reflection-analyser

Reflective-writing analysis — the lens-family member that reads a learning journal / reflection / portfolio entry as reflection, not just as prose.

document-analyser reads readability; this reads reflective depth. Different signals from the same words. Explicit-only (auto_routable: false) — same pattern as conversation-analyser: text and prose extensions auto-route to document-analyser; invoke reflection-analyser deliberately when you want the reflective-depth interpretation.

Built around the markers commonly used in reflective-writing rubrics (Moon's depth scale, Gibbs' reflective cycle, the SOLO taxonomy): metacognition, criticality, evidence linkage, affect language, and forward-looking action.

Install

pip install reflection-analyser

Optional: read .pdf / .docx / .pptx journals (otherwise plain-text / .md only):

pip install 'reflection-analyser[documents]'

Use

Python:

from reflection_analyser import ReflectionAnalyser

# From text directly
result = ReflectionAnalyser().analyse_text("Looking back, I realised that…")

# From a file (composes on document-analyser for binary docs when [documents] is installed)
result = ReflectionAnalyser().analyse("journal.md")
result = ReflectionAnalyser().analyse("journal.docx")  # requires [documents]

print(result.depth_band)               # "dialogic"
print(result.composite_depth_score)    # 0.62
print(result.metacognition.count)      # 7
print(result.criticality.count)        # 3

CLI:

reflection-analyser journal.md
reflection-analyser journal.txt --json
reflection-analyser journal.docx                 # needs [documents] extra
echo "Looking back…" | reflection-analyser -
reflection-analyser serve
reflection-analyser manifest

HTTP (reflection-analyser serve on port 8015):

curl -F file=@journal.md http://localhost:8015/analyse

Signals

For a piece of reflective writing:

• Metacognition — first-person + cognitive verbs (I realised, I noticed, looking back, on reflection). Surface depth indicator.
• Criticality — contrast/qualification phrases (however, in contrast, that said, on the other hand). Marker of dialogic vs descriptive reflection.
• Evidence — references to specific moments, sources, dates, quotes — proper-noun and citation density. Concrete vs abstract.
• Affect — emotion words (frustrated, surprised, confident, uncertain). Too few = clinical; presence indicates engagement.
• Action / forward-looking — next time, going forward, I will, future-tense intent. Marker of transformative reflection.

Composite depth score (0–1) combines per-marker coverages; mapped to a Moon-style band:

Band	Score	Description
descriptive	0.0–0.25	"What happened" only — events recounted, little interpretation
dialogic	0.25–0.5	Some self-questioning + critical thought
critical	0.5–0.75	Multiple perspectives, evidence linkage, qualification
transformative	0.75–1.0	Forward-looking insight, evidence-linked, change-oriented

The score is a signal, not a grade — it's meant to inform human judgement, not replace it.

The family

What you want	Use
Document text + readability	document-analyser
Reflective depth on that text	reflection-analyser (this)
Human-AI conversation analysis	conversation-analyser
Any file → right engine	auto-analyser

Limits

• Lexicon-based v1 — fast, transparent, but catches phrasing not meaning. A reflective sentence without our trigger words underscores; a non-reflective sentence with I realised overscores.
• English-only for v1.
• Calibrated against generic reflective-writing rubrics; tune the band thresholds in _BAND_THRESHOLDS for your unit's specific rubric if needed.
• Vision/LLM-augmented depth scoring is a possible follow-on; not in v1.

License

MIT