Research State Snapshot

• Theory Version: TH-TYPO-ADAPTIVE-002
• Knowledge Base Version: KB-ATLAS-2026-07-21
• Highest Confidence Areas:
  • typography operates through multiple layers
  • local perceptual improvements do not guarantee reading improvements
  • print size and display capacity create interacting constraints
  • reader compensation can mask typographic cost
  • individual differences materially affect optimal settings
• Lowest Confidence Areas:
  • predicting transfer before testing
  • quantifying cognitive effort independently of speed
  • generalizing personalized font effects beyond study populations
  • linking reading improvements to real-world task outcomes
• Largest Remaining Unknown: How to diagnose the active bottleneck for a person, task, and environment with a short practical test.
• Active Research Streams:
  • human confusion matrices
  • personalized typography
  • compensation-cost metrics
  • cross-script generalization
  • task-weighted typography
• Recently Invalidated Ideas:
  • better glyph recognition automatically improves reading
  • one best font can be selected for a population
  • slower reading caused by disfluency reliably improves memory
  • reading speed alone is an adequate outcome
• Priority Changes:
  • personalized and task-specific models moved to highest priority
  • universal font ranking moved to low priority
  • cross-layer measurement moved ahead of additional anatomy cataloging

Executive Summary

What Was Accomplished

This phase investigated the largest uncertainty left by the prior Project Atlas typography research:

When does a measurable improvement at the glyph or font level transfer upward to word recognition, reading speed, comprehension, memory, navigation, or action?

The research reviewed evidence from vision science, psycholinguistics, eye-tracking, low-vision research, learning science, and human-computer interaction. It examined both successful and failed transfers.

Six major hypothesis cycles were completed:

1. Whether glyph-level improvement is sufficient for reading improvement.
2. Whether reading speed captures the cost of typography.
3. Whether font effects are stable across readers.
4. Whether typography-induced difficulty improves memory or comprehension.
5. Whether layout constraints can reverse local typographic benefits.
6. Whether adaptation and personalization should replace universal optimization.

Major Discovery

Cross-layer transfer is conditional.

A local typography change transfers upward only when:

1. it improves the process that is currently limiting performance;
2. the improvement is not cancelled by a new cost at another layer;
3. the higher-level task depends materially on the improved process;
4. the measurement captures the relevant benefit;
5. the reader has not already compensated enough to hide the difference.

This produces the proposed Transfer Alignment Principle:

A typography intervention produces system-level benefit only when the intervention, active bottleneck, task dependency, and outcome measure are aligned.

Confidence

Moderate to high.

The individual components are supported by multiple independent studies. The integrated predictive model remains unvalidated as a whole.

Remaining Uncertainty

The most important unresolved issue is practical diagnosis. Atlas needs a short protocol that can distinguish whether a reader is currently constrained by:

• visual acuity
• contrast
• crowding
• glyph confusability
• line capacity
• eye-movement strategy
• lexical processing
• comprehension
• fatigue
• interface navigation

Original Objective

Determine when and why typography improvements transfer from lower perceptual layers to higher reading and task outcomes.

Scope

Included:

• glyph recognition
• word recognition
• continuous reading
• visual span
• eye movements
• reading speed
• comprehension
• memory
• subjective workload
• personalization
• display and layout interaction

Excluded from this phase:

• full cross-script comparison
• production of new fonts
• new human-subject experiments
• branding and expressive typography except where relevant to transfer
• complete interface-navigation research

Repository Context

This package extends three prior artifacts:

1. The visual-information foundation established that typography can be modeled as a noisy channel only at clearly defined layers.
2. The letter-confusion audit established that similarity ratings, confusion matrices, visual-span measures, and synthetic datasets answer different questions.
3. The autonomous typography report established the active-bottleneck model and rejected universal monotonic rules for spacing, size, weight, and width.

The present package tests the missing link between those lower-level findings and real reading behavior.

Current Understanding

Typography changes can produce at least four distinct outcomes:

1. Capacity improvement
   The reader can recover more visual information.

2. Efficiency improvement
   The same task is completed faster or with fewer eye movements.

3. Compensation reduction
   Performance remains similar, but less effort is required.

4. Task improvement
   Comprehension, recall, navigation, or decision quality improves.

These outcomes are related but not interchangeable.

