Thesis Quality Gate

Thesis Quality Gate

Use the deterministic CLI in this Skill as the control plane for thesis state and delivery readiness. Keep project evidence, decisions, reports, and private data in the thesis project; keep only reusable code and sanitized templates in this Skill.

Locate the project

Run commands from the thesis project root, which must contain .project/. Prefer the project-local launcher when present:

python3 scripts/thesis_quality.py state refresh

Otherwise invoke the Skill entry directly:

python3 <skill-path>/scripts/thesis_quality.py --project <project-root> state refresh

Read references/project-contract.md before onboarding a new project or changing paths and schemas.

Operate the quality loop

1. Refresh state before relying on branch, commit, version, page count, stage, risk, or last-build information.
2. Refresh the numeric ledger after changing thesis numbers. Treat automatic classification as provisional; never auto-mark evidence as verified.
3. Sync the literature registry after changing ref.bib. Mark a source verified only after checking a real bibliographic source.
4. Run Doctor after meaningful source, environment, build, or delivery changes.
5. Before formal delivery, require confirmed research anchors, an accepted current-stage artifact contract, a closed revision ledger, a current deterministic de-AI report, and an isolated AI review report.
6. Run the gate matching the requested delivery stage. A blocked gate is a result to explain and resolve, not an instruction to weaken policy. Interpret the machine decision as PASS, REOPEN_WRITING, or REOPEN_ANALYSIS; when both kinds fail, analysis takes precedence while writing failures remain listed.

python3 scripts/thesis_quality.py evidence refresh
python3 scripts/thesis_quality.py evidence validate
python3 scripts/thesis_quality.py literature sync
python3 scripts/thesis_quality.py doctor
python3 scripts/thesis_quality.py gate --mode predefense

Use --json on evidence, literature, Doctor, and gate commands when another tool needs machine-readable output.

Preserve validation integrity

• Never invent raw data, calculation scripts, enterprise records, citation verification, or author confirmation.
• Preserve reviewed ledger entries when rescanning; retire reviewed entries that disappear from the manuscript instead of silently deleting their audit history.
• Keep human-controlled state fields intact when refreshing machine-derived fields.
• Treat repository HEAD as an observed snapshot; bind manuscript artifacts and reviews to the stable manuscript_commit plus the current source-tree digest.
• Reuse an existing Doctor only when engine version, Skill commit, script digest, source tree, policy, evidence, literature, PDF, and DOCX provenance all match.
• Treat PDF parsing or OCR failure as a failed or skipped check; never reconstruct missing content from inference.
• Do not copy project data, company information, or absolute user paths into this Skill.
• Do not change gate policy merely to obtain PASS.

Read references/evidence-ledger.md when classifying evidence. Read references/gate-policy.md when adding checks, changing gate modes, or interpreting a blocked delivery. Read references/stage-artifact-contracts.md when binding a stage handoff to exact inputs, outputs, hashes, acceptance checks, and reopen route.

Integrate with other thesis Skills

• Let thesis-orchestrator decide when to call this Skill; keep check implementation here.
• After thesis-analyze changes statistics, refresh and validate the evidence ledger.
• After thesis-literature verifies sources, update and validate the literature registry.
• After mem-thesis-build or thesis-docx creates artifacts, refresh state and run Doctor.
• Before thesis-review approves delivery, run the appropriate gate and cite its report.

Do not duplicate the scripts inside those Skills. Add only short routing hooks and commands.