seven of about 272, 350 times seven over nine. The gap to the target is 350 minus 210, 140 applications, over the two months remaining, August and September, about 70 a month. The current run-rate, the July increment over June, is about 55 a month, the friction fixes from the April cohort’s abandonment data and the settlement-visibility prototype beginning to move the number, but not enough. At the current run-rate, the two months produce about 110 more applications, landing near 320, about 30 short of the 350 target, about two percent of the eligible cohort, and the forecast is the sentence that names the gap: at the current flow, the benefit lands about 30 applications short of the target, and the decisions that close the gap are the quarter’s decisions, the friction fixes accelerated, the settlement-visibility work built for the September cohort, the run-rate lifted to about 70 a month by August, with a mid-August check on the reading. The benefit forecast is the windshield on the adaptive clock, and its instrument is the same as the flow forecast’s instrument, the range with the assumptions and the decisions attached, applied to the outcome rather than the output.

The two forecasts, the flow forecast and the benefit forecast, are kept separate on the page, because they answer different questions and they can disagree, and the disagreement is information. The flow forecast says when the work will be done. The benefit forecast says whether the work is working. A project that reports only the first is the project that ships the wrong work on time, the KijaniPay June wall in two colors, the delivery dashboard green and the benefit clock amber, which is why the July review carries both pages and why the board’s demand for a date is answered with the range on the flow page and the gap on the benefit page, and the decision on both: the quarter’s backlog is ordered to close the benefit gap, and the release dates flex inside the flow range. That is the adaptive windshield, and it is the same windshield the corridor is looking through, built from different instruments, reading the same future.

Scenarios are the honest shape of the future

The forecast that the review produces is a single number on a page, and the honest forecast is a set of scenarios, because the future is not a point and the forecast’s job is to make the future’s shape visible before it arrives. The minimum viable set is three: the base case, the pessimistic case, and the optimistic case, each with its named assumptions, its triggers, and the decision it would require. The base case is the current evidence-based expectation, the forecast of chapter 15’s vocabulary, built from the trends and the bottom-up estimates, the number the project would defend in a quiet month. The pessimistic case is the tail that the register and the pre-mortem have already identified, the wet-season resumption slipping, the re-run failing again, the milestone the trend has been watching, and its trigger is the earliest evidence that the tail is arriving, the two-week slip, the failed rehearsal, the approval condition that lands late. The optimistic case is the headroom the evidence supports, the acceleration holding, the commissioning backlog clearing at the fast rate, the resumption landing on its first dates, and its trigger is the earliest evidence that the headroom is real.

The corridor’s completion-date range is the scenario set in public form, and it is the answer the chair is owed: the central and northern segments open at month twenty-five, the phased opening on the earned evidence of chapter 31, and the eastern segment opens in month twenty-seven in the base case, month twenty-eight if the resumption slips, the optimistic case absorbed into the base because the base already carries the good news the evidence supports. The range carries its arithmetic with it — the reserve band, the 80-percent range on the remaining work running to about 2,630 inside the total authorized limit of 2,720, the forecast at 2,551 inside the reserve, the cash curve peaking at about 400 in the phased-opening month — the numbers chapters 18 and 31 established, now assembled into the shape of the future rather than the record of the past.

The scenario discipline is the discipline of chapter 23’s pre-mortem and contingency playbook, applied to the forecast’s own instruments, and it has three rules. First, the scenarios are built from the project’s own evidence, the risk register, the assumption inventory, the reference classes, the trend charts, not from the market’s mood or the committee’s hope, and a scenario that cannot name its evidence is a hope wearing a forecast’s clothes. Second, every scenario names its trigger, the earliest observable evidence that the world is moving toward that scenario, the resumption’s two-week slip, the application run-rate failing to lift by mid-August, the milestone slipping a second consecutive review, because a scenario without a trigger is a paragraph, and a scenario with a trigger is a decision waiting on evidence. Third, the scenario set is revised on the forecast’s cadence, the monthly control cycle in predictive work, the iteration review in adaptive work, and it is never revised to comfort a stakeholder, the frozen forecast disease of chapter 31 wearing a scenario chart’s clothes. The scenario set is the windshield’s honest shape because it says what the future would have to be for the project’s date to change, and the statement of what would have to be true is the most powerful sentence a project can produce, because it turns the meeting from an argument about the date into an argument about the evidence.

The hybrid register translates the scenario discipline across the seam, and Meridian’s clinic program is the example, the construction network’s gate dates and the training waves’ flow forecast speaking to each other through the integration milestones of chapter 33. The network’s scenario set carries the clinic certificate dates, the regulatory evidence, the physical sequence, with its triggers, the certificate delays and the construction hold points. The flow system’s set carries the training-wave completions, the clinical super-user availability, the adoption measures of chapter 11, with its triggers, the wave cycle times and the point-of-work readiness readings. The seam discipline is the translation: the network’s pessimistic certificate date becomes the flow system’s service target, and the flow system’s pessimistic wave forecast becomes the network’s evidence for the training gate, and the forecast reported to the steering committee is one page with both scenarios and the seam row that owns the translation, because the hybrid forecast that cannot speak across the seam is the hybrid failure of chapter 33 in forecast form, the construction green and the training amber, the clinic opening with nobody trained. The crisis register compresses the same discipline to its minimum, and Northstar’s daily huddle is the scenario set running in hours: the base case, the next day’s plan; the pessimistic case, the supplier failure and the route closure; the trigger, the morning report; and the decision, the afternoon re-plan, with the floors, the safeguarding, the cold chain, the sanctions, never moving, the same separation of the promise and the model that chapter 31 carried, at emergency speed.