A study may show a capacity gain without an efficiency gain. A reader may maintain speed while making longer fixations. A font may feel easier without improving comprehension. A more difficult font may slow reading without creating deeper processing.

Key Discoveries

KD-001: Transfer is bottleneck-dependent

Glyph improvements matter most near perceptual limits, in peripheral reading, under low contrast, for low-vision readers, and in low-context tasks.

When lexical, semantic, temporal, or oculomotor processes dominate, improved glyph distinction may not change reading speed.

KD-002: Compensation can preserve output

Readers alter:

• fixation duration
• fixation count
• saccade length
• regressions
• reading strategy
• use of context

As a result, average reading speed can remain stable while effort changes.

KD-003: Individual differences are not noise

Large-scale personalized-font studies report that different readers achieve their best reading performance with different typefaces. Preference does not reliably predict performance.

The appropriate unit of analysis may therefore be the reader-font-task combination rather than the typeface alone.

KD-004: Local difficulty is usually not a desirable difficulty

Disfluent fonts reliably increase perceived difficulty, but effects on learning and memory are inconsistent or null. Additional orthographic effort does not necessarily create semantic or relational processing.

KD-005: Layout can cancel local gains

Larger print improves recognition until fewer characters fit per line or screen. Wider forms may improve local shape availability while increasing line length or reducing text capacity. More spacing can reduce crowding while fragmenting words.

Typography must therefore be evaluated within its display geometry.

KD-006: Personalization is more promising than universal ranking

Evidence from adaptive and individualized typography suggests meaningful reading speed gains can be obtained by tuning font family, size normalization, character spacing, and line spacing to the reader.

The strongest current direction is not “find the best font,” but “identify a small effective configuration for this reader and task.”

Research Log

Cycle 1: Is glyph-level improvement sufficient for continuous reading?

Objective

Test whether a font that improves isolated or peripheral letter recognition should improve continuous reading.

Hypothesis

Improved glyph recognition will reliably increase reading speed.

Evidence Found

Fonts designed for peripheral distinction can improve:

• isolated letter recognition
• peripheral word recognition
• some low-vision outcomes

Luciole showed small advantages over some comparison fonts for some readers with low vision.

Evidence Against

A font designed specifically for peripheral vision improved peripheral letter and word recognition but did not improve eye-mediated sentence reading.

Training that reduced crowding did not always produce proportional changes in visual span or reading speed.

Sources

• Bernard et al. (2016)
• He et al. (2017)
• Galiano et al. (2023)
• Legge et al. visual-span research

Analysis

The local benefit was real, but the continuous-reading system was able to compensate or was limited elsewhere.

The original hypothesis assumed a serial pipeline:

better letters → better words → faster reading

The evidence supports a conditional network:

better letters
    ├─ improves reading if letter information is limiting
    ├─ produces no change if another process is limiting
    └─ may be cancelled by spacing, width, familiarity, or eye-movement costs

Conclusion

Rejected as a universal claim.

Confidence

High

Next Step

Determine whether common reading metrics conceal compensation.

Cycle 2: Does reading speed capture typographic cost?

Objective

Test whether equal reading speed implies equivalent typography performance.

Hypothesis

If two text settings produce the same reading speed, they are functionally equivalent.

Evidence Found

Reading speed is strongly related to many practical reading outcomes and remains an important measure.

Evidence Against

Font-width studies found similar aggregate speed with different combinations of fixation counts, fixation durations, and saccade behavior.

Webpage typography studies found effects on eye movements even when higher-level performance differences were limited.

Bionic-reading studies found little or no speed advantage and changed effort or reading time inconsistently.

Sources

• Minakata and Beier (2021)
• Scaltritti et al. (2019)
• Beelders et al. (2025)
• eye-movement studies of reading strategy

Analysis

Readers can maintain throughput by paying a higher internal cost.

