Representation by Timothy J. Harland Same Picture, Different Lens · Book Three Piece 3 of 6 — Chapters Five–Six From the Author’s Portal · showyourmap.com Free for weary travelers. E UNUM PLURIBUM. ------------------------------------------------------------ CHAPTER FIVE Persistence At 8:06 the following morning, someone had erased the whiteboard. Lena stopped in the doorway. Jonah nearly walked into her. “What?” She moved aside. He looked into the conference room. “Oh.” The board was clean. Not completely. A faint gray residue remained where yesterday’s questions had been: NONCOMPLIANT → ? CONFUSED → ? DROWSY → ? REFUSING → ? And beneath them, barely visible: A USEFUL LABEL BECOMES DANGEROUS WHEN ITS PROVISIONAL STATUS DISAPPEARS. Jonah stepped into the room. “Housekeeping?” “Probably.” “You look offended.” “I’m not.” “You stopped walking.” “The board contained work.” “We have pictures.” “I know.” “Notes.” “Yes.” “Carla probably has three copies.” “Probably.” Jonah looked at the erased board. “So nothing was lost.” Lena turned toward him. He smiled. “Sorry.” “No, you aren’t.” “No.” She put her bag on the table. “Nothing was lost from storage.” Jonah nodded. “But?” “The working state disappeared.” His smile faded. He looked at the board again. “Oh.” Lena uncapped a marker. Then stopped. Jonah noticed. “You’re not rewriting it.” “No.” “Why?” “Because I want to know what I remember.” “That seems dangerous.” “It’s bounded.” “Still dangerous.” She handed him another marker. “You too.” “Why?” “Independent reconstruction.” Jonah looked at the marker. Then at her. “You’ve made remembering a test.” “Yes.” “That is deeply on brand.” “Write.” They divided the board vertically. Lena took the left. Jonah took the right. Neither looked at the other’s side. Lena wrote: OBSERVATION → INTERPRETATION → LABEL → DECISION Then: LABEL MAY FUNCTION AS A PROXY FOR CURRENT UNDERSTANDING Then: PROVISIONAL She stopped. Something else had been there. She could reconstruct the concept. Not the wording. She left space. On the other side Jonah wrote: WHAT CAN’T WE AFFORD TO LOSE? Then: LABELS CHANGE WHAT PEOPLE LOOK FOR Then: THE NAME ISN’T THE THING Lena glanced sideways. “You looked.” “I did not.” “That last one is suspiciously concise.” “I contain multitudes.” “Unfortunately.” He continued. TEMPORARY ANSWERS HAVE A WAY OF BECOMING PERMANENT Lena stopped writing. “What?” “What?” “Say that again.” “Temporary answers have a way of becoming permanent.” “That wasn’t on the board.” “No.” She looked at him. “Why did you write it?” Jonah shrugged. “Because that’s what yesterday felt like.” Lena looked at her own word. PROVISIONAL Then at his sentence. Different representation. Same— She stopped herself. “Maybe,” she said. Jonah looked over. “I didn’t ask.” “I know.” He smiled. Carla arrived at 8:19. She carried coffee and a folder. She stopped when she saw the divided whiteboard. “No.” Jonah turned. “Good morning.” “I left you alone for thirteen minutes.” “Fourteen,” Lena said. Carla looked at her watch. “That does not improve this.” She put the folder down. “What are you doing?” “The board was erased,” Lena said. “I see that.” “We reconstructed it independently.” “Why?” Jonah pointed at the clean portions of the board. “Apparently because nothing can simply happen anymore.” Carla looked at Lena. “Accurate?” “Approximally.” Jonah turned slowly toward her. Lena saw him. “Don’t.” Carla looked between them. “I’m going to regret asking, but what does that mean?” “Optimal approximation,” Jonah said. Carla waited. “That did not help.” “It will.” “No.” Lena pointed to the folder. “What’s that?” Carla looked down. Her expression changed. “Right.” She opened it. “I brought you something.” The document was twelve pages long. PATIENT MOBILITY SAFETY IMPROVEMENT INITIATIVE Carla flipped to page seven. A table listed common contributing factors in inpatient falls. Medication. Environmental obstruction. Improper footwear. Failure to request assistance. Confusion. Noncompliance. Lena pointed to the last two. “When was this written?” “Four years ago.” “Current?” “Yes.” “Used for?” “Fall reviews. Training. Some quality reporting.” Jonah looked at Carla. “You came here because of noncompliance.” “Yes.” “Yesterday?” “Yes.” “No.” Carla frowned. “What?” “You didn’t come because of yesterday.” “I absolutely came because of yesterday.” “You found something.” Carla looked at Lena. Lena