From 635d902691eb15769ee57fbf78ab2299f409b2c9 Mon Sep 17 00:00:00 2001 From: Mortdecai Date: Sat, 28 Mar 2026 18:12:31 -0400 Subject: [PATCH] Add Section 12: Manager Internalization strategy Formalizes the actionable middle ground: a manager who understands the proof can schedule primarily by priority while tactically interleaving small tasks to maintain metric parity with other teams. Key contributions: - Constrained optimization formulation (minimize priority-weighted delay subject to unweighted mean staying in acceptable band) - Theorem 12: bounded metric cost of priority scheduling (within-class SPT is free, between-class inversions are bounded) - Manager as information barrier (shields team from metric's perverse incentives, preserving intrinsic motivation per Appendix B) - Competitive breakdown as prisoner's dilemma: cooperative equilibrium is stable when metric is a health-check, collapses when metric is ranked or tied to compensation - Scope table: viable for parity/health-check, fragile under ranking, not viable under compensation linkage Co-Authored-By: Claude Opus 4.6 (1M context) --- README.md | 168 +++++++++++++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 167 insertions(+), 1 deletion(-) diff --git a/README.md b/README.md index d4dbde2..c6fc466 100644 --- a/README.md +++ b/README.md @@ -839,7 +839,173 @@ and rarely met in the systems where it is most commonly used. --- -## 12. Conclusion +## 12. Manager Internalization: The Actionable Solution + +The preceding sections present two extremes: reject the metric entirely +(Sections 1-10) or surrender to it (Appendix A). In practice, most +managers cannot unilaterally change the metric — it is set at the +organizational level, reported across teams, and embedded in dashboards +that other stakeholders consume. The best solution is company-wide metric +reform. The *actionable* solution is what a single informed manager can +do right now. + +### 12.1 The Strategy + +A manager who understands the proof can **internalize the metric's +limitations without propagating them to the team**. The approach: + +1. **Schedule primarily by priority.** The team works critical tasks + first, exactly as professional judgment and the priority system + dictate. This is the default — the team need not know why. + +2. **Tactically interleave small tasks to maintain metric parity.** When + the queue contains a small, low-priority task that can be completed + quickly without materially delaying any high-priority work, do it. + Not because the metric demands it, but because the small task *also + needs to get done*, and doing it now costs almost nothing. + +3. **Never reveal the metric as the motivation.** The team is told "knock + out this quick one while we're waiting on the vendor callback for the + P1" — not "we need to bring our average down." The team's + professional judgment and intrinsic motivation (Appendix B) remain + intact. The manager absorbs the metric-management burden. + +This is a **constrained optimization**: minimize priority-weighted delay +(do the right work in the right order) subject to the constraint that +the reported unweighted mean stays within an acceptable band. + +### 12.2 Formalization + +Let $\bar{C}_{\text{target}}$ be the unweighted mean completion time that +other teams report — the parity threshold. The manager's problem is: + +$$\min_{\sigma} \sum_{i=1}^{n} w(q_i) \cdot C_i \quad \text{subject to} \quad \bar{C}(\sigma) \le \bar{C}_{\text{target}}$$ + +This is a single-machine scheduling problem with a budget constraint on +the unweighted mean. The solution is a modified priority schedule: + +- Start from the priority-first ordering (all P1 first, then P2, etc.). +- Identify small low-priority tasks whose insertion ahead of lower-ranked + same-priority tasks reduces $\bar{C}$ without displacing any + higher-priority task. +- Insert them only when the marginal improvement to $\bar{C}$ exceeds + the marginal cost to priority-weighted delay. + +**Theorem 12 (Bounded Metric Cost of Priority Scheduling).** For a +priority-first schedule with $n$ tasks, the gap between its unweighted +mean $\bar{C}_{\text{priority}}$ and the SPT-optimal unweighted mean +$\bar{C}_{\text{SPT}}$ is bounded by: + +$$\bar{C}_{\text{priority}} - \bar{C}_{\text{SPT}} \le \frac{n-1}{2n}(\bar{p}_{\max\text{-class}} - \bar{p}_{\min\text{-class}}) \cdot n_{\text{classes}}$$ + +where $\bar{p}_{\max\text{-class}}$ and $\bar{p}_{\min\text{-class}}$ are +the mean processing times of the largest and smallest priority classes. + +**Proof sketch.