Equivalent speed can coexist with:

• longer fixations
• more regressions
• reduced spare attention
• lower robustness under distraction
• higher fatigue
• reduced persistence

Reading speed is an output variable, not a complete cost function.

Conclusion

Rejected.

Confidence

High

Next Step

Investigate whether individual readers exhibit stable differences in font response.

Cycle 3: Is there a universally superior font for fluent readers?

Objective

Test whether population averages can identify one generally best-performing typeface.

Hypothesis

Once fonts are normalized for size, one or a small set of fonts will consistently outperform others across readers.

Evidence Found

Some fonts perform better on average in specific experiments. Familiar and conventional structures often support fluent reading.

Evidence Against

Individual-difference studies found substantial reader-specific variation in font preference and reading speed.

Preference and familiarity did not reliably identify the fastest font for each reader.

Studies of school-age readers found that personalized font selection could increase reading speed while preserving comprehension.

Adaptive-font systems demonstrated the feasibility of learning a reader-specific font configuration.

Sources

• Wallace et al. (2020)
• Wallace et al. (2022)
• Kadner et al. (2021)
• Nedeljković et al. (2020)

Analysis

Population means can conceal crossover interactions:

Reader A: Font X > Font Y
Reader B: Font Y > Font X
Average: X ≈ Y

This is not random measurement noise if the preference remains reproducible.

Universal defaults are still needed, but they should be viewed as robust starting points rather than optima.

Conclusion

Rejected as a universal claim.

Confidence

Moderate to high

Next Step

Test whether deliberate difficulty can improve higher-level outcomes.

Cycle 4: Does typographic disfluency improve memory or comprehension?

Objective

Evaluate whether making text harder to read creates deeper processing.

Hypothesis

A difficult-to-read font increases mental effort, which improves memory and comprehension.

Evidence Found

A small number of studies reported benefits under limited conditions, especially for distinctive material or novel-word learning.

Distinctive typography can direct attention when selectively applied.

Evidence Against

Multiple studies found no benefit of Sans Forgetica or other disfluent fonts for recall or comprehension.

A one-week delayed-recall study found a strong testing benefit but no font benefit, despite a large perceived-difficulty manipulation.

The authors argued that disfluency increases local orthographic work rather than semantic-relational processing and may interfere with integration across words.

Recent work also reports null effects on comprehension and attention.

Sources

• Wetzler et al. (2021)
• Taylor et al.
• Geller et al.
• Tietz et al. (2025)
• disfluency meta-analytic debate

Analysis

Effort is not fungible.

More perceptual effort
≠
more semantic processing

Difficulty helps only when the extra work engages processes that are useful for the later task.

This is consistent with transfer-appropriate processing and the broader Atlas transfer model.

Conclusion

Rejected as a general theory.

Confidence

High

Next Step

Investigate interactions between local legibility and available layout space.

Cycle 5: Can local legibility improvements be reversed by layout constraints?

Objective

Test whether larger, wider, or more spaced type remains beneficial when display capacity is considered.

Hypothesis

Improving local visual availability will improve total reading performance.

Evidence Found

Larger print improves reading below critical print size.

Spacing can reduce crowding.

Wider or larger-x-height forms may improve recognition under small-size conditions.

Evidence Against

On small displays, increasing print size reduces characters per line and lines per screen. Reading performance requires both sufficient angular size and enough text capacity.

Research found a minimum approximate character count per line for maintaining a criterion level of reading speed.

Increased spacing and width can introduce more eye movements and navigation cost.

Sources

• Atilgan et al. (2020)
• Legge and Bigelow (2011)
• Yu et al. (2007)
• Minakata and Beier (2021)
• Sawyer et al. (2025)

Analysis

Typography and layout share a finite spatial budget.

A local change reallocates that budget:

larger characters
    → more visible detail
    → fewer characters per fixation or line
    → more navigation

more spacing
    → less crowding
    → lower density
    → greater peripheral extent

The correct objective is not maximum local legibility. It is maximum task performance under a spatial constraint.

Conclusion

The hypothesis was rejected in its simple form.

Confidence

High

Next Step

Determine whether adaptive typography is theoretically and practically superior to fixed settings.

