New Section 4.5 proves that completing old tasks is actively punished
by the unweighted mean: a single 26-day-old task hurts the average
more than 26 one-day tasks help it (same total wait resolved, worse
metric). The rational response is not starvation (Theorem 3) but
abandonment — closing aged tasks as "won't fix" to protect the average.
Changes:
- New Section 4.5 with Theorem 6.1 and Corollary 6.2
- Old Section 4.5 (Compound Effect) renumbered to 4.6, table updated
- Conclusion updated with new item 3, subsequent items renumbered
- Edition 1 backed up to .backup/README.md.v1
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Major restructure into four parts with clear argumentative arc:
Part I: Mathematical Foundation (Theorems 1-7)
Part II: Priority Systems (Theorems 8-11, IT example)
Part III: Organizational Dynamics (info asymmetry, psychology, manager strategy)
Part IV: Assessment (devil's advocate, related work, conclusion)
Structural changes:
- Added Section 1 (Introduction) framing the contribution
- Promoted Appendices A/B to full Sections 7/8 (load-bearing content)
- Merged Little's Law as a remark in Section 3.2 (was a detour)
- Merged "When Valid" into Devil's Advocate Section 10.5
- Added Section 11 (Related Work) situating the paper
- Cleaned up "Hmm" and "Wait" language in Theorems 11/WSJF
- Renumbered all sections and cross-references
- Net reduction of 400 lines while adding new content
New citations [18-27]:
- Austin (1996) - measurement dysfunction (most important predecessor)
- Muller (2018) - The Tyranny of Metrics
- Coffman/Shanthikumar/Yao (1992) - conservation laws in scheduling
- Angel/Bampis/Pascual (2008) - SPT fairness criteria
- Bansal/Harchol-Balter (2001) - SRPT unfairness
- Wierman/Harchol-Balter (2003) - fairness classifications
- Campbell (1979) - Campbell's Law
- Ferreira et al. (2024) - moral injury in business
- Bevan/Hood (2006) - gaming in public health
- Moore (2012) - moral disengagement (complementary to our argument)
Citations woven into body: Austin referenced in Sections 4.1, 5.3;
scheduling fairness papers in Section 4.2 note; Campbell/Muller in
Section 7.4; moral injury extension in Section 8.4; all contextualized
in Related Work Section 11.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
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) <noreply@anthropic.com>
Covers all sources referenced or relied upon in the paper:
Scheduling theory: Smith (1956) for SPT/WSJF/exchange argument,
Conway/Maxwell/Miller (1967) for scheduling textbook, Little (1961,
2011) for queueing law, Reinertsen (2009) for WSJF terminology.
Measurement/incentives: Goodhart (1984) and Strathern (1997) for
Goodhart's Law and its generalization.
Behavioral economics: Kahneman & Tversky (1979) for loss aversion.
Game theory: Akerlof (1970) for information asymmetry/adverse
selection, Holmstrom (1979) for moral hazard.
Psychology: Festinger (1957) for cognitive dissonance, Deci & Ryan
(1985) and Ryan & Deci (2000) for Self-Determination Theory,
Seligman & Maier (1967) and Seligman (1975) for learned helplessness,
Shay (1994) and Litz et al. (2009) for moral injury.
Each citation includes DOI where available, ISBN for books, and a
brief annotation mapping it to where it is used in the paper.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Explores what happens when team members understand the proof but are
required to optimize the synthetic metric anyway. Draws on established
psychology frameworks:
- Cognitive dissonance (Festinger): proof eliminates the ambiguity
that would normally provide rationalization cover
- Self-Determination Theory (Deci & Ryan): all three intrinsic needs
(autonomy, competence, relatedness) are violated by awareness
- Moral injury (Shay, Litz): structural conditions met when team
knowingly deprioritizes critical work for metric optimization
- Learned helplessness (Seligman): repeated failed advocacy produces
metric fatalism and disengagement
Derives the adversarial selection spiral: the metric selects against
competent team members who recognize its flaws and for those who
don't, producing invisible competence degradation that the metric
itself cannot detect.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Explores the case where the unweighted mean is reported directly to the
client, making the metric itself the source of satisfaction. Under this
model the entire paper's conclusion inverts: SPT genuinely maximizes
client satisfaction at zero marginal cost.
Analyzes this as a moral hazard / pooling equilibrium using game theory,
identifies three fragility conditions (client inspects own ticket,
competitor offers per-ticket SLAs, team internalizes the metric), and
maps the pattern across domains (education, healthcare, finance, software).
Concludes: the incentive exists, the equilibrium is real, and it holds
until it doesn't.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Corrections:
- Theorem 4: Restated from "maximizes slowdown inequality" (wrong) to
"uniquely assigns max completion time to largest task" (correct).
SPT actually compresses slowdown variance; harm is in absolute delay.
- Theorem 5: Completely rewritten. Old claim that LPT minimizes slowdown
variance was backwards (verified: tasks [1,5,10] give SPT var=0.06,
LPT var=42.2). New theorem correctly states SPT concentrates absolute
delay on the largest task.
- Theorem 10: Removed draft language ("Wait —"), corrected cross-term
analysis. Old claim that SPT is Pareto-dominated when p_H > 8p_L was
wrong (verified: n_H=2,n_L=2,p_H=10,p_L=1 gives D_SPT=275 < D_pri=283).
Replaced with correct WSJF exchange argument.
- IT example: Fixed PWCT arithmetic (9.225→10.2, 6.633→10.167). Added
honest discussion that aggregate PWCT fails to distinguish schedules;
per-priority-class metrics are needed.
- Section 5: Added caveat that Little's Law batch-case application is
not straightforward; clarified what Theorem 2 actually proves.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Sections 9-11: Prove that unweighted mean completion time becomes
adversarial under priority classification (Theorems 8-10), propose
PWCT/WSJF as alternatives with a worked IT service desk example,
and present honest counterarguments establishing the narrow conditions
under which the unweighted metric remains defensible.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Sections 7-8: Prove that optimizing unweighted mean completion time
maximizes slowdown inequality (Theorem 4), maximizes satisfaction
variance across clients (Theorem 5), provides zero throughput gain
(Theorem 6), and therefore simultaneously degrades client experience
while failing to improve productivity (Theorem 7).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Mathematical proof that unweighted average task completion time
is gameable by scheduling policy (SPT), while work-weighted
completion time is schedule-invariant. Demonstrates that SPT's
apparent advantage is an artifact of the metric, not genuine
throughput improvement.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Mortdecai