Skip to main content
Thesis is the optional top level of the three-level mirror (../../docs/design/00-meta-spec.md §3.1): thesis ⊃ papers ⊃ hypotheses. A repo that is not a doctoral thesis simply omits it — its top is the portfolio. Where it exists, the thesis binds a subset of the portfolio’s papers into one examinable document via a synthesizing framing chapter (the Nordic “kappa”), plus the appended papers themselves. It is deliberately a partial mirror (ADR-0005, ../../decisions/0005-three-level-mirror.md). The lower levels have a generate/resolve pair driving a high-throughput flywheel; a thesis has no such flywheel — there is one thesis, framed once and refined. So framing (the degenerate generate) and synthesis (the resolve) are one thesis skill, not two. The roll-up target is narrative coverage of the aims, never a paper count (there is no universal N; the binding norm is scope). A monograph is the degenerate case: one paper spanning the whole thesis.

​
When to use

  • framing (occasional). Establish or refine the thesis aims and narrative through-line, and decide which portfolio papers compose the thesis. Run this once early (a prospectus/proposal), then only when the scope shifts.
  • synthesis (the resolve). Assemble the kappa framing chapter over the chosen papers, then clear the defensibility gate. Run as submission and defense approach.
  • Never, for a repo that is not a thesis. The top level is optional; do not scaffold a thesis/ folder unless the author is writing a doctoral thesis.
This skill does not create new scientific findings, adjudicate paper quality, or aggregate across repositories (cross-repo aggregation is out of scope, ../../docs/design/00-meta-spec.md §1).

​
Modes

​
framing (occasional)

Define the aims — the small set of overarching questions/objectives the thesis answers — and the narrative through-line that makes the papers read as one project rather than a stapled stack. Then map which portfolio papers compose the thesis and which aim(s) each supports.
  • Lives in docs/research/thesis/aims.md: the aims, the through-line, and the chapter↔paper map (aim → supporting paper-ids drawn from the papers.md registry).
  • Proposes and records the author’s framing; it does not select papers on the author’s behalf. Which work belongs in the thesis is a scoping judgment the author makes and signs.
  • Refined, not churned. Re-run when the program’s scope genuinely shifts (a new aim, a paper dropped from scope), not per paper.

​
synthesis (the resolve)

Assemble the kappa (see below) over the framed papers, then run the defensibility gate. This mode reads the portfolio state and the aims, drafts the framing chapter, surfaces coverage gaps, and stages the mock viva — the author authors the prose and decides defensibility. The full thesis-level staged progression spans both modes — the first three are framing outputs, the last is this mode: prospectus → aims/narrative → chapter↔paper map → kappa + defensibility.

​
The kappa

The kappa (Swedish “overcoat”) is the synthesizing framing chapter that turns a stack of papers into one thesis. It is more than concatenated introductions and introduces no new findings — it recontextualizes existing, already-resolved work. Draft it in docs/research/thesis/kappa/. Components (the artifact schema): Then the appended papers themselves. The per-paper contribution statement is load-bearing for co-authored papers (authorship/credit norms) and is the author’s to write — the skill drafts a scaffold, never asserts the split. Hard constraint: no new findings in the kappa. If synthesis would require a result the papers do not already establish, that is a hypothesis for hypothesis-testing inside a paper, not kappa prose.

​
Milestones

Program milestones are a small, configurable, time-based list in docs/research/thesis/milestones.yml — institution gates vary and are deadline-driven, so they are never hard-coded. A typical progression:
Track per milestone: status (not-started / scheduled / passed), date, and the next binding deadline. These are calendar gates, distinct from the coverage/defensibility state below — progress surfaces both. Milestones are reported, never scored: an overdue milestone is a surfaced gap, not a penalty (anti-Goodhart, ../../docs/design/00-meta-spec.md §3.6).

​
Defensibility gate

The thesis is done when it is defensible: every aim is covered by ≥1 paper and the kappa states a coherent through-line + the original contribution. The roll-up target is narrative coverage of the aims (does each aim have supporting papers; is the through-line stated) — not a paper count, never a percentage or average (../../resources/references/thesis-and-progress-tracking.md, Part A). Defensibility is a material human decision, recorded with a named human sign-off (../../docs/design/00-meta-spec.md §2.1; lifecycle §8). Before it is recorded, the defend skill runs a full MOCK VIVA (../defend/SKILL.md; ../../docs/design/01-lifecycle.md §6) across all three targets (claim | cited-work | methodology). This gate is non-blocking but escalated (ADR-0021, ../../decisions/0021-thesis-gate-per-gap-confirmation.md). As the highest-stakes, least-reversible decision, it replaces the single blanket override used at other gates with per-gap acknowledged sign-off: the author must explicitly acknowledge each surfaced gap in writing before proceeding.
  • The AI never adjudicates a gap as “critical” and never hard-blocks — it surfaces gaps as observed facts; the human acknowledges each and decides.
  • Waving a thesis through with known gaps is therefore deliberate and fully on record, per gap.

​
Composition

  • Reads down — the papers.md registry and each paper’s decision/status frontmatter (which papers are done), and portfolio-backlog.md. It consumes resolved papers; it does not resolve them.
  • literature position --level thesis — for the kappa’s extensive independent related-works synthesis (thesis scope, unioned/deduplicated across every aim and paper, broader than a single paper’s positioning).
  • defend — the mock viva at the defensibility gate; also self-invokable on any aim or contribution statement.
  • progress — reads the thesis status block in aims.md (readiness / defensibility + the aim list + last-updated) and rolls it into dashboard.md as a pure projection. Thesis status = uncovered-aims list + through-line-stated flag + milestone state, never a score.
  • The engineering backend — thesis prose is authored, not engineered, so there is nothing to delegate at this level (unlike a paper’s outline/plan).

​
Guardrails

Hard rules, not preferences.
  • Optional top level. Do not scaffold or invoke thesis for a non-thesis repo. Its absence is the normal case; the portfolio is the top.
  • No new findings. The kappa recontextualizes resolved work only. Any needed new result routes back to hypothesis-testing within a paper.
  • Coverage, not count. Report narrative coverage of the aims and the through-line; never a paper count, %-complete, or productivity score. No universal N — the binding norm is scope.
  • Author authors, skill drafts. The aims, the scope (which papers compose the thesis), the per-paper contribution statements, and the defensibility judgment are the author’s. The skill drafts scaffolds and surfaces gaps.
  • Defensibility is a signed human decision. The mock viva must fire first; the gate is non-blocking with per-gap acknowledged sign-off (ADR-0021). The AI never grades a novel claim’s substance or decides a gap is “critical.”
  • No cross-repo aggregation. Papers living in a separate repo are out of scope (../../docs/design/00-meta-spec.md §1); each repo’s top level is self-contained. A possible future extension is sketched in ../../docs/design/proposals/cross-repo-thesis-aggregation.md.

​
Commit attribution

When you commit artifacts produced by this skill, add these git trailers — discovery + provenance (see ../../resources/commit-attribution.md):