Cycle 6: Should typography be modeled as an adaptive control problem?

Objective

Evaluate whether personalization and situational adjustment offer a stronger model than universal design values.

Hypothesis

A small adaptive system can identify settings that improve reading for a specific reader without harming comprehension.

Evidence Found

Individualized studies show measurable reading-speed differences across fonts.

Generative and adaptive systems have optimized font shapes or configurations using human-in-the-loop feedback.

Personalized selection can preserve comprehension while improving speed.

Recent situational systems report improvements in efficiency and perceived workload when typography responds to context.

Evidence Against

Personalized testing has costs:

• measurement noise
• practice effects
• short-term optimization
• preference instability
• overfitting to one text type
• implementation complexity

The evidence base is still small compared with conventional reading research.

A personalized optimum for speed may not optimize comfort, retention, or accessibility.

Sources

• Wallace et al. (2020, 2022)
• Kadner et al. (2021)
• adaptive typography HCI research
• individual-difference studies

Analysis

The evidence supports constrained adaptation, not unlimited customization.

A practical system should:

1. begin with a robust accessible default;
2. test a small controlled set of alternatives;
3. measure more than preference;
4. preserve user override;
5. optimize for the current task;
6. avoid continuous visual instability.

Conclusion

Provisionally supported.

Confidence

Moderate

Next Step

Develop and validate a short bottleneck-diagnostic and personalization protocol.

Confirmed Findings

CF-TR-001

Local perceptual improvements do not guarantee higher-level reading benefits.

Evidence: EV-TR-001, EV-TR-002, EV-TR-003
Confidence: High

CF-TR-002

Reading speed alone can conceal changes in eye-movement strategy and effort.

Evidence: EV-TR-004, EV-TR-005
Confidence: High

CF-TR-003

Typographic effects vary meaningfully across readers.

Evidence: EV-TR-006, EV-TR-007, EV-TR-008
Confidence: Moderate-High

CF-TR-004

Font preference is not a reliable substitute for measured performance.

Evidence: EV-TR-006
Confidence: Moderate-High

CF-TR-005

Disfluent fonts do not reliably improve memory or comprehension.

Evidence: EV-TR-009, EV-TR-010
Confidence: High

CF-TR-006

Print size, spacing, width, and display capacity interact.

Evidence: EV-TR-011, EV-TR-012, EV-TR-013
Confidence: High

CF-TR-007

Personalized typography can improve reading performance while preserving comprehension in some populations.

Evidence: EV-TR-006, EV-TR-007, EV-TR-008
Confidence: Moderate

Evidence Registry

EV-TR-001

Citation: Bernard et al. (2016), font designed for peripheral vision.
Finding: Improved peripheral letter and word recognition, but not eye-mediated reading performance.
Supports: HY-TR-002, TH-TYPO-ADAPTIVE-002
Quality: High, peer-reviewed experimental study.

EV-TR-002

Citation: He et al. (2017), linking crowding, visual span, and reading.
Finding: Challenges a simple causal pathway from reduced crowding to reading improvement.
Supports: HY-TR-002
Quality: High.

EV-TR-003

Citation: Galiano et al. (2023), Luciole low-vision font.
Finding: Small, population- and comparison-dependent advantages rather than universal superiority.
Supports: HY-TR-001, HY-TR-004
Quality: Moderate-High.

EV-TR-004

Citation: Minakata and Beier (2021), font width and eye movements.
Finding: Font width changed eye-movement strategies and exposed tradeoffs.
Supports: HY-TR-003
Quality: High.

EV-TR-005

Citation: Scaltritti et al. (2019), typography on real webpages.
Finding: Typographic variables affected eye movements and performance across reader groups.
Supports: HY-TR-003
Quality: High.

EV-TR-006

Citation: Wallace et al. (2020), individual font preference and effectiveness.
Finding: Readers differ in fastest font; normalization matters; preference and familiarity do not fully predict performance.
Supports: HY-TR-004, TH-TYPO-PERSONALIZED-001
Quality: Moderate-High.