was already reading page eight. Jonah smiled. “There it is.” Carla sighed. “Fine.” She turned the page. A decision tree. PATIENT FAILED TO FOLLOW ESTABLISHED FALL- PREVENTION PROTOCOL Below it: Was patient cognitively capable of understanding instruction? YES → NONCOMPLIANCE NO → COGNITIVE CONTRIBUTOR UNKNOWN → FURTHER REVIEW Lena stared at it. “How is cognitive capability determined?” “That’s the problem.” “By whom?” “Usually the reviewer.” “Based on?” “Chart documentation.” Jonah leaned closer. “And if the chart already says noncompliant?” Carla nodded. “Exactly.” Silence. Lena traced the tree with one finger. Patient behavior generated documentation. Documentation generated a label. The label entered a review. The review used the documentation to determine which label best explained the behavior. And then— “What happens to the result?” she asked. “Falls database.” “And?” “Quarterly reporting. Training priorities. Unit-level interventions.” “Does it return to the chart?” “Not directly.” “Does it affect future practice?” “Yes.” “How?” “If a unit has a pattern of noncompliance-related falls, education may focus on patient adherence.” Jonah sat back. Samantha arrived at 8:37. Carla had called her without asking. She read the decision tree. “Oh.” Jonah looked at her. “That seems to be happening a lot.” “It’s a useful syllable.” Lena asked, “Can you trace the field?” “Which field?” “Noncompliance. From chart to falls database.” Samantha sat. “Probably.” “Probably?” “There may be transformations.” “What transformations?” “Mapping. Normalization. Category consolidation.” Lena stared at her. Samantha sighed. “Yes, I hear it.” “Good.” Jonah looked between them. “I don’t.” Samantha pointed at the decision tree. “The word in the chart may not be the same field in the reporting database.” “But it means the same thing?” “Hopefully.” Lena said, “Show us.” The first transformation was simple. Clinical review: NONCOMPLIANCE became: PATIENT ADHERENCE CONTRIBUTOR Jonah pointed. “Different words.” “Same reporting category,” Samantha said. Lena asked, “Why change it?” “Standardization. Different units use different terms.” The second transformation combined several categories. REFUSAL NONCOMPLIANCE FAILURE TO FOLLOW INSTRUCTION became: ADHERENCE / BEHAVIORAL FACTOR Jonah frowned. “Those aren’t the same.” “No,” Samantha said. “But for aggregate reporting, they’re considered similar.” “Why?” “Otherwise the categories become too small to analyze.” Lena nodded. Necessary compression. Again. Samantha continued. At the enterprise level: ADHERENCE / BEHAVIORAL FACTOR joined: COGNITIVE FACTOR COMMUNICATION FACTOR OTHER PATIENT-SPECIFIC FACTOR under: PATIENT FACTOR Jonah stared. “Now they’re definitely not the same.” “They aren’t supposed to be.” He looked at her. Samantha pointed to the reporting hierarchy. “This level isn’t asking why an individual patient fell. It’s asking how many falls had staff, environmental, equipment, medication, or patient contributors.” Lena nodded. “Different decision.” “Yes.” “Different required resolution.” “Yes.” Jonah looked at the hierarchy. “So each level is accurate for what it’s trying to do.” “Approximally,” Lena said. Jonah looked at her again. She ignored him. Samantha smiled. “I’m starting to like that word.” “Don’t encourage them,” Carla said. Lena drew the transformations on the board. RUTH KELLER ↓ NONCOMPLIANT WITH WALKER ↓ PATIENT ADHERENCE CONTRIBUTOR ↓ ADHERENCE / BEHAVIORAL FACTOR ↓ PATIENT FACTOR Jonah stared at it. “By the bottom, Ruth is gone.” Lena looked at him. “Yes.” Carla said, “She’s supposed to be.” Neither answered. Carla continued. “That’s aggregate data. We cannot retain individual narratives in enterprise reporting.” “I know,” Lena said. “Then what’s wrong?” “Nothing.” Carla waited. Lena pointed at the hierarchy. “Yet.” Carla closed her eyes. Jonah smiled. “She really hates that.” “I know.” They pulled the previous quarter’s fall report. PATIENT FACTORS: 31% Under recommendations: Continue patient education regarding fall-prevention protocols. Reinforce adherence to call-light and assistive- device expectations. Lena read it once. Then again. “Why education?” Carla answered. “Because adherence was a major contributor.” “According to which level?” Carla stopped. Samantha opened the underlying data. Twenty-six incidents had been classified under patient factors. Nine involved adherence