Shape

Description

The abstract structural form of a problem or system, independent of surface content. “This has the shape of a scheduling problem” is often more useful than the explicit taxonomy of scheduling problems — it captures the essential structural relationships without requiring committal to a specific category. Shape is the recognition mechanism that lets cross-domain analogies feel right before they can be articulated.

Encounters

• “The shape of the feedback” — KCC parent-feedback work: the structural form of complaints reveals what’s really going on under surface variations.
• “The shape of the task drives the right context-management choice” — agent orchestration: structural form (focused vs. expansive, single-pass vs. iterative) decides whether elephant-goldfish dispatch fits.
• “Different sentences but consistent shape” — replicated-phrase doctrine: when a sentence shows up at multiple abstraction levels with consistent shape, it’s a sign of well-carved underlying structure.
• “Same shape as X” — diagnostic move: recognizing a new problem has the structural form of a familiar one.

When it applies / triggers on

User-initiated: User describes a concrete situation — a UI behavior, a code snippet, a domain problem, a proposed feature — and the agent reaches for shape to extract structure. Unlike load-bearing, the trigger is rarely a single verb; the lexical signal “like” appears in 28% (7/25) of real-user-msg shape triggers — usually as the user gestures at an existing pattern (“is this like the…”, “kind of like X”). Three recurring sub-shapes:

• Cross-domain recognition — user describes a domain-X problem; agent maps it to a structural class with same-shape instances from domains Y, Z. (Example: “another way we can substantially cheapen source routing is upstream. Provider pruning.” → agent recognizes a negative-result suppression list with re-evaluation policy, citing DNS negative caching, malware blocklists, CDN no-route lists as same-shape.)
• Same-shape-across-surfaces — user surfaces a specific decision; agent recognizes it as an instance of a pattern the codebase has already absorbed and applies the doctrine where it earns its keep. (Example: discussion of inline-text affordances vs 44px hit targets → “this is the same shape as several patterns the codebase has already absorbed.“)
• Hidden-shape diagnostic — user describes confusion or friction with an existing abstraction; agent identifies that the abstraction is hiding the real shape, refactor cue follows. (Example: “we’re bundling a bunch of stuff under the extractor interface … turning Extractors inward back on the products of other extractors” → agent: “the abstraction is hiding the real shapes rather than revealing them.“)

Agent-initiated: Engine notices the target situation is being framed in a way that may obscure (or misidentify) its structural form. Candidate inference: “what’s the structural shape of this independent of surface content? Does the current framing match the underlying structure?” Also surfaced in <task-notification> contexts where the elephant is evaluating a goldfish’s design output (“the shape of what just landed” framing — 5/30 of T2’s task-notification flavor matches).

Vocabulary cues: “same shape as,” “shape of,” “like X,” “kind of like,” “structurally a,” “this is structurally,” “reminds me of,” “the abstraction is hiding,” “mis-shaping the problem,” “structural form.”

Situation-shape signals: A new architecture/feature proposal, a refactor question, or an interface design discussion where multiple implementations are being lumped under one abstraction (a hidden-shape candidate). Also: user’s description of a problem that “rhymes” with something they’ve seen before but can’t quite name — the agent’s job is to surface the name.

Composes with

• load-bearing — shape recognition is the move that surfaces the load-bearing structural elements (which roles, which relations actually carry the analogy).
• cargo-cult (anti-pattern relationship) — cargo cult is shape without the underlying force-dynamic. “Shape as label” without verification is the cargo-cult precursor.
• container — many shapes are container-shaped (encapsulation, scope, boundary).

When it doesn’t apply

• “Same shape as X” as a label — when “shape” is used as a decorative claim rather than a structure-mapping with verifiable role correspondences. (Failure mode flagged in saved-insights synthesis.)
• Surface-content-load-bearing domains — branding, art, performance: the surface IS the substance; treating it as “just shape” misses the point.

Sources

• Schema-theoretic lineage (Bartlett, Schank, Rumelhart — schemas as abstract structural forms).
• Engineering taste tradition (“this has the shape of …”).

Canonical exemplars from corpus (T2 2026-05-17)

• Interface overloading hides the real shapes (cwd: campconnect, session: idx=0): “The abstraction is hiding the real shapes rather than revealing them… the threshold is incompatibility, not multiplicity.”
• Semantic shape of #615 is precision/recall (cwd: campconnect, session: idx=1): “The semantic shape of #615 is a precision/recall problem… This is a generalizable pattern across web-scraping pipelines: URL-shape is fast but unreliable; rendered-page-shape is slow but accurate.”
• 44px doctrine applied where it earns its keep (cwd: campconnect, session: idx=7): “This is the same shape as several patterns the codebase has already absorbed — apply doctrines where they earn their keep, not as universal mandates.”
• Negative-result suppression list shape (cwd: campconnect, session: idx=8): “What you’re describing is structurally a negative-result suppression list with re-evaluation policy — the same shape as DNS negative caching, malware blocklists, and CDN no-route lists.”

Trigger pattern (T2): Shape surfaces when the user proposes a new architecture/feature/handling rule and the agent either recognizes it as having the same structural form as a known pattern, or finds the proposal’s framing is mis-shaping the actual problem — rarely a single verb in the user message; the lexical signal “like” appears in 28% of real-user-msg triggers.