EV-TR-007

Citation: Wallace et al. (2022), different fonts increase reading speed for different individuals.
Finding: Individual font tuning can improve reading speed while maintaining comprehension.
Supports: HY-TR-004, HY-TR-007
Quality: Moderate-High.

EV-TR-008

Citation: Kadner et al. (2021), AdaptiFont.
Finding: Human-in-the-loop optimization can generate reader-specific font configurations.
Supports: HY-TR-007
Quality: Moderate.

EV-TR-009

Citation: Wetzler, Pyke, and Werner (2021), Sans Forgetica delayed recall.
Finding: Strong manipulation of perceived difficulty; no one-week recall benefit; testing effect remained.
Supports: HY-TR-005
Quality: High.

EV-TR-010

Citation: Tietz et al. (2025), text disfluency, attention, and comprehension.
Finding: Disfluency did not improve comprehension or reduce mind wandering.
Supports: HY-TR-005
Quality: Moderate-High.

EV-TR-011

Citation: Atilgan et al. (2020), print-size and display-size constraints.
Finding: Adequate print size and sufficient characters per line jointly constrain reading.
Supports: HY-TR-006
Quality: High.

EV-TR-012

Citation: Yu et al. (2007), letter spacing.
Finding: Spacing reduces crowding but increases peripheral extent; effects are non-monotonic.
Supports: HY-TR-006
Quality: High.

EV-TR-013

Citation: Legge and Bigelow (2011), print-size review.
Finding: Critical print size and fluent range; no unlimited benefit from larger text.
Supports: HY-TR-006
Quality: High.

Hypothesis Registry

HY-TR-001: Direct Transfer

Claim: Better glyph recognition directly improves reading.
Status: Rejected as universal; retained conditionally.
Confidence: High.

HY-TR-002: Active Bottleneck

Claim: A local typography change transfers only when it affects the current limiting process.
Status: Supported.
Confidence: Moderate-High.

HY-TR-003: Compensation Masking

Claim: Readers can preserve speed by changing eye movements and effort, masking typography cost.
Status: Supported.
Confidence: High.

HY-TR-004: Stable Individual Optima

Claim: Readers have reproducible differences in effective font configuration.
Status: Provisionally supported.
Confidence: Moderate.

HY-TR-005: Disfluency Benefit

Claim: Perceptual difficulty improves learning by inducing deeper processing.
Status: Rejected as a general claim.
Confidence: High.

HY-TR-006: Spatial Budget

Claim: Local visual improvements can create layout costs that reverse their benefit.
Status: Supported.
Confidence: High.

HY-TR-007: Adaptive Typography

Claim: Controlled personalization can outperform one fixed default for some readers and tasks.
Status: Provisionally supported.
Confidence: Moderate.

HY-TR-008: Transfer Alignment

Claim: System-level benefit requires alignment among intervention, active bottleneck, task dependency, and outcome measure.
Status: New theory candidate.
Confidence: Moderate-High.

Failed Assumptions

1. Assumption: Letter recognition is the base variable governing all reading.
   • Failure: Other processes can dominate.
2. Assumption: Reading speed is a complete performance measure.
   • Failure: Eye-movement and effort changes can occur without speed changes.
3. Assumption: Population averages identify the best font.
   • Failure: Crossover effects among readers are substantial.
4. Assumption: More perceptual effort creates deeper learning.
   • Failure: Effort is often consumed locally at the orthographic layer.
5. Assumption: Larger or more open text is always better.
   • Failure: Spatial capacity and navigation introduce opposing costs.
6. Assumption: Preference can guide personalization.
   • Failure: Preferred and fastest fonts do not reliably match.

Proposed Models

TH-TYPO-ADAPTIVE-002: Adaptive Layered Reading Theory

Typography performance emerges from interaction among:

visual signal
× observer capability
× learned familiarity
× language redundancy
× display geometry
× task demand
× compensation strategy

No typeface has one context-free readability value.

CN-TR-001: Transfer Alignment Principle

A local improvement produces a higher-level benefit when:

Intervention affects active bottleneck
AND
task depends on that bottleneck
AND
new costs do not cancel the gain
AND
measurement detects the gain

CN-TR-002: Compensation Reserve

Readers possess a finite capacity to compensate for weak typography.

