Shorts Mode
An impasse-driven remediation modality: 1-minute video flashcard shorts targeting specific impasses identified by the Gate. Each short addresses exactly one breakdown point — the precise step where the learner’s reasoning failed — not a broad topic summary.
Shorts are one modality in Lugh’s try-catch remediation cascade (see Impasse-Driven Remediation). Blog/text is the first-line content; shorts offer a visual/spatial channel for gaps that benefit from that encoding; full remediation episodes are the most expensive fallback.
The Core Insight
Different cognitive channels serve different kinds of impasses. Audio provides narrative flow; text allows self-paced re-reading; video shorts offer visual anchoring and spatial encoding. The goal is impasse-driven remediation, not re-teaching — each short covers exactly one breakdown point identified by the Gate. No narrative arc, no episode structure — just the concept, a definition, and one example.
Pipeline Architecture
Layer 1 — Content generation (local LLM)
After the Gate identifies missed/shaky concepts, the LLM generates per-concept content:
- A card title (the concept name)
- A one-sentence definition (plain language, no jargon)
- One concrete example or analogy
- An optional “why this matters” line
Input: the Gate assessment output + the episode’s learning objectives Output: a structured JSON manifest of cards
Layer 2 — Layout system
Each card follows a fixed visual template (deterministic, not generated):
- Looping background (subtle motion or abstract still)
- Title appears with fade + slight upward motion
- Definition appears after a short delay
- Example appears last
- Gentle zoom over the 5-second display window
Deterministic layout means: consistent quality, batchable, zero variation cost.
Layer 3 — Rendering (FFmpeg)
FFmpeg composes the final video from template + card content. Advantages:
- Runs entirely locally
- Parallelizable across available cores
- Near-zero marginal cost per card
A 1-minute short contains ~6–8 cards at roughly 8 seconds per card.
Layer 4 — Voice (optional)
Piper TTS reads the card text aloud over the animation. Keeps the short accessible for audio-primary learners and reinforces the definition through dual encoding.
Layer 5 — Backgrounds
Two options, escalating by effort:
- Pre-downloaded loops — royalty-free abstract motion, fast, always available
- Generated once — Stable Diffusion creates a small set of category-themed backgrounds, reused across all shorts in that topic area
Integration with the Gate
The Gate already categorizes outcomes as Solid / Shaky / Missed. Shorts mode hooks into this:
| Gate result | Audio option | Shorts option |
|---|---|---|
| Solid | Advance | — |
| Shaky | Note for next episode | Optional short for review |
| Missed | Remediation episode | Short before or after remediation |
The learner chooses. Some will prefer to re-listen; others will want a fast visual pass. Both paths feed back into the same tutor session before advancing.
Design Constraints
- One concept per short — no multi-topic shorts; this forces precision
- One minute max — forces selection of what actually matters
- No narration required — voice is optional; the visual alone should be complete
- Generated from Gate output — shorts are always scoped to gaps, never generated speculatively
What This Is Not
Not a content engine or a YouTube pipeline. The shorts are for the individual learner reviewing their own gaps. The value is specificity — you get a short on exactly what you missed, not a general overview of the topic.
Related
- Impasse-Driven Remediation — the overarching remediation philosophy and try-catch modality cascade
- The Gate — source of missed/shaky concepts that trigger short generation
- Episode Anatomy — the upstream episode structure shorts are derived from
- Shareables — completed shorts could optionally become shareable artifacts