The Monte Carlo is an arithmetic engine, not a fortune-teller

The scenario set is three views of the future chosen by judgment, and the Monte Carlo method is the machine that shows the future’s full distribution, and the discipline is knowing exactly what the machine does and exactly what it does not do. The method takes its name from the casino, and its origin is documented: Stanislaw Ulam and Nicholas Metropolis, working at Los Alamos in the aftermath of the Second World War, published “The Monte Carlo Method” in 1949 in the Journal of the American Statistical Association, describing the technique of solving problems by constructing a stochastic process that follows the problem’s logic and observing the process’s outcomes through repeated random sampling. The idea had served the war’s neutron-diffusion physics, the calculations no one could do by hand; the 1949 paper gave the method its name and its shape. In project forecasting the shape is the same: the model of the project, the schedule network, the cost structure, the backlog, is run many thousands of times, and each run draws its inputs, the durations, the costs, the throughputs, from the distributions the team provided, and the collection of outcomes, the thousands of completion dates and final costs, is the forecast’s true shape.

The intuition the method provides is worth more than the number it produces, and the corridor’s example carries it. Take a small commissioning sequence of three chains, the kind the eastern segment’s remaining work resembles, each chain estimated three-point in the manner of chapter 15: the first, the resumption and the slab, optimistic 6 weeks, most likely 8, pessimistic 12; the second, the seam re-integration, optimistic 5, most likely 7, pessimistic 10; the third, the operator certification, optimistic 4, most likely 6, pessimistic 9. The deterministic plan sums the most likely values: 8 plus 7 plus 6, 21 weeks. The Monte Carlo draws each chain’s duration from its three-point distribution, sums the three, and repeats, and the distribution of the summed outcomes has three properties the deterministic sum cannot show. The median of the summed distribution sits above the sum of the most likely values, because the distributions are right-skewed, the pessimistic tails longer than the optimistic ones, and a sum of skewed distributions inherits the skew; the deterministic 21 weeks is not the 50th percentile, it is the mode of a fiction. The 85th percentile, the number a public commitment would need, sits noticeably higher, about 25 weeks, and the gap between the median and the 85th is the schedule risk in one number, the width of the honest range. And the distribution’s tail is fatter than the independent-sum model suggests when the chains share a cause, the wet season that delays the slab, the re-integration, and the certification together, because correlated inputs widen the distribution beyond the independent case, and the model that treats the three as independent understates the risk of the bad season. The intuition, in one sentence: the sum of the most likely values is not the most likely sum, the tails are longer than the middle suggests, and the shared cause is the tail’s real engine.

The discipline of reading the Monte Carlo is the discipline of refusing its false sophistication, and the rules are the rules the book has built for every instrument. The output is only as honest as its inputs: the simulation draws from the distributions the team provided, and if the distributions came from optimism, the reference-class-blind inside view of chapter 15, the simulation is a precise computation of an imprecise guess, the garbage-in-garbage-out warning wearing a 10,000-run suit. The output’s percentiles are properties of the model and its assumptions, not promises about the world: the statement “85 percent confident by month twenty-eight” is true only if the model’s logic matches the work and the input distributions match the evidence, and the calibration check is the one chapter 15 built, the reference class, the distribution of how comparable work actually went, the outside view the inside view cannot produce. The simulation’s date distribution is not a schedule: the model is only as good as the decomposition of chapter 14 and the logic of chapter 17, and a Monte Carlo run on a network with the wrong dependencies, the load profile that chapter 31’s dry run exposed, is a confident computation of the wrong future. And the output is not the decision: the P50, the P85, and the tail are the raw material for the scenario set and the reserve band, and the decision, which percentile to carry, which reserve to hold, which trigger to set, belongs to the people who own the risk, the discipline chapter 22 built and chapter 23 ran. On the flow clock the same engine adds what the division cannot: the frequency of the tail, the share of simulated worlds in which the backlog takes four weeks or more, which is the probability the project should report instead of the date it should promise. The Monte Carlo is an arithmetic engine, and the project is the driver, and the driver who blames the engine for the destination is the driver the engine cannot save.

Read the trio before you act

The three instruments are now on the table at BlueLine: the earned-value indices, the seam package’s 0.69 and 0.50; the milestone trend, the eastern segment holding at month twenty-seven for two reviews; and the flow, the commissioning backlog clearing at nearly twice the rate of two months ago. The chair wants one date. The room has three readings. And the discipline that answers the chair is the discipline of reading the trio before acting, because the trio is disagreeing on purpose, and the disagreement is the diagnosis.

The first reading is the composition of the indices. The seam package’s SPI of 0.69 and CPI of 0.50 are point-in-time snapshots of a package whose past contains a one-time shock, the procurement-heavy early spend and the failed dry run, and chapter 31 already established the reading: the indices are signals to examine, the shock is priced, the repair estimated, the bottom-up forecast of 42 inside the ceiling. The corridor’s cumulative index carries the same composition, and a rate-based extrapolation of its CPI would compound a past event into the future, the exact error the EAC forms were built to expose. The indices say the corridor is behind where it planned to be; the composition says the behind-ness is concentrated, priced, and past-shaped, which is a different statement from “the corridor is slipping.”

The second reading is the trajectory. The milestone trend says the forecast dates have stopped slipping: the eastern segment recovered from month twenty-nine to month twenty-eight to month twenty-seven and has held for two reviews, and the milestone dictionary behind the chart says the milestones are evidence, the acceptance criteria fixed, the hold points real. The trend is the direction, and a stable direction with a priced position is the forecast’s foundation, the accumulated variance examined and the trajectory read.

The third reading is the flow. The commissioning team’s throughput has accelerated into the finish, the backlog clearing at nearly twice the rate of two months ago, the near-term reality the schedule index cannot see and the milestone trend can only lag. And the opening’s operational readiness, the readiness measures of chapter 37, the drills, the competency checks, the counter calibrations, are the flow the forecast must not forget, because the corridor can be physically complete and operationally unready on the same day, and the forecast that counts only the construction misses the passenger.

The three readings disagree because they measure three different truths: the accumulated position, the direction, and the near-term reality. They are not in conflict. They are in sequence, and the forecast is built from the sequence: the position corrected by the direction, the direction confirmed by the flow, the whole checked against the scenario set and the reserve. The disagreement the room feels is not a contradiction to resolve by choosing the flattering instrument. It is the information the instruments were built to provide, the three views of the same future that together make the range, and the range is the answer the chair is owed.

The action that follows the reading is the discipline of interpreting signals before taking action, and its rules are the rules every instrument in this book carries. First, explain before you act: the index moved, the trend shifted, the flow changed, and the explanation, the cause, the composition, comes before the response, because the response to a priced shock, a reserve release, a re-sequencing, a change request, is different from the response to a systemic decline, and acting on the number without the explanation is acting on noise. Second, check the data before you explain: the chapter 37 discipline, the counting rule, the data quality, the calibration, the owner, the source, because an index computed from a physical claim is a claim wearing an index’s clothes, and the explanation of a wrong number is a wrong explanation. Third, name the signal’s persistence before you extrapolate: the one-month wobble is noise, the two-review hold is a signal, the three-month trend is a pattern, and the forecast’s confidence lives in the distinction, the same distinction that separates the seam’s one-time shock from the systemic inefficiency it could have been. And fourth, decide on the cadence, not on the demand: the chair wants a date today, and the honest answer is the range with the assumptions and the trigger, the forecast revised on the monthly cycle, the trigger watched on the evidence, the decision made when the trigger fires, because the forecast is a model, and the model is only as true as the cadence that revises it.

The month-twenty-five review resolves the way the instruments say it should, and the resolution is the chapter’s proof in the corridor’s own room. The committee accepts the range: the central and northern segments open at month twenty-five, on the earned evidence and the readiness measures of chapter 37, the phased opening the control cycle has been holding since the failed dry run; the eastern segment is forecast at month twenty-seven, with the month-twenty-eight risk carried in the register under the trigger, the resumption’s two-week slip, owned and dated; the forecast at completion holds at about 2,551 million units inside the reserve band, the VAC of about 11 million absorbed by the reserve, the cash peak of about 400 in the opening month carried by the funding staircase; and the sentence to the city is the sentence that names the range and the assumptions, the windshield in public form, the date with its width, the trigger with its evidence, the reserve with its arithmetic. The chair leaves with a range instead of a date, and the range is the better answer, because the date would have been a promise the evidence did not support, and the range is a promise the evidence does support, revised on a cadence, watched at the trigger, owned by the people who can move the number.

The forecast theater

Every practice in this book has its theater, the plausible behavior that makes the practice ineffective while keeping its appearance, and forecasting’s theater is the most expensive one, because it produces confidence about the future at exactly the moment the future is least known. The theater has six acts, and each one is a competent person doing a reasonable thing badly.

The single-number answer is the first act, and it is the act the chair almost induced: the demand for one date, met by the supply of one date, the median quoted as a commitment, the range deleted because the range is uncomfortable in a headline. The single-number answer feels decisive and is a lie, because the median is a bet that the project will be average, and the project that promises the median without saying it is a median has promised a coin flip. Its field signal is the announcement that quotes a date with no range, no assumption, and no trigger, and its cost is the credibility that pays when the coin lands the other way.

The frozen forecast is the second act, and it is the disease chapter 31 named, wearing a new coat: the forecast that never moves because moving it would require explaining it, the number held to match the baseline while the evidence accumulates against it, the report that says “forecast: unchanged” for six months while the index falls and the trend slips. Its field signal is the review that discusses the variance and never the forecast, the forecast column that repeats the baseline row, the date that is defended rather than revised.

The rear-view report is the third act, and it is the report this chapter opened against: the variance explained in exhaustive detail, the month’s story told completely, the future never mentioned, the meeting that closes with no forecast and no decision because the forecast was never on the agenda. It feels rigorous and is useless, because it answers the question the project has already lived and avoids the question it is still living, and its field signal is the meeting that is comfortable, the one where everyone agrees on the past and no one commits to the future.

The false precision is the fourth act, and it is the chapter 15 lesson in forecast form: the EAC reported to the last unit, 2,551.3 million, the completion date to the day, the 85th percentile quoted to two decimal places, the precision manufactured by the spreadsheet and mistaken for knowledge. Its field signal is the forecast whose digits outnumber its assumptions, the number more precise than the evidence that produced it.