Compensation reserve is consumed through:

• longer fixation
• additional fixation
• regression
• contextual inference
• working-memory effort
• slower navigation

A design may appear adequate in easy conditions but fail under distraction, fatigue, or secondary-task load.

CN-TR-003: Spatial Typography Budget

Every display allocates a finite visual area among:

• character size
• character width
• letter spacing
• word spacing
• line length
• number of lines
• margins
• hierarchy

Improvement in one dimension consumes capacity elsewhere.

DF-TR-001: Typography Intervention Decision Framework

1. Define the task.
2. Define the critical error.
3. Identify the likely bottleneck.
4. Choose an intervention that targets that bottleneck.
5. Measure local effect.
6. Measure eye-movement or effort effect.
7. Measure task-level transfer.
8. Test under degraded conditions.
9. Test individual variation.
10. retain a user override.

Theory Impact Assessment

Affected Theory Records

• TH-TYPO-CHANNEL-001
• TH-TYPO-ADAPTIVE-001
• TH-TYPO-BOTTLENECK-001

New Principle Candidates

• CN-TR-001 Transfer Alignment Principle
• CN-TR-002 Compensation Reserve
• CN-TR-003 Spatial Typography Budget
• TH-TYPO-PERSONALIZED-001 Reader-Specific Typography

Deprecated Principles

• Universal Font Superiority
• Direct Glyph-to-Reading Transfer
• Disfluency as General Desirable Difficulty

Confidence Changes

• Active bottleneck theory: Moderate → Moderate-High
• Personalized typography: Low → Moderate
• Universal readability score: Low → Very Low
• Compensation-cost model: Moderate → Moderate-High

Predictions Created

1. Personalized settings will produce larger gains under perceptually demanding conditions than under easy reading.
2. Readers with equal speed across fonts will differ in fixation and workload.
3. Preference will predict adoption better than peak performance, but not peak reading speed.
4. Low-context tasks will show stronger transfer from glyph distinction than ordinary prose.
5. Font gains measured on a large display will shrink or reverse on a small display when line capacity falls below a critical range.
6. Disfluent typography applied selectively may improve attention to marked elements, while applying it globally will not improve learning.

Predictions Invalidated

• A font engineered for isolated-character distinction will necessarily improve sentence reading.
• Greater perceived difficulty predicts better delayed recall.

Required Theory Registry Updates

Add:

• TH-TYPO-ADAPTIVE-002
• CN-TR-001
• CN-TR-002
• CN-TR-003
• DF-TR-001

Deprecate:

• TH-TYPO-DIRECT-TRANSFER-001
• TH-TYPO-DISFLUENCY-001

Open Questions

Critical

1. Can a five-minute test diagnose a reader's active typography bottleneck?
2. How stable are personalized font gains across days, devices, and content?
3. Which compensation metrics best predict fatigue and failure under load?
4. How should Atlas optimize multiple objectives without reducing them to one score?

High

5. Do personalized settings improve comprehension over long reading sessions?
6. How much gain comes from font family versus size normalization, spacing, grade, and line layout?
7. Which reader characteristics predict response to typography?
8. Can a computational model narrow the personalization search space reliably?
9. How do personalized settings interact with browser zoom and accessibility overrides?

Medium

10. Can selective disfluency serve as an attention signal without harming comprehension?
11. Are individual optima stable across languages and scripts?
12. How much familiarization is needed before a new font reaches stable performance?

Recommended Next Research

Research Backlog

• Extract effect sizes from individualized-font studies.
• Record whether fonts were normalized by x-height, cap height, or nominal size.
• Compare preference, speed, comprehension, and workload within participants.
• Build a taxonomy of transfer failures.
• Review pupillometry and dual-task methods for reading effort.
• Review standards for safety-critical labels and identifiers.
• Investigate font grade as distinct from weight.
• Compare static adaptation with continuous adaptive typography.
• Analyze accessibility risks of changing typography automatically.
• Review long-session reading and fatigue studies.