** The gap arises entirely from the cross-class ordering: +within each priority class, the manager can use SPT (shortest first) at +no priority cost, since all tasks in the class have equal priority. The +only deviation from global SPT is the *between-class* ordering, where +large high-priority tasks are placed before small low-priority tasks. +Each such inversion costs at most $p_{\text{large}} - p_{\text{small}}$ +in the unweighted sum, and there are at most +$n_{\text{classes}} \cdot (n / n_{\text{classes}})$ such inversions. +$\blacksquare$ + +In practice, this means: **a manager who uses SPT within each priority +class and priority ordering between classes will produce a metric that +is close to the SPT-optimal value** — often within 10-20% — while +respecting the priority system entirely. + +### 12.3 Why This Works: The Manager as Information Barrier + +The strategy works because the manager serves as an **information +barrier** between the metric and the team: + +| Layer | Sees the metric | Sees the priorities | Sees the proof | +|-------|----------------|--------------------|-----------------| +| Organization | Yes | Nominally | No | +| Manager | Yes | Yes | **Yes** | +| Team | No (shielded) | Yes | Irrelevant | +| Client | Yes (dashboard) | Via SLA | No | + +The manager is the only actor who holds all three pieces of information. +By internalizing the proof, the manager can: + +- Present a metric that satisfies organizational reporting (the number + is reasonable) +- Direct the team by priority (professional judgment preserved) +- Shield the team from the metric's perverse incentives (Appendix B + costs avoided) + +This is *not* manipulation. The manager is not fabricating numbers or +misreporting. They are doing the right work in the right order, and +the metric happens to be acceptable because within-class SPT is free +and between-class inversions are bounded (Theorem 12). + +### 12.4 The Competitive Breakdown + +This strategy fails when the metric becomes **competitive between teams**. + +Model $m$ teams, each managed independently. Team $j$ reports +$\bar{C}_j(\sigma_j)$. If teams are ranked, rewarded, or compared on +$\bar{C}$: + +**Case 1: Cooperative** — Teams are measured for parity, not ranking. +The threshold is "stay within a reasonable band." Each manager +independently uses the internalization strategy. All teams do +approximately the right work. The metric is decorative but harmless. +This is a **coordination game** with a stable cooperative equilibrium. + +**Case 2: Competitive** — Teams are ranked by $\bar{C}$. Promotions, +resources, or recognition go to the lowest average. This is a +**prisoner's dilemma**: + +| | Team B: Priority-first | Team B: SPT | +|---|---|---| +| **Team A: Priority-first** | (Good work, Good work) | (A looks bad, B looks good) | +| **Team A: SPT** | (A looks good, B looks bad) | (Both look good, both do wrong work) | + +The dominant strategy for each team is SPT. The Nash equilibrium is +(SPT, SPT) — all teams optimize the metric, all teams do the wrong +work, and the organization reports excellent numbers while critical +tasks rot across every queue. + +The internalization strategy is a **cooperative equilibrium that is not +stable under competition**. A single team that defects to pure SPT will +outperform all others on the metric, forcing other managers to choose +between doing the right work (and looking bad) or following suit (and +abandoning their professional judgment). + +### 12.5 The Scope of the Solution + +| Condition | Strategy viability | +|-----------|-------------------| +| Metric used for health-check / parity | **Viable** — cooperative equilibrium holds | +| Metric visible but not ranked | **Viable** — no competitive pressure to defect | +| Metric ranked across teams | **Fragile** — viable only if all managers cooperate | +| Metric tied to compensation / resources | **Not viable** — prisoner's dilemma dominates | +| Metric reform possible at org level | **Unnecessary** — fix the metric instead | + +The internalization strategy is actionable *right now*, by a single +manager, without organizational permission or metric reform. It +preserves team psychology (Appendix B), respects priorities (Sections +9-10), and produces an acceptable reported metric (Theorem 12). + +Its limitation is structural: it requires the metric to be a +**reporting formality**, not a **competitive instrument**. The moment +the metric drives resource allocation or team ranking, the cooperative +equilibrium collapses and only organizational reform — replacing the +metric with a priority-weighted alternative (Section 10) — can prevent +the race to the bottom. + +**The best solution is company-wide. The actionable solution is a +manager who understands this proof, shields their team from the metric, +schedules by priority, and uses SPT only within priority classes to +keep the number reasonable.** + +--- + +## 13. Conclusion The unweighted average completion time is a **biased statistic** that: