| name | foveal-forge-protocol |
| description | Declarative specification for the foveal-forge perceptual vision training protocol. Defines the drills, their psychophysical parameters, session structure, scoring, and progression rules. This is the single source of truth for the training contract; code implements this spec. |
| version | 0.1.0 |
| status | draft |
Foveal Forge Training Protocol
Purpose
Train the neural correlates of near vision (contrast sensitivity, saccadic efficiency, reading fluency) in presbyopic adults via perceptual learning. This spec defines what the training does, not how it is implemented. Code in src/ implements this contract.
Evidence base
Each drill maps to a published mechanism in references/evidence.md. Drills without citations do not ship.
Session structure
A standard session is 12 minutes, composed of:
| Phase | Duration | Drill |
|---|
| Warm-up | 1 min | Fixation stability |
| Block A | 4 min | Gabor contrast sensitivity |
| Rest | 30 sec | Distance gaze cue (20-20-20) |
| Block B | 3 min | Saccade + crowding |
| Rest | 30 sec | Distance gaze cue |
| Block C | 2.5 min | Reading fluency (RSVP) |
| Cooldown | 30 sec | Free near-far gaze |
Sessions are intended 5 days per week. The app does not push notifications; the user opts into a daily local alarm.
Drills
D1. Fixation stability (warm-up)
- Stimulus: central black fixation cross on neutral-gray field
- Task: maintain gaze for 60 sec
- Purpose: baseline attentional state; not scored
- Citation: N/A (preparatory)
D2. Gabor contrast sensitivity (Block A)
The core mechanism. Lateral-masking Gabor patch paradigm based on Polat et al. (2012).
- Stimulus: central Gabor patch flanked by two collinear Gabor patches at variable separation (1.5λ to 12λ)
- Spatial frequencies tested: 3, 6, 12 cycles per degree
- Orientations: vertical, ±45°, horizontal (randomized)
- Task: 2AFC — did the central patch appear or not? (yes/no with confidence)
- Adaptive procedure: 3-down-1-up staircase, terminating at 79% threshold
- Trials per session: ~80 (variable, time-boxed to 4 min)
- Scored metric: contrast threshold (Michelson) per spatial frequency
- Progression: rising spatial frequency mastery; track threshold over sessions
- Citation: Polat et al., 2012, Scientific Reports 2:278
D3. Saccade + crowding (Block B)
Trains rapid foveation and reduces crowding interference.
- Stimulus: Sloan letter triplets (target flanked by distractors) appearing at random retinal eccentricities up to 8°
- Task: identify the center letter
- Presentation: 150ms exposure, masked
- Crowding spacing: adaptive — starts at 1.5× critical spacing, narrows on correct response
- Trials per session: ~60
- Scored metric: mean reaction time, accuracy, critical spacing
- Citation: Levi, 2008, Vision Research 48:635 (crowding); Yu et al., 2010 (perceptual learning of crowding)
D4. Reading fluency / RSVP (Block C)
Rapid Serial Visual Presentation trains word-level processing speed.
- Stimulus: common-word passages (CEFR B1 level, 60-word blocks) presented one word at a time
- Initial rate: 200 words per minute
- Adaptive procedure: rate increases 5% on correctly answered comprehension question; decreases 10% on failure
- Font: sans-serif, 24pt baseline, adjustable
- Trials per session: as many 60-word blocks as fit in 2.5 min
- Scored metric: sustained WPM at 80% comprehension
- Citation: Calabrèse et al., 2014, Vision Research 103:33
Baseline assessment (one-time, on first launch)
Before training begins, the user completes:
- Pelli-Robson-style contrast sensitivity test at standard reading distance (40 cm self-reported)
- Standardized reading speed test (MNREAD-inspired) — measures critical print size and maximum reading speed
- Self-report: age, current correction (glasses/contacts/none), known ocular conditions (with disclaimer: this app is not a substitute for an eye exam)
Baseline establishes per-user starting parameters and the comparison line for progress.
Progression and metrics
The app surfaces three primary metrics per drill family:
- Contrast threshold (lower = better) — log scale chart, per spatial frequency
- Critical crowding spacing (lower = better) — degrees of visual angle
- Sustained reading rate at 80% comprehension (higher = better) — WPM
A composite "neural near-vision score" is not computed. We do not want users equating a single number with their actual vision.
Re-baseline
Every 30 completed sessions, the app prompts a fresh baseline run to track real progress against the initial measurement.
What this protocol does NOT include
- Eye-movement "exercises" without a perceptual learning basis (e.g., Bates palming, figure-eight tracing)
- Accommodation drills (no evidence they affect lens elasticity)
- Pinhole training (insufficient evidence)
- Any gamification that rewards continued use over actual learning (no streaks-as-currency, no leaderboards)
Safety stops
The app must:
- Surface a "stop and rest" prompt if the user reports eye strain (post-session 0-10 scale; ≥7 triggers a recommended rest day)
- Include in onboarding: discontinue use and consult an ophthalmologist if you experience pain, sudden vision changes, flashes, or floaters
- Refuse to start a session within 30 minutes of the last completed session
Versioning
This spec uses semantic versioning. Code references the protocol version it implements. A spec change with a major-version bump invalidates prior session comparability — the app warns the user before applying.