or behavioral factors. Seven cognitive. Four communication. Six other. Jonah looked at the recommendation. “All twenty-six became education?” “Not exactly,” Carla said. “There are other recommendations.” “But this one came from the category.” “Yes.” Lena asked, “What did the nine adherence cases contain?” Carla knew where this was going. “You want all nine.” “Yes.” “That’s going to take time.” “How much?” “An hour. Maybe two.” Lena looked at Jonah. He looked at the clock. “Bounded?” “Quarter only.” “Relevance?” “Whether the aggregate recommendation remains supported by the source cases.” He nodded. “Clinical constraint acknowledged.” Carla stared at him. “What does that mean?” Jonah smiled. “It means she didn’t ask for four years.” Carla considered that. “Fine.” They reviewed nine. Three were straightforward. Patients understood instructions and deliberately chose not to follow them. Two involved cognitive impairment that had not yet been recognized at the time of the incident. One involved a patient who could not reach the call light. One involved language access. One involved urgency related to diarrhea. The last involved Ruth Keller. Jonah counted. “Three.” Carla looked at him. “Three what?” “Three where education about adherence clearly matches the source problem.” Lena corrected him. “Three where the available documentation supports that interpretation.” “Right.” “Different.” “I know.” Samantha looked at the remaining six. “The reporting classification isn’t necessarily wrong.” “No,” Lena said. “Because the patient’s behavior contributed to the fall risk.” “Yes.” “But the recommendation assumes the behavior is modifiable through education.” Lena looked at her. “Yes.” Carla sat back. “Oh.” Jonah pointed at her. “That one.” She ignored him. Lena returned to the whiteboard. Under the hierarchy she wrote: REPRESENTATION 1 → supports decision A REPRESENTATION 2 → supports decision B REPRESENTATION 3 → supports decision C Then: WHAT HAPPENS WHEN REPRESENTATION 3 IS USED TO MAKE DECISION A? Samantha stood. “That.” Carla looked at her. “What?” “That may be the problem.” Lena waited. Samantha pointed at the hierarchy. “Each abstraction level is valid for its intended use. But once the data exists, people reuse it.” “Reports,” Carla said. “Dashboards. Benchmarking. Staffing. Training. Policy.” Jonah added, “Questions it wasn’t built to answer.” Nobody spoke. Lena looked at the board. A representation did not have to be wrong. It only had to travel. At 10:02, they moved to the larger conference room. Megan joined them. Vasquez arrived five minutes later. Lena entered first. She chose the chair at the end nearest the whiteboard. The monitor sat directly ahead of her. Everyone else fell within the same field of view. Jonah stopped before sitting. Lena looked at him. “What?” “You picked an anchor.” “I picked a chair.” He looked at the board. Then the monitor. Then the room. “Approximally.” Lena followed his glance. Realized what she had done. Looked back at him. “Don’t.” He sat. “I didn’t say anything.” “You were going to.” Megan whispered to Carla, “What just happened?” Carla whispered back, “I stopped asking.” Samantha presented the fall-report hierarchy. Vasquez understood immediately. “You’re mixing levels of abstraction.” Lena looked at her. “Possibly.” Vasquez smiled. “Of course.” Carla said, “The enterprise category is valid.” “Yes.” “The unit-level classification is valid.” “Potentially.” “The individual documentation is valid.” “Potentially.” Carla looked frustrated. “Then where exactly is the failure?” Jonah looked at Lena. She waited. He understood. This one was his. He pointed to the quarterly recommendation. “What question was this report built to answer?” Carla looked at it. “What contributes to inpatient falls.” “And what question did the recommendation answer?” “How do we reduce them.” “Same question?” Carla started to answer. Stopped. Vasquez leaned back. “No.” Jonah nodded. “The first is descriptive.” Samantha added, “The second is intervention.” Lena watched the room. Nobody needed her yet. Good. Megan said, “Knowing something contributed doesn’t tell you how to change it.” “Exactly,” Vasquez said. Carla looked at the nine cases. “So we had valid data.” “Yes,” Lena said. “A valid category.” “Yes.” “A valid report.” “Yes.” “And still reached a recommendation the source data doesn’t fully support.” Lena considered