The cherry-picked percentile is the fifth act, and it is the sandbag in probability’s clothes: the P50 reported as a commitment when the decision needs the P85, the P85 quoted to justify a date that was wanted anyway, the percentile chosen after the fact to defend the number rather than before the fact to inform the decision. Its field signal is the percentile that changes with the audience, the 85th for the sponsor, the 50th for the public, the tail for the budget request.

And the simulation worship is the sixth act, and it is the newest theater: the Monte Carlo run as a ritual, the 10,000 runs as a substitute for judgment, the output percentile quoted as proof, “the model says 85 percent,” with the inputs unexamined, the logic unverified, the reference class unconsulted, the machine’s fluency mistaken for the project’s knowledge. Its field signal is the room that defers to the model, the question “what does the model say” asked without the question “what does the model assume,” and the decision surrendered to an arithmetic engine that was built to serve it.

The theater’s common root is the same root every theater in this book shares: the instrument mistaken for the judgment. The forecast is not the number. It is the assumptions, the range, the triggers, and the decision, and the instrument that produces the number is a servant of the judgment that reads it, and the room that remembers the hierarchy, the judgment above the instrument, the assumptions above the arithmetic, the decision above the distribution, is the room that keeps its theater off the stage.

What the machine can run

The forecasting instruments are arithmetic, and arithmetic is exactly what the machine can hold, and the boundary of the machine’s role is the boundary the book has drawn since chapter 15, drawn again in chapter 31, and now drawn for the forecast: the machine can compute, draft, and simulate, and the assumptions, the interpretation, and the decision stay human. The machine can compute the earned-value worksheet from the provided records, the planned value, the earned value, the actual cost, the variances, the indices, the naive extrapolations, with the formulas and the inputs stated so the arithmetic can be checked. It can run the Monte Carlo from the provided input distributions and model logic, producing the distribution of completion dates and final costs with their percentiles, the output labeled as a computation of the model, not a prediction of the world. It can draw the milestone trend chart and the burnup chart from the provided data, flagging the slips, the holds, the scope movements, the aging items, the pattern made visible for the room to read. And it can draft the scenario paragraphs from the provided register and assumptions, each with its trigger, the raw material the room corrects, because the scenario’s truth is in the project’s evidence, not in the machine’s fluency.

The boundaries are the boundaries the book has built. The assumptions are not discoverable by the machine: which EAC form matches the future of the remaining work, whether the shock is past or persistent, which percentile the decision needs, these are judgments about the project’s composition and its appetite, and the machine’s proposed assumption is a hypothesis until the people who own the risk confirm it. The data is not self-grounding: the machine computes from what it is given, and a forecast computed from a physical claim, an uncalibrated counter, or a gamed metric is a precise computation of a wrong input, which is why the chapter 37 discipline, the counting rules, the data-quality log, the owners, precedes the chapter 38 machinery. The source data is the approved, non-confidential project records, redacted before prompting, and no confidential, personal, regulated, or contract-restricted information enters an unapproved system, the standard the book has held since chapter 15. The verification is the room: the outputs are drafts until a named owner verifies the inputs against the records, the formulas against the arithmetic, the assumptions against the evidence, and the audit record says what was generated, from what, checked by whom, and decided by whom, the same record the control cycle has carried since chapter 31. And the decision is not delegable: the machine can produce the distribution, and the people decide the percentile, the reserve, the trigger, the range, and the date, because the decision is an acceptance of obligation under the governance of chapter 8, and no simulation is an obligation. The machine accelerates the arithmetic, and the windshield remains human, which is the only division of labor that keeps the forecast honest, because the future is the one instrument the machine cannot be given, and the judgment that reads the future is the judgment this book exists to build.

Practice

One. A quick check: name the instrument. For each statement from a project report, name the instrument it belongs to, baseline, actual, variance, trend, or forecast, and say what decision it supports. (a) “The eastern culvert works earned 34 percent of their value this month against the 40 percent the plan scheduled.” (b) “The seam repair finished in the sixth week, one week past the five-week estimate.” (c) “The commissioning backlog has cleared from 120 open items to 40 over the last four reviews.” (d) “At the current completion rate, the pilot’s actionable backlog clears in about three weeks, with the range two and a half to four.” (e) “The corridor’s forecast at completion is about 2,551 million units against the 2,540 authorization.” (f) “Weekly completions over the last six weeks: eight, nine, ten, nine, eleven, ten.”

(a) is the variance, earned against planned, and the decision it supports is the schedule question, is the package earning at the planned pace, with the composition, the wet season, the permit, the mobilization, examined before the forecast moves. (b) is the variance of the actual against the commitment, the estimate of chapter 15, and the decision it supports is the learning question, what did the estimate miss, with the answer feeding the next estimate rather than the current forecast. (c) is a trend, the flow read across reviews, and the decision it supports is the near-term question, is the work accelerating into the finish, with the cadence and the definition of done checked against chapter 37’s rules. (d) is a forecast, the evidence-based expectation of the future with its range, and the decision it supports is the commitment question, what date can the project stand behind, with the percentile named. (e) is a forecast against the baseline, the variance at completion wearing its full name, and the decision it supports is the governance question, does the reserve hold, with the VAC of about 11 million inside the reserve band. (f) is a list of actuals, and the trap is naming it a trend: actuals become a trend only when read across time with a decision attached, and the statement as given supports no decision, which is the rear-view report in miniature. The common error in all six is naming the instrument by the number’s flavor rather than by the decision it serves, and the repair is the same in all six: the question the number answers is the instrument.

