Methods that recur
across projects.

What it is. A staged compression exercise. Start with an unfiltered paragraph answering "tell me about yourself." Cut it to one sentence. Cut the sentence to one word. Each cut forces a choice: what you keep is the signal.

Why it works. Compression-via-deletion is cognitively cheaper than synthesis-from-scratch. Starting with too much and carving away is easier than starting from a blank page — and the carving is where the thinking happens.

Run it in pairs. Solo compression produces vague abstractions ("Leader", "Driven"). Compression in dyads — where you have to explain each cut to a partner — produces specific, personal language ("Translator", "Connector", "Witness"). The peer is not a nice-to-have. See field note.

Failure modes. (1) Students cheating toward job titles ("Engineer", "Consultant"). Block these up front. (2) Single-word results that are too abstract ("Leadership") — push for the next cut. (3) Insisting on the word as a permanent identity rather than a working hypothesis.

What it is. Three card-sort exercises run in sequence — values, strengths, skills — each with a forced top-N. Adapted from established instruments: Knowdell Career Values, CliftonStrengths, and Motivated Skills. The cards are physical, not digital.

Why three. Any single instrument produces a flattering self-portrait. Three instruments triangulate — and the disagreements between them are more useful than the agreements. A student who picks "Creativity" as a top value but doesn't have it in their top skills is in the productive territory.

Why physical. Handling, sorting, and repositioning cards changes the quality of attention. Drag-and-drop on a screen is faster, but slower thinking is the point.

Failure modes. (1) Free-sort with no forced top-N (students pick "everything matters"). Always force a top 5 → top 2. (2) Reading the cards as a personality test instead of a discussion prompt.

What it is. An LLM layer that compares what a student says about themselves (their compressed paragraph, their card-sort choices) against external evidence (their resume, prior writing). It surfaces the gap. It does not resolve it.

The crucial framing. The model is not a judge — it's a witness. A student who picks "Creator" but has a resume full of operations work doesn't get told they're wrong. They get told "your card sorts suggest Creator; your resume suggests Organizer; what's that gap about?" That question is where reflection happens.

Output style matters. Confident, declarative outputs ("You are a…") shut down reflection. Tentative, evidence-cited outputs ("Your resume foregrounds X; you self-identified as Y; the difference might be…") open it.

Failure modes. (1) Model agreement = no friction = no reflection. Tune for productive disagreement. (2) Students treating the AI as authoritative rather than provisional. Surface the model's limits explicitly in facilitation.

What it is. Present results (occupations, possibilities, options) as a spatial layout where similar items sit near each other, instead of a ranked list sorted by fit score.

Why it works. Ranked lists ask "which is best?" and students stop at the top five. Spatial maps ask "what's near what?" and students start moving. The behavior we want is exploration, and the format determines the behavior.

What "similar" means. In Synapse: 13 dimensions of working preference, with adjacency computed by distance in profile space. In DartWorld: clustered by occupational family and shared career narratives. The specific similarity metric is less important than that nearness is meaningful.

Failure modes. (1) A map that looks like a map but functions like a list (top match highlighted, others greyed out). (2) Too many dimensions — the map becomes random. Aim for 8–15 meaningful axes. (3) Ignoring the adjacency information in the UI — make it the primary signal, not a side panel.

What it is. A short reflective onboarding (~9 questions) that clusters a respondent into a small number of named archetypes (originally Explorer / Seeker / Achiever in DartWorld). The archetype is a starting frame for the session, not a personality verdict.

Why few. Five labels produce ambiguous self-identification ("I'm sort of two of those"). Three labels force a clearer choice and produce stronger downstream engagement.

Why archetypes break. The Achiever label, in our case, produced sustained pushback from the students it was supposed to capture. Field note. Lesson: archetypes carry connotations the designer can't fully control. Replace them with elicitation prompts when the label becomes the friction.

Failure modes. (1) Treating the archetype as a result rather than a starting point. (2) Labels that match cultural stereotypes too closely (the "Achiever" problem). (3) Too many archetypes (≥5) producing low-confidence self-placement.

What it is. A persistent journaling interface where entries from previous sessions remain visible — not hidden behind a date filter. Each new entry sits next to its predecessors, so the student can see (and reference) what they thought last term.

The threshold effect. Single entries are surface-level — "what I'm thinking about this week." Three or more entries from the same student begin to reference earlier entries, name contradictions, and track shifting priorities. That referential behavior is the signal that longitudinal reflection is happening.

Why a pinboard. A chronological feed encourages "what's new"; a pinboard encourages "what's still there." The spatial layout is doing cognitive work — same logic as M04.

Failure modes. (1) Reset-style prompts ("What's new today?") that don't reference prior entries. (2) Hiding old entries behind a "history" view. (3) Encouraging completion or streak metrics, which produce ritual posting without reflection.

What it is. Structured peer conversation — usually dyads, occasionally triads — as the engine that turns a private reflection exercise into a productive one. The conversation isn't a debrief; it's where the work happens.

Why. Saying something out loud to another human forces clarity that internal monologue never demands. The student is not just reporting — they're constructing the thought in real time, in response to a face.

Pairs > groups. Groups of 4+ produce one-speaker-at-a-time dynamics that suppress the people who most need to talk. Dyads force participation. Triads add a witness — useful for some prompts, distracting for others.

Failure modes. (1) "Share with your partner" with no constraint — produces small talk. Give explicit prompts and time limits. (2) Group sizes ≥ 4. (3) Letting facilitators substitute for the peer — kills the mechanism.

What it is. Using the existing studio-critique format that architecture, design, and art students are already fluent in — but pointing it at career decisions instead of design projects. The student "presents" a career direction, dilemma, or option to a peer group, and the group critiques it.

Why it works. Architecture students learn through critique. Importing an unfamiliar reflection framework requires learning the framework first. Repurposing the framework they already use removes the activation cost.

What changes. The vocabulary: "What's the parti of this career direction?", "Where's the structure?", "What's the relationship between figure and ground?" The questions are familiar; the object is novel.

Failure modes. (1) Importing pedagogical baggage along with the format (e.g., harsh critique culture). (2) Forcing the metaphor where it doesn't fit. (3) Working in disciplines without a critique tradition — this method doesn't transfer cleanly to STEM fields without a parallel ritual.

What it is. Forced either/or choices instead of Likert scales or sliders. "Which energizes you more: manifesting or deciphering?" "Are you more driven by money or meaning?"

Why binaries. Likert scales produce middle clustering — most respondents pick 3 or 4 out of 5. That's not modesty; it's noise. Binaries force the respondent to commit, and the friction of committing is where the self-knowledge happens.

Why "wrong" prompts are productive. A binary that almost-but-not-quite fits produces better reflection than one that fits perfectly. The student thinks "that's not exactly me — I'm more…" and is now articulating their actual position.

Failure modes. (1) False dichotomies that respondents resent ("Are you more left-brained or right-brained?"). Pick binaries that feel like real choices. (2) Too many in a row — fatigue sets in around 15–20. Cap the instrument.