the wording. “Approximally.” Carla stared. Jonah laughed. “Now she’s doing it.” At 10:39, the room had stopped discussing noncompliance. That was another sign. The question had moved. Samantha pulled a readmission dashboard. “Same architecture.” Medication-related readmission. Access. Jonah pointed around the room. “We’re about to follow every place this could happen.” Samantha closed the readmission dashboard. Vasquez smiled. “He’s right.” Lena looked at the board. Falls. Readmissions. Infections. Pain. Satisfaction. Five branches already. Diminishing returns. She erased four. Left falls. Carla watched. “That physically hurt me.” “Why?” “You just erased potentially useful investigations.” At 11:06, the working group had a bounded question. Not: Are labels bad? Not: Is aggregation dangerous? Not: Does healthcare oversimplify patients? Lena rejected all three. Instead: CAN A REPRESENTATION REMAIN VALID WHILE BECOMING UNSUITABLE FOR A NEW DECISION? Vasquez read it. “Yes.” Lena looked at her. “That was not a request for an answer.” “I know. But yes.” Jonah smiled. Lena ignored both of them. Samantha photographed the board. Carla said, “If the answer is obviously yes, why are we investigating it?” Jonah looked at Lena. She nodded. He answered. “Because we don’t know when people notice the decision changed.” Carla went still. Lena added beneath the question: WHERE IS THE DECISION BOUNDARY? Samantha stared at it. “That’s measurable.” “How?” Megan asked. “Trace where data changes purpose.” Lena nodded. “Not where it changes format.” Samantha looked at her. “Right.” Jonah added, “The words can stay exactly the same and the purpose can change.” Lena turned toward him. “Yes.” He smiled. She did not tell him to stop. At 11:34, Carla received an email. She read it. Then read it again. Jonah noticed. “Problem?” “I don’t know.” Lena looked over. “Relevant?” “Possibly.” Carla turned the laptop. A department administrator had requested the current fall- risk dashboard for a staffing review. Samantha frowned. “Why?” “Budget planning.” Lena looked at the email. “What does the dashboard measure?” “Fall rates. Contributing factors. Interventions.” “What staffing decision?” Carla read further. “They’re reviewing whether overnight sitter utilization can be reduced on two units.” “You can’t just walk into Finance.” Lena stopped at the door. “Why?” “Because—” Carla paused. Jonah waited. Carla sighed. “I don’t actually have a good answer.” “Good.” Jonah stood. Carla looked at him. “Where are you going?” “With her.” “I assumed that. Why?” He picked up his notebook. “Different decision.” Lena opened the door. Jonah followed. Carla called after them. “Do not interrogate Finance.” Lena stopped. Turned. “What would you prefer?” Carla thought. “Ask questions like normal people.” Lena looked at Jonah. He looked at her. Neither spoke. Carla closed her eyes. “Just go.” The elevator doors closed. Jonah leaned against the wall. “You know what this means.” “No.” “We found the next layer.” “Possibly.” “Same data.” She looked toward the administrative corridor. “Reuse.” Jonah stopped. The word was simple. Almost disappointingly so. He smiled. “Keep it?” Lena continued walking. “For now.” Behind them, the elevator doors closed. Upstairs, the whiteboard still held the question: CAN A REPRESENTATION REMAIN VALID WHILE BECOMING UNSUITABLE FOR A NEW DECISION? For the first time, they were about to follow the representation somewhere the patient never went. CHAPTER SIX The Decision Finance was quieter than Lena expected. Not silent. Quiet in the way spaces became quiet when most conversations happened through spreadsheets. Jonah noticed her looking. “What?” “Nothing.” “You expected something.” “No.” “That was fast.” Lena kept walking. Carla had called ahead despite saying she wouldn’t. The department administrator was waiting for them in a glass-walled office overlooking the parking structure. Daniel Cho. Director of Resource Planning. He stood when they entered. “Carla said you had questions about the fall dashboard.” “Yes,” Lena said. Daniel looked at Jonah. “And you are?” “Jonah Reed.” A beat. Daniel waited. Jonah waited back. Lena said, “He’s with me.” Daniel nodded. “That somehow created more questions.” “It usually does,” Jonah said. They sat. Daniel turned his monitor. “Which part of the dashboard?” Lena remained standing. “The part you intend to use.” Daniel