Two. A field drill: build the earned-value worksheet. Take the largest work package on the project you lead, or the one you work on, and build the worksheet this chapter worked for the seam package. (a) Write the budget at completion, the planned value at the last review date, the earned value from the accepted evidence under the counting rule of chapter 31, and the actual cost from the ledgers of chapter 18. (b) Compute the schedule variance, the cost variance, the schedule performance index, and the cost performance index, and write one sentence for each index saying what it is made of and whether its composition makes the extrapolation honest. (c) Write the three estimate-at-completion forms, rate-based, schedule-inclusive, and bottom-up, each with the sentence that names what it assumes persists, and choose the form the evidence supports. (d) Compute the to-complete performance index against the budget, and say what the number means as a decision rather than as a math fact.

The drill passes when the worksheet names the assumptions and the choice is argued from the evidence, not from the flattery of the number. The most common failure is the earned value taken from the physical claim or the percent-complete guess rather than the accepted evidence, the chapter 31 failure wearing a worksheet’s clothes; the repair is the counting rule, the accepted milestones, the independent evidence, written before the value is earned. The second failure is the index read as a verdict, the 0.50 answered with alarm or denial instead of examination; the repair is the composition question, what is this index made of, the procurement profile, the shock, the rework, and is the cause past or persistent. The third failure is the EAC chosen because it is the number the sponsor wants; the repair is the what-persists sentence written for every form before the choice, because the form is an assumption about the future, and the assumption is the argument. The fourth failure is the TCPI read as a call to work harder, a 5.5 answered with overtime; the repair is the reading this chapter taught, the TCPI is a statement about the remaining budget, and a statement about the budget is a decision about funding and scope, not an order for effort. A worksheet that cannot say what it assumes is a worksheet that has deleted its judgment.

Three. A numbers drill: reproduce the seam arithmetic, then run the sensitivity. (a) Verify from the chapter’s numbers: schedule variance minus 8 million units, cost variance minus 18 million, SPI about 0.69, CPI 0.50, rate-based EAC 80 million, rest-at-budget EAC 58 million, schedule-inclusive EAC about 100 million, bottom-up EAC 42 million, TCPI against the budget 5.5, and the corridor VAC of about 11 million against the 2,540 authorization. (b) Suppose the review comes a month later and the repair has progressed but not completed: earned value 24 million units, actual cost 40 million, planned value 31 million. Recompute the indices, the rate-based and bottom-up estimates, and the TCPI against the contract ceiling of 48, and say what changed. (c) Recompute the KijaniPay benefit gap if the August reading reaches 270 cumulative applications instead of the July 210, with the same target and the same two-month clock.

(a) The arithmetic is reproduced from the text: 18 minus 26 is minus 8; 18 minus 36 is minus 18; 18 over 26 is about 0.69; 18 over 36 is 0.50; 40 over 0.50 is 80; 36 plus 22 is 58; 36 plus 22 over 0.346 is about 100; 36 plus 6 is 42; 22 over 4 is 5.5; 2,540 minus 2,551 is minus 11. (b) The new indices: SPI is 24 over 31, about 0.77; CPI is 24 over 40, 0.60; the rate-based estimate is 40 over 0.60, about 67; the bottom-up estimate is 40 plus 6, 46, inside the ceiling of 48; the TCPI against the ceiling is 16 over 8, 2.0. What changed is the lesson of the chapter: the indices are recovering from the shock, the CPI from 0.50 to 0.60, the SPI from 0.69 to 0.77, and the rate-based estimate, about 67, still over-penalizes the future with the past, because the past rate is dominated by the procurement-heavy early spend and the failed run, while the bottom-up estimate, 46, is the honest forecast, because the remaining work, the repair’s tail and the acceptance plan, is estimable and the shock is priced; the corridor’s forecast holds as the 43.6 price impact of chapter 31 did, inside the commercial structure. (c) The gap closes: 350 minus 270 is 80 over the two months, 40 a month, and the current run-rate of about 55 a month would now overshoot the target, which is the forecast changing because the evidence changed, the windshield moving on its cadence, and the decision moving from closing a gap to protecting the guardrails while the run-rate holds.

Four. A decision room: the city wants one date. It is month twenty-five at BlueLine, the chair of the transport committee demands a single public date for the eastern segment, and the instruments disagree in the way this chapter opened with: the seam package’s indices at 0.69 and 0.50, the milestone trend holding at month twenty-seven for two reviews, the commissioning flow clearing its backlog at nearly twice the rate of two months ago, the register carrying the wet-season resumption as the swing factor, and the reserve band with its 2,630 range inside the 2,720 limit. The options: (a) announce month twenty-seven as a single date, on the strength of the trend and the flow, with the range deleted for the public; (b) announce month twenty-seven as the base case with the month-twenty-eight risk named, the trigger, the resumption’s two-week slip, the assumption, the wet season, and the reserve, stated in the public sentence; (c) announce month twenty-eight, building a buffer against the pessimistic case; (d) announce no date, citing the uncertainty, until the resumption evidence arrives. Decide the move, defend the trade, and say what the record must carry.

