Thesis-by-Publication & Multi-Level Progress Tracking — digest

​

Part A — Thesis-by-publication & PhD milestones

• The model. Traditional monograph vs thesis by publication / cumulative / “stapler” (papers bound by a framing text). The cumulative model is the papers→thesis nesting → treat monograph as the degenerate case (one “paper” = the whole thesis), not a separate mode.
• N-paper convention. No universal N; most-cited concrete figure is Swedish medicine 3–5 articles (Karolinska). The binding rule is scope, not count (“total scope corresponding to four years full-time”). → make paper-count a soft, configurable target, never a hard gate.
• The framing chapter (“kappa”). Swedish for “overcoat” — the synthesizing chapter turning a stack of papers into one thesis; more than concatenated introductions (Stockholm Univ Kappa Guidelines; Karolinska; Lund). Components (reusable as the thesis-artifact schema): introduction/background; unifying narrative/aims; extensive independent related work; per-paper summary + explicit statement of the candidate’s contribution to co-authored papers; comprehensive concluding discussion; future work. Introduces no new findings — it recontextualizes.
• Coherence / “original contribution to knowledge.” The examinable through-line (Oxford MPLS integrated-thesis guidance; QUT; UNE): examiners judge (a) coherence of the whole, (b) how each paper contributes to the overarching project, (c) that it reads as a single coherent document. → thesis→papers roll-up target is narrative coverage of the aims + kappa through-line, not “N papers done.” Surface gaps in the argument, not counts.
• Milestones/gates (UC Davis, Berkeley, Michigan, McGill handbooks): proposal/prospectus → committee formation → qualifying/comprehensive exam → advancement to candidacy (pivotal) → annual progress review → submission → defense/viva. Track per milestone: status (not-started/scheduled/passed), date, next binding deadline. Institution-specific + time-based → a small configurable milestone list at the thesis level, not hard-coded.

​

Part B — Multi-level progress tracking (core design input)

• Stage-Gate / phase-gate (Cooper) — closest match; each stage ends at a gate with explicit go/kill/hold criteria + required deliverables. Maps to our resolve gates: hypothesis (evidence → verdict), paper (submission-ready), thesis (defensible). Borrow “gate = exit criteria + decision” — more honest than % burndown.
• OKRs — the hierarchical roll-up pattern (parent progress from children); adapt cautiously (research outcomes are binary-ish, not KPI-measurable).
• Avoid burndown/velocity — they assume known decomposable scope, which pre-verdict research lacks.

• Hypothesis verdicts → paper readiness: not “80% at 4/5 resolved”; a single refuted load-bearing hypothesis can invalidate the claim. Roll up as coverage + blocker status, not an average.
• Paper states → thesis narrative coverage: map each paper to the aims; thesis status = “uncovered aims?” + “through-line stated?” (exactly what examiners check).
• Keep verdict and readiness as distinct axes — a refuted hypothesis is done, successful science; conflating refuted with failed/incomplete is a modeling error.

• Goodhart’s law / Campbell’s law — a self-tracking PhD tool is especially Goodhart-prone (user sets and games the metric).
• DORA (2012) and the Leiden Manifesto (Hicks & Wouters et al. 2015) — metrics support, never replace, qualitative judgment.
• Design rule — surface, don’t score. Show state + gaps (unresolved hypotheses, uncovered aims, overdue milestones, staleness/last-touched), not an aggregate productivity score.
• Do NOT count: paper/word counts, citation/impact proxies, commit counts, %-complete on unresolved research, hypothesis “success rate” (punishes refutation). Refuted = green/done, not red.

1. status = cross-cutting read verb, not a fourth level. Each artifact owns a small status block in its markdown frontmatter (verdict/readiness/coverage + last-updated). status <level> [id] reads + rolls up — never invents a separate progress artifact to drift.
2. One thin dashboard view aggregating frontmatter into a generated overview markdown (never hand-edited) — a projection of the source of truth, nothing to game or sync.
3. Roll-up = semantic function of children + explicit gate criteria, surfaced as coverage + blockers, never %.
4. Bake in the anti-Goodhart stance (cite Goodhart/Leiden/DORA); refuted = done; document it so it isn’t quietly turned into a score.

​

Sources

​

Flags