looked at her. “All of it?” “For what decision?” “Sitter utilization.” “Specifically?” Daniel leaned back. “You really do start there.” “Where?” “With the decision.” “Yes.” Jonah took the chair nearest the door. Daniel glanced at him. Jonah shrugged. “She does.” Daniel opened a staffing model. Two inpatient units. Four quarters. Fall rates. Patient acuity. Sitter hours. Overtime. Agency utilization. Labor cost. Lena moved closer. “What are you trying to determine?” “Whether current overnight sitter utilization is justified.” “Define justified.” Daniel looked at her. Jonah smiled. Daniel noticed. “Is that funny?” “No.” “Yes,” Jonah said. Lena ignored him. Daniel turned back to the model. “We’ve increased sitter hours seventeen percent over eighteen months. Falls haven’t decreased proportionally.” “So the question is whether sitters work.” “No.” Lena looked at him. Daniel continued. “The question is whether the additional sitter hours are producing enough benefit to justify the additional cost.” Jonah stopped smiling. Lena noticed. “Good,” she said. Daniel raised an eyebrow. “Good?” “That is a different question.” “From what?” “Whether sitters work.” “Yes.” He waited. “So?” Lena sat. “Show me the model.” Daniel started with labor. Average sitter hours per night. Hourly cost. Overtime exposure. Agency coverage. Vacancy rates. Then fall outcomes. Falls per thousand patient-days. Falls with injury. Falls during sitter coverage. Falls without sitter coverage. Jonah watched the columns appear. “This is more complicated than the dashboard.” “Of course.” “Why?” Daniel looked at him. “Because dashboards are for people who don’t want to look at this.” Jonah nodded. “I respect the honesty.” Daniel scrolled. “We’re not trying to eliminate sitters.” “What are you trying to do?” Lena asked. “Determine whether we can reduce baseline overnight coverage and deploy sitters more selectively.” “Based on?” “Risk.” “Defined how?” Daniel opened another tab. A fall-risk score. Age. Mobility. Medication. Cognition. Prior falls. Assistance requirements. Lena recognized it. “Where does patient behavior enter?” Daniel pointed. “Here.” ADHERENCE / BEHAVIORAL FACTOR Jonah leaned forward. “There it is.” Daniel looked at him. “What?” “Nothing.” Lena said, “Not nothing.” Jonah looked at her. “Growth.” “Don’t.” Daniel waited. Lena pointed at the field. “Where does that value come from?” “The safety data.” “Which comes from?” “I assume clinical review.” “You assume?” Daniel paused. “Yes.” “Have you verified it?” “No.” “Why not?” “Because I don’t own the source system.” Lena stopped. That answer mattered. Not because it was wrong. Because it was ordinary. Daniel folded his arms. “Is the data bad?” “No.” “Then I’m missing something.” Jonah looked at Lena. She nodded once. He took it. “The data was built to answer a different question.” Daniel looked at him. “What question?” “What contributed to a fall.” “And?” “You’re using it to predict who needs a sitter.” “Yes.” “Same question?” Daniel started to answer. Stopped. Lena watched him reconstruct it. “No.” Jonah nodded. Daniel turned toward the screen. “But contribution should still have predictive value.” “Possibly,” Lena said. “There it is.” “What?” “That word.” “Which?” “Possibly.” Jonah leaned back. “You get used to it.” “No, you don’t,” Carla said from the doorway. Everyone turned. She held up a coffee. “I decided this required supervision.” “You called ahead,” Lena said. “Yes.” “That was supervision.” “It was insufficient.” She sat beside Jonah. Daniel looked at her. “Did you know about this?” “I knew enough to become uncomfortable.” “That’s reassuring.” “It wasn’t intended to be.” Lena returned to the model. “How much weight does adherence carry?” Daniel opened the scoring logic. “Not much.” “How much?” “Four percent of the composite risk score.” Carla relaxed slightly. Jonah saw it. He said nothing. Lena did too. She also said nothing. Daniel noticed both of them noticing. “What?” “Nothing,” Carla said quickly. Jonah looked at Lena. Lena continued. “If we remove it?” Daniel changed the weight to zero. The model recalculated. Almost nothing happened. Jonah looked disappointed. “So it doesn’t matter.” Daniel nodded. “Not materially.” Lena looked at the rankings. Two patients changed position. Neither crossed the staffing threshold. “Correct.” Carla exhaled. “There.” Lena looked at her. “There what?” “The