Suggested Specialized Research Agents

Agent A: Personalized Typography Analyst

Focus:

• individual-difference datasets
• reliability
• effect sizes
• personalization algorithms
• overfitting risk

Agent B: Eye-Movement Compensation Analyst

Focus:

• fixation duration
• regressions
• saccade length
• pupil response
• dual-task cost
• fatigue

Agent C: Task-Critical Typography Analyst

Focus:

• medication labels
• safety signage
• serial numbers
• control panels
• glance reading

Agent D: Spatial Budget Modeler

Focus:

• print size
• viewport size
• characters per line
• reflow
• zoom
• responsive layout

Parallel Research Opportunities

• typography personalization
• visual hierarchy transfer
• icon and symbol confusion
• chart-label legibility
• wayfinding
• alert design
• multilingual interfaces
• low-vision adaptive systems

Risks

• personalization may overfit short tests
• speed optimization may reduce comprehension or comfort
• adaptive changes may create instability
• study fonts may not reflect production implementations
• individual effects may be inflated by repeated-measures designs
• publication bias may favor positive personalization results
• accessibility systems may conflict with authored adaptation
• laboratory reading may not predict real-world persistence

Cross-Discipline Opportunities

Control Theory

Model the reader as a controller that changes sampling behavior to preserve performance.

Economics

Treat attention and screen area as scarce resources with opportunity costs.

Decision Theory

Weight typography errors by consequence rather than frequency alone.

Reliability Engineering

Use degradation testing and failure thresholds instead of ideal-condition ranking.

Personalized Medicine

Use robust defaults plus individual response testing rather than population averages alone.

Information Theory

Measure uncertainty reduction at each layer, without equating symbol accuracy with meaning.

Knowledge Relationships

Glyph robustness
    ↓ conditional transfer
Word recognition
    ↓ constrained by
Eye movements + language inference
    ↓ constrained by
Layout and display capacity
    ↓ produces
Reading efficiency and compensation cost
    ↓ affects
Comprehension, navigation, action

Personalization affects every transition rather than only glyph selection.

Repository Updates

Create or update:

/research/typography/rep/RP-ATLAS-TYPO-TRANSFER-001.md
/theory/TH-TYPO-ADAPTIVE-002.md
/concepts/CN-TR-001-transfer-alignment.md
/concepts/CN-TR-002-compensation-reserve.md
/concepts/CN-TR-003-spatial-typography-budget.md
/frameworks/DF-TR-001-typography-intervention.md
/evidence/typography-transfer-registry.yaml

Website Updates

Add a Typography Transfer section with:

• layered transfer diagram
• local-versus-system outcome examples
• personalized typography evidence
• rejected disfluency claim
• interactive spatial-budget illustration
• confidence indicators

Do not present a universal “best font” list.

AI Consumption Notes

Agents consuming this package should:

1. Treat reading speed, comprehension, and preference as separate outcomes.
2. Never infer transfer from a lower-level metric alone.
3. Preserve individual-level data when available.
4. Record whether font sizes were physically or optically normalized.
5. Distinguish typeface, font, grade, weight, spacing, and layout.
6. Record display size and characters per line.
7. Separate null effects from evidence of equivalence.
8. Avoid recommending disfluent fonts for learning without task-specific evidence.
9. Prefer profiles and conditional rules over global scores.
10. Reference evidence and hypothesis IDs in future work.

Handoff Instructions

The next agent should begin with Open Question 1:

Can a short diagnostic determine the active typography bottleneck?

Required next steps:

1. Review adaptive psychophysical methods such as qReading.
2. Review short font-tuning methods used by individualized-font studies.
3. Identify the minimum set of tasks needed to distinguish:
   • size limitation
   • crowding
   • glyph confusion
   • line-capacity limitation
   • lexical limitation
4. Propose an experiment that can run in five to ten minutes.
5. Define stopping criteria and reliability requirements.
6. Produce RP-ATLAS-TYPO-DIAGNOSTIC-001.

Research Journal

JR-ATLAS-TR-001

Reviewed previous Atlas artifacts and identified cross-layer transfer as the largest unresolved uncertainty.