The defensible answer is (b), and the reasoning is the chapter’s discipline in one decision: the instruments are read as a system, the position corrected by the direction, the direction confirmed by the flow, the whole checked against the scenario set, and the honest output is the range with its assumptions and its trigger, the windshield in public form. The single date, (a), is the surrender to the demand this chapter opened against: the median announced as a commitment is a coin flip wearing a headline, and the corridor has already lived the cost of announced numbers, the ninety-two percent that the mayor’s office quoted and the earned evidence that did not support it; the range is the answer that can be defended when the trigger fires, because the trigger was named when the date was given. Month twenty-eight, (c), is the sandbag in calendar form: it buys buffer by paying the price of a false promise, the city planning around a date the evidence does not require, and it borrows from the reserve without earning the release, the chapter 15 sandbag wearing a forecast’s clothes. No date, (d), is reasonable and incomplete: the uncertainty is real, and the honest response to the chair is not the refusal of the date but the giving of the date with its width, because the range is the answer that holds the uncertainty and the commitment in one sentence. The record must carry the forecast with its percentile, the base case and the pessimistic case with their named assumptions, the trigger with its evidence and its owner, the reserve arithmetic with the 2,630 range inside the 2,720 limit, and the sentence to the city, the range, the assumption, the trigger, and the cadence of revision, because the public date is a commitment, and the commitment is only as durable as the record that says what it assumed.

Five. A decision room: the benefit clock at KijaniPay. It is July at KijaniPay, the benefit clock reads about 210 cumulative applications against the linear path of about 272, the target is 350 by September, the current run-rate is about 55 a month against the about 70 a month required, the gap at the current rate is about 30 applications, the flow forecast holds at about three weeks for the actionable backlog, and the growth experiments have returned, the settlement-visibility prototype is showing the merchant feedback moving, the fraud loss and the settlement promise hold inside their guardrails. The options: (a) reorder the quarter’s backlog to the application-friction and settlement-visibility work, set the run-rate target of about 70 a month by August with a mid-August check and a named owner for the lift, and report the benefit forecast as the range, on target if the lift holds, about 30 short if it does not; (b) hold the current ordering, which serves the flow, and report the forecast as it stands; (c) ask the team to work faster, extending hours and pulling the cohort ahead without changing the work; (d) move the target, renegotiating the 350 with the sponsors on the evidence of the first two quarters. Decide the move, defend the trade, and say what the record must carry.

The defensible answer is (a), and the reasoning is the two-forecast discipline of this chapter: the flow forecast says when the work will be done, and the benefit forecast says whether the work is working, and the two are kept separate on the page, with the flow range about three weeks holding while the benefit gap becomes the quarter’s decision. The reordering is the June lesson of chapter 32 applied in July: the backlog is ordered by outcome, risk, and evidence, and the application-friction and settlement-visibility work are the items the benefit clock names, the work that changes the run-rate rather than the work that flatters the flow; the run-rate target of 70 a month by August is the forecast made decision, the what-would-have-to-be-true of the scenario set, and the mid-August check is the trigger, the earliest evidence that the lift is holding. Holding the current ordering, (b), is reporting without deciding, the rear-view report in adaptive clothes, the forecast quoted and the quarter’s work unchanged, and it fails the chapter’s own test, the forecast exists to produce a decision, and a forecast that produces none is a forecast that has stopped serving. Asking the team to work faster, (c), is the effort-not-leverage error that chapter 30 and chapter 32 named: the run-rate is a property of the work’s friction and the system’s flow, and extending hours without changing the friction raises utilization, not application volume, the utilization trap of chapter 17 wearing a benefit target’s clothes; the settlement-visibility build is the leverage, and the leverage is the reordering. Moving the target, (d), is the baseline drift of chapter 31 in benefit form: the 350 is the product goal’s measurable shape, the fixed point that holds while the means flex, and a target moved on the first shortfall is a target that was never a target; the honest move is to hold the target, name the gap, and reorder the work, with the target revisited only through the governance that set it, on evidence, with a record, the discipline chapter 32 taught. The record must carry the reordered backlog with its outcome links, the run-rate target with its owner and its mid-August check, the benefit forecast with its range, the guardrail readings, the settlement promise and the fraud loss, unchanged, and the decision of the council, because the benefit clock is the adaptive windshield, and the windshield’s record is the quarter.

Six. The mastery drill: explain the conflicting signals. A platform project in its seventh month of ten carries the release the board has announced for month ten. Three instruments disagree. The earned-value schedule performance index reads about 0.85, the project is behind plan by the arithmetic. The milestone trend shows the last three releases shipping on their forecast dates, the trend flat for three consecutive reviews. And the team’s throughput has risen from about six to about nine completed items a week over the last six weeks, with the cycle time down and the work in progress down. The board asks the only question: are we on track for the month-ten release? Explain the disagreement, name the data to check before the forecast moves, and decide what to report.

The mastery is not the single explanation, it is the discipline of the trio, and the test is the one this chapter gave the corridor: every instrument is read for what it measures, and the disagreement is the diagnosis, not a tie-breaker vote. The three readings are consistent with at least two credible diagnoses, and the drill passes when both are on the table before the forecast moves. Diagnosis one: the SPI’s planned value is stale. The baseline was never re-sequenced after the scope change, the planned value counts work the plan scheduled in months the plan no longer schedules, the denominator is a record of an old future, and the index, 0.85, is a snapshot of a plan that is no longer the plan, while the milestone trend and the flow, reading the real work, agree the project is healthy; the data to check is the baseline-change record, the re-sequencing decisions, the milestone dictionary, and whether the planned value was recomputed when the scope moved, because an index measured against a stale denominator is a claim wearing an index’s clothes. Diagnosis two: the flow is inflated by smaller items. The throughput rose because the backlog was split, the items got smaller, the completion count rose while the value earned did not, and the milestone trend is flat because the releases are the evidence that matters and they are holding, while the SPI is the honest reading of the value gap; the data to check is the item sizes, the definition of done, the release burnup, the value line against the scope line, and whether the throughput is counting completion or counting slicing, the chapter 17 utilization lesson wearing a flow chart’s clothes. Both diagnoses are credible, the evidence distinguishes them, and the unsafe move is choosing the flattering reading, the flow, and announcing the date, or the alarming reading, the SPI, and re-sequencing the work without the explanation; the safe move is the chapter’s sequence, check the data, name the composition of every instrument, let the release burnup and the milestone dictionary arbitrate, and report the range, on track in the base case with the assumptions named and the trigger set on the evidence that would move it, because the forecast is built from the sequence of the instruments, and the sequence is only as honest as the data it is fed. Credit belongs to any answer that diagnoses before deciding, names the data to check, and reports the range; the wrong answer is the one that picks a single instrument, flattering or alarming, and announces a date on it.

Seven. The transfer question. On the project you lead, or the one you work on, where does your report end: at the variance or at the forecast? Does your steering meeting explain the month and never commit to the future, and is the meeting comfortable because the future is absent? What would a windshield page look like for your project: the trend, the forecast range, the assumptions, the triggers, and the decision each trigger would require, and who owns the page? What does your estimate at completion assume persists, and do you believe it: is your shock past or persistent, is your index made of a one-time event or a systemic rate, and which of the three forms, rate-based, schedule-inclusive, or bottom-up, does your evidence support, or is your forecast the number the spreadsheet printed? When did your milestone trend last move, and what did the movement say: was the slip explained, the recovery evidenced, the hold tested, and is your milestone dictionary fixed, or are your milestones moving to match the achieved dates? Does your flow forecast carry a percentile, and who chose it: the median, the 85th, the tail, and was the choice made for the decision or for the audience, and does your burnup draw the scope line, or does your burndown hide the scope growth? What is your scenario set: the base, the pessimistic, the optimistic, each with its named assumptions and its triggers, and what would have to be true for your date to move, and is that sentence written anywhere, or is the date defended without the conditions? Is your Monte Carlo a calibrated instrument or a ritual: do your input distributions come from evidence and reference classes, is your model’s logic the project’s logic, and is the machine computing or deciding? Where is your theater: the single-number answer, the frozen forecast, the rear-view report, the false precision, the cherry-picked percentile, the simulation worship, and which act is on your stage this month? And the question underneath all of them, the one the corridor learned in its month-twenty-five season: when the instruments disagree, do you read the trio or do you choose the instrument, because the durable principle of this chapter is the one the review proved: the variance is the rear-view mirror and the forecast is the windshield, the index is a snapshot and the trend is a trajectory and the flow is the near-term truth, the forecast is a set of assumptions about what persists, chosen by judgment, carried as a range, watched at the trigger, and revised on a cadence, and mastery of forecasting is not the production of the date, it is the production of the honest range and the courage to give it, because the date that hides its range is a promise, and the range that names its assumptions is a decision. The most common next failure is the one the corridor will face in the months after this chapter ends, and it is the failure every forecast faces when the future arrives: the trigger fires and the meeting does not convene, the range narrows and the reserve is treated as a surplus to be spent, the forecast is revised and the decision is postponed, the instruments are read and the action is deferred, which is why the forecast has its cadence, its owners, and its triggers, the same discipline every instrument in this book carries, the instrument that is read and not acted on is the instrument that becomes the theater. And the next chapter turns to the path the forecast opens: the new information that arrives after the forecast, the issue, the change, the decision, and the configuration that the new information must travel through, the Meridian policy change that will be logged as a software change and will turn out to be a training change, a workflow change, a data change, and an acceptance change, which is the subject of the next chapter, the control of issues, changes, configuration, and decisions that turns the forecast of this chapter into the actions of the next, the evidence-to-decision path that chapter 38 built and chapter 39 will run.*