questionable category isn’t affecting the decision.” “Correct.” Carla waited. Lena returned to the screen. Jonah smiled. Carla pointed at her. “I know that look.” “I don’t have a look.” “You absolutely have a look.” Daniel said, “What am I missing?” Lena pointed at the model. “The field isn’t the question.” Daniel looked at the screen. “What is?” “The threshold.” He opened the staffing threshold. Patients above a certain composite score qualified for recommended overnight sitter coverage. Below it, standard fall precautions. Jonah read the number. “Why forty-two?” Daniel answered immediately. “Because that’s where the model showed the best balance between fall reduction and staffing cost.” Lena looked at him. “When?” Daniel paused. “During implementation.” “When was that?” “Three years ago.” “Using what data?” “The prior two years.” “So the threshold represents conditions from three to five years ago.” “Yes.” “Has it been recalibrated?” Daniel hesitated. “No.” Carla looked at him. “Why not?” “The model has been performing adequately.” Lena asked, “Defined how?” Daniel looked at her. Then laughed once. “Right.” He opened another report. Overall falls had remained within target. Injury falls had decreased slightly. Sitter cost had increased. Overtime had increased. Agency utilization had increased. Patient acuity had also increased. Jonah pointed. “That seems relevant.” “It is,” Daniel said. “How much?” “That’s what we’re trying to determine.” Lena looked at the model again. “Then why are you reducing sitter utilization?” “We haven’t decided to.” “The email said staffing review.” “Yes.” “That isn’t the same thing.” “No.” Carla looked at Jonah. Jonah whispered, “She likes him.” “I heard that,” Lena said. Daniel looked between them. “I have no idea what’s happening.” “Normal,” Carla said. Daniel pulled the original implementation analysis. Three years earlier, the model had estimated that targeted sitter coverage would reduce falls enough to offset a portion of the increased labor cost. Not eliminate the cost. Justify it. Expected: 11% reduction in falls 7% reduction in falls with injury 9% increase in sitter labor cost Actual: 6% reduction in falls 8% reduction in falls with injury 17% increase in sitter labor cost Jonah read the figures. “So it failed.” Daniel shook his head. “No.” Jonah looked at him. Daniel pointed. “Injury falls improved more than expected.” “But total falls improved less.” “Yes.” “And cost almost doubled the expected increase.” “Yes.” “So?” Daniel leaned back. “So the answer depends on what return you care about.” Lena looked at him. There. Jonah looked at her. He had heard it too. Carla hadn’t. “What?” she asked. Lena pointed at the original model. “What return was this designed to optimize?” Daniel answered. “Fall reduction relative to labor cost.” “Total falls?” “Primarily.” “Not injury severity?” “It was included.” “How?” Daniel opened the old workbook. A weighted cost estimate. Emergency imaging. Extended length of stay. Treatment. Liability exposure. Jonah frowned. “You put a price on someone getting hurt?” Daniel didn’t react. “We put an estimated cost on the consequences associated with injury falls.” “Different.” “Yes.” Lena glanced at Jonah. He nodded. Correction accepted. Daniel continued. “The model needs comparable variables. That doesn’t mean the variables are equivalent.” Lena looked at him again. “You should come upstairs.” Carla nearly choked on her coffee. Daniel laughed. “I’ll pass.” Lena wrote on Daniel’s whiteboard. EXPECTED RETURN Below it: ACTUAL RETURN Then: VARIANCE Daniel stood beside her. “That’s what we’re doing.” “No.” He looked at her. “You’re comparing current cost to current outcomes.” “Yes.” “That’s not the same thing.” “Why?” “What did you expect to happen?” Daniel pointed at the original analysis. “That.” “What actually happened?” “The newer report.” “What assumptions explain the difference?” He stopped. Jonah looked at the board. “That’s the missing step.” Daniel nodded slowly. Carla said, “You mean they shouldn’t decide whether the program is worth continuing until they know why actual performance diverged from expected performance.” “Not exactly,” Lena said. Carla closed her eyes. “Of course not.” Lena continued. “They may still need to make a decision before they know why.” Daniel nodded. “Budget cycle.” “So uncertainty does not remove the decision.” “No.” “It changes the confidence.” “Yes.” Jonah looked at the three words. Expected. Actual. Variance. “So you don’t need the perfect answer.” Lena turned. “No.” “You need the best estimate after incorporating what happened.” “Yes.” Daniel said, “A revised forecast.” Lena considered. “An updated approximation.” Jonah smiled. Carla pointed at him before he could speak. “Whatever you’re about to say, don’t.” “I wasn’t going to say anything.” “Now you understand how I feel,” Lena said. Carla stared at her. Jonah laughed. At 12:17, Samantha joined by video. Her face appeared on Daniel’s monitor. “Why am I in Finance?” Carla answered. “Poor choices.” “Fair.” Lena showed her the original model. Samantha read the assumptions. Then the current results. “Patient acuity changed.” Daniel nodded. “Significantly.” “Documentation changed too.” Lena looked at her. “How?” “Two years ago we changed fall-event review templates.” Daniel sat forward. “What changed?” “More structured contributing-factor fields.” “Would that affect the risk model?” “Maybe.” Lena waited. Samantha sighed. “Possibly.” Jonah smiled. “She’s learning.” Samantha ignored him. “If the staffing model imports those fields, the distribution could shift even if patient behavior didn’t.” Daniel opened the interface specification. Carla looked at Lena. “Do you know what any of that means?” “Yes.” “That was a mistake. I shouldn’t have asked.” The interface had changed. Not dramatically. Three fields had been consolidated. One renamed. One newly required. The staffing model still accepted the data. No errors. No warnings. No broken imports. Jonah looked at the mapping. “So everything worked.” “Yes,” Samantha said. “And that might be a problem.” “Possibly.” Lena leaned toward the screen. “Does the model know the source representation changed?” “No.” “Does Finance?” Daniel answered. “No.” “Does clinical quality know Finance uses it?” Carla answered. “I didn’t.” Silence. Nobody had made a mistake. That made it more interesting. Daniel stood and walked to the whiteboard. He added beneath Lena’s words: EXPECTED RETURN ACTUAL RETURN VARIANCE Then: CHANGED CONDITIONS Lena nodded. He added: CHANGED INPUTS Samantha said through the monitor, “Good.” Then Daniel wrote: CHANGED DECISION? Jonah looked at Lena. She didn’t move. Carla noticed. “Why did everyone stop?” Jonah answered. “Because that’s the first question on the board that isn’t about whether the old model was right.” Daniel looked at it. “What is it about?” Lena answered. “Whether it is still suitable.” They spent the next forty minutes not solving the staffing problem. That was deliberate. They identified what they did not know. How much of the sitter-cost increase came from wage changes? How much from increased acuity? Had patient mix shifted? Had the relationship between sitter coverage and injury severity changed? Did the altered clinical documentation affect the model’s inputs? Were there unit-specific differences hidden by the hospital- wide aggregate? Carla looked at the growing list. “This is getting bigger.” “Yes,” Lena said. “That usually worries me.” “It should.” Jonah looked at her. “Too broad?” “Not yet.” Carla sighed. “How do you know?” Lena pointed at the questions. “Each one can change the decision.” Daniel nodded. “Material variables.” “Yes.” Jonah pointed to another question he had written: Does overnight sitter presence reduce total falls or primarily injury severity? “This one?” Daniel considered. “Yes.” “Why?” “If the return is primarily reduced severity, then total fall count is the wrong primary measure for evaluating the program.” Jonah nodded. “And if it doesn’t?” “Then the original measure may still be useful.” Lena looked at the board. They were not collecting everything. They were collecting what could change the decision. Different boundary. Same discipline. At 1:08, Daniel ordered lunch. Carla looked surprised. “You’re keeping them?” Daniel looked at Lena. “I want to know whether I’m about to cut something that’s working because I’m measuring the wrong return.” Lena nodded. Jonah looked at Carla. “That seems reasonable.” Carla pointed at him. “You’re enjoying this.” “A little.” Daniel opened a new spreadsheet. “Let’s start with wages.” Lena stopped him. “No.” He looked up. “What?” “Before you calculate anything.” She pointed at the board. “What decision are you actually authorized to make?” Daniel stared at her. Then at