JR-ATLAS-TR-002

Compared successful local recognition improvements with failed reading-speed transfer. Direct-transfer theory weakened.

JR-ATLAS-TR-003

Reviewed eye-movement evidence. Added compensation masking and compensation reserve concepts.

JR-ATLAS-TR-004

Reviewed individual-difference and adaptive-font studies. Personalized typography confidence increased from low to moderate.

JR-ATLAS-TR-005

Reviewed disfluency and Sans Forgetica evidence. General desirable-difficulty hypothesis rejected.

JR-ATLAS-TR-006

Reviewed display-size and print-size interaction. Spatial typography budget model added.

JR-ATLAS-TR-007

Integrated findings into Transfer Alignment Principle and adaptive layered reading theory.

Research Quality Metrics

• Primary Sources: 13 core experimental or review sources
• Independent Source Families: 7
• Counterexamples Reviewed: 9
• Competing Viewpoints Reviewed: 5
• Hypotheses Tested: 8
• Failed Hypotheses: 4 universal claims, 2 simple assumptions
• Research Completeness: 82% for current objective
• Confidence Gain: Moderate → Moderate-High
• Open Questions Reduced: 1 major question decomposed into 12 testable questions

Research Debt

Missing Evidence

• raw individualized-font datasets
• long-term personalization stability
• fatigue and workload evidence
• safety-critical task transfer
• non-Latin transfer evidence

Missing Experiments

• short bottleneck diagnostic
• within-person stability study
• low-context versus prose comparison
• dual-task compensation study
• small-display personalization study

Missing Disciplines

• occupational ergonomics
• clinical low-vision rehabilitation
• safety engineering
• educational measurement
• adaptive user interfaces

Tool Limitations

• several publisher pages did not expose full datasets
• some current HCI results are available only as abstracts
• no new human experiments were conducted
• effect sizes were not pooled meta-analytically

Assumptions Awaiting Evidence

• compensation reserve is finite and measurable
• personalized font effects are stable enough for production use
• active bottlenecks can be diagnosed quickly
• computational proxies can reduce human testing substantially

Appendix A: Bibliography

Academic

• Atilgan, N. et al. (2020). Reconciling Print-Size and Display-Size Constraints on Reading.
• Bernard, J. B. et al. (2016). A New Font Specifically Designed for Peripheral Vision Improves Peripheral Letter and Word Recognition, but Not Eye-Mediated Reading Performance.
• Galiano, A. R. et al. (2023). Luciole, a New Font for People with Low Vision.
• He, Y. et al. (2017). Linking Crowding, Visual Span, and Reading.
• Kadner, F. et al. (2021). Increasing Individuals' Reading Speed with a Generative Font Model and Bayesian Optimization.
• Legge, G. E., and Bigelow, C. A. (2011). Does Print Size Matter for Reading?
• Minakata, K., and Beier, S. (2021). The Effect of Font Width on Eye Movements During Reading.
• Nedeljković, U. et al. (2020). You Read Best What You Read Most.
• Scaltritti, M. et al. (2019). Investigating Effects of Typographic Variables on Webpage Reading Through Eye Movements.
• Wallace, S. et al. (2020). Individual Differences in Font Preference and Effectiveness as Applied to Interlude Reading in the Digital Age.
• Wallace, S. et al. (2022). Different Fonts Increase Reading Speed for Different Individuals.
• Wetzler, E. L., Pyke, A. A., and Werner, A. (2021). Sans Forgetica Is Not the Font of Knowledge.
• Yu, D. et al. (2007). Effect of Letter Spacing on Visual Span and Reading Speed.

Books

• Legge, G. E. (2007). Psychophysics of Reading in Normal and Low Vision.

Industry

• Google Fonts accessibility and readability research initiatives.
• Browser and operating-system text customization documentation.

Patents

No patent evidence was decisive in this phase.

Standards

• WCAG text spacing and reflow requirements.
• ISO 9241 human-system interaction standards.

Historical

• Bouma, H. letter recognition and crowding research.
• Reicher-Wheeler word-