Notes

• The composite cases remain author-created illustrative material. The BlueLine month-twenty-five control review, the transport committee chair’s demand for a single eastern-segment date, Daniel Osei’s range on the screen, the corridor’s three instruments disagreeing, the milestone trend with its drift and recovery and hold, the commissioning flow clearing at nearly twice the rate of two months ago, the sentence to the city, and all named characters and roles are teaching constructions consistent with the facts established in earlier chapters: the corridor’s four segments, twenty-four stations, the phased opening of the central and northern segments at month twenty-five with the eastern segment at month twenty-seven or twenty-eight, Lena Voss’s month-twenty-five completion obligation, the mayor’s month-twenty-four promise, the contractor’s ninety-two percent counting rule, and the seam package’s planned value of 26 million units against earned 18 and actual 36, the indices at 0.69 and 0.50, the repair estimate of 6 million and five weeks, the bottom-up forecast actual of 42 inside the ceiling of 48, the shared-savings price of 43.6 million against the at-target 43.2, and NCR-031, from chapters 7, 15, 18, 20, 31, and 37; the corridor totals, the 2,400 million-unit capital envelope, the 140 million-unit mitigation line, the 2,540 million-unit authorization, the forecast at about 2,551 inside the reserve band, the 80-percent range on the remaining work running to about 2,630 inside the total authorized limit of 2,720, the cash peak of about 400 million units in month twenty-five, the eastern segment’s variation of 38 million units in the baseline with the two-week mobilization and the wet-season break, and the systems package’s earned-value arithmetic from chapter 18, the budget at completion of 120 million units, the planned 60, the earned 45, the actual 48, and the three estimate-at-completion forms at 128, 154.7, and 126, are unchanged from the chapters that established them; the KijaniPay June numbers, the twenty-seven actionable items, the nine completions a week with the observed range seven to eleven, the cycle time about 4.7 days, the forecast range two and a half to four weeks, the 155 applications against the 233 linear path, the 350 target against the 1,400-merchant cohort, the 78-application gap, and the 65-a-month requirement against the 43-a-month run-rate, are unchanged from chapters 17, 30, and 32. The teaching numbers introduced here are fully reproducible from the text: the seam worksheet’s schedule variance of minus 8 million units, 18 minus 26, the cost variance of minus 18 million, 18 minus 36, the rate-based estimate at completion of 80 million, 40 over 0.50, the rest-at-budget estimate of 58 million, 36 plus 22, the schedule-inclusive estimate of about 100 million, 36 plus 22 over about 0.346, the bottom-up estimate of 42 million, 36 plus 6, the to-complete performance index of 5.5, 22 over 4, and the corridor variance at completion of about minus 11 million, 2,540 minus 2,551; the sensitivity run, the earned value of 24 against the planned 31 and the actual 40, gives the schedule performance index about 0.77, 24 over 31, the cost performance index 0.60, 24 over 40, the rate-based estimate about 67, 40 over 0.60, the bottom-up estimate 46, 40 plus 6, and the to-complete performance index against the ceiling of 48 equal to 2.0, 16 over 8, an author-created teaching scenario presented with its assumptions; the KijaniPay July benefit-clock readings, the cumulative applications of about 210 against the month-seven linear path of about 272, 350 times seven over nine, the remaining gap of 140 over two months, 70 a month, the current run-rate of about 55 a month, and the shortfall of about 30 at the current rate, 350 minus 210 minus 110, are author-created teaching numbers extending the established June facts, presented with their assumptions; the eastern segment’s milestone trend, the baseline date of month twenty-seven with the forecast dates of month twenty-nine, month twenty-eight, month twenty-seven, and month twenty-seven across the four reviews, and the commissioning-flow clearing figure, are author-created teaching numbers consistent with the established month-twenty-seven forecast and the month-twenty-eight risk; the Monte Carlo toy, the three-chain commissioning sequence with the three-point durations, the deterministic sum of the most likely values at 21 weeks, the median above it, and the 85th percentile at about 25 weeks with the correlated wet-season cause widening the distribution, is a generic author-created teaching construction stated as such, not a fact from the case’s earlier chapters; the KijaniPay flow distribution with its simulated tail frequency, the median at three weeks with the 85th at about four and the pessimistic tail at five or six, is a teaching interpretation of the established flow numbers under stated assumptions; and the mastery drill’s platform project, the SPI at 0.85, the flat milestone trend, and the throughput rising from about six to about nine, is an author-created composite scenario for the drill, with the two credible diagnoses, the stale planned value and the smaller items, presented as alternative readings the evidence must distinguish. The standards and sources are described in the book’s own words: the earned-value management practice, the planned value, the earned value, the actual cost, the variances, the performance indices, the estimate-at-completion forms, and the to-complete performance index, follows the earned value management standard ISO 21508:2018, Earned value management in project and programme management, and the earned-value criteria of ANSI/EIA-748, both introduced in chapter 31, and the three estimate-at-completion forms follow the treatment established in chapter 18, the rate-based, the schedule-inclusive, and the bottom-up, each presented here as an assumption about what persists rather than as a mechanical output; the flow forecasting, the throughput, the cycle time, the cycle-time percentiles, the aging work, and the probabilistic interpretation of the flow forecast, follows John D. C. Little, “A Proof for the Queuing Formula: L = λW,” Operations Research 9(3), 1961, and Daniel S. Vacanti, Actionable Agile Metrics for Predictability: An Introduction (Neptune Township, NJ: DZone, 2015), as cited in chapters 15 and 17, with the percentiles and the range described here in the book’s own words; the Monte Carlo method follows Nicholas Metropolis and Stanislaw Ulam, “The Monte Carlo Method,” Journal of the American Statistical Association 44(247), 1949, pages 335-341, which introduced the method’s name and its published shape, described here in the book’s own words as the repeated random sampling of a stochastic model, with the caveat that the method’s output is only as honest as its input distributions and its model logic, and the calibration check via the reference classes of chapter 15; the three-point estimation and its distributional properties follow the PERT treatment and the reference-class discipline established in chapter 15; the general monitoring, measurement, and performance-evaluation frames follow ISO 21502:2020, Project, programme and portfolio management: Guidance on project management, and the PMBOK Guide, Eighth Edition, Project Management Institute, November 2025, per the book’s reference baseline of 1 August 2026, which treat measurement and performance analysis among their general practices and performance domains, all described here in the book’s own words as general frames rather than quoted. This book remains independent of PMI, ISO, and all standards and framework bodies, and no proprietary certification manual, commercial text, or framework guide is reproduced or paraphrased here.