Carla. Carla stared back. “I don’t know,” he said. Jonah slowly put down his pen. Daniel opened the budget directive. They read it together. The language was simple: Each department shall identify opportunities to reduce variable labor expenditure by 5% while maintaining established quality and safety targets. Jonah read it again. “You aren’t required to reduce sitter hours.” Daniel shook his head. “No.” “You’re required to identify opportunities to reduce variable labor.” “Yes.” Carla leaned forward. “So why sitters?” Daniel looked at the staffing report. “Because they’re one of our fastest-growing variable labor categories.” Lena nodded. Reasonable. Then: “Fastest-growing by percentage or dollars?” Daniel stopped. Carla looked at the ceiling. Jonah whispered, “Here we go.” Daniel opened another report. Agency nursing. Overtime. Sitters. Temporary clerical coverage. Transport. The largest percentage increase was sitters. The largest dollar increase was agency nursing. By a lot. Daniel stared at the screen. Carla said nothing. Jonah said nothing. Lena waited. Daniel finally spoke. “We sorted the report by percentage.” “Yes.” “And started reviewing from the top.” “Yes.” He looked at her. “That isn’t necessarily wrong.” “No.” “But if the objective is five percent of variable labor expenditure—” “Then percentage growth and absolute expenditure answer different questions.” Daniel leaned back. “Right.” Nobody laughed. At 1:26, the staffing question changed. Not because anyone discovered new data. The data had been there. The representation had been there. The directive had been there. The error—if it was an error—was smaller than that. They had begun with: Which growing labor category should we reduce? The directive had actually asked: Where can variable labor expenditure be reduced while maintaining safety and quality? Jonah looked at both sentences on the board. “Those feel almost identical.” “Yes,” Lena said. “But they aren’t.” “No.” Daniel stared at them. “The first assumes reduction begins with growth.” “Yes.” “The second doesn’t.” “No.” Carla looked at Lena. “So what now?” Lena looked at the board. Expected return. Actual return. Variance. Changed conditions. Changed inputs. Changed decision. Then the two questions. “We don’t evaluate the sitter program yet.” Daniel nodded. “We evaluate whether it belongs in the candidate set.” “Yes.” “And if it does?” “Then we evaluate it against the correct return.” Jonah added, “Which we haven’t defined.” Daniel looked at him. “Not completely.” Carla laughed. “What?” “Nothing.” “No, tell me.” She shook her head. “I thought you came down here to ask whether a dashboard was being misused.” Lena looked at her. “We did.” “And now?” Jonah looked at the whiteboard. “Now we know we were asking the wrong question.” Lena considered that. “No.” Jonah looked at her. She pointed to the original dashboard. “It was the right question to get us here.” He smiled. “Different?” “More precise.” “I’ll take it.” At 1:41, they left Finance. Daniel remained behind with a new candidate-analysis request and a much shorter list of questions than the one they’d almost built. In the elevator, Jonah leaned against the wall. “You know what’s bothering me?” “Yes.” “Stop doing that.” Lena waited. He continued. “Nothing failed.” “No.” “The dashboard worked.” “Yes.” “The import worked.” “Yes.” “The staffing model worked.” “Yes.” “The fall report worked.” “Yes.” “The budget directive was clear.” “Reasonably.” “And we still almost evaluated the wrong intervention against the wrong return.” “Yes.” The elevator descended. Jonah looked at the numbers above the door. “So where was the failure?” Lena thought about it. Not because she lacked an answer. Because she had several. “The handoff.” “Between systems?” “Between questions.” Jonah looked at her. She continued. “Each representation remained usable long enough to reach the next decision.” “But nobody checked whether the next decision was the one it had been built to support.” “Yes.” The doors opened. Neither moved immediately. Jonah looked back at her. “What do we call that?” Lena stepped into the hallway. “We don’t.” He followed. “Why not?” “Not yet.” “Because?” She kept walking. “We have one example.” Jonah smiled. “Lucky guess?” Lena stopped. Turned. He looked entirely too pleased with himself. “That doesn’t mean what you think it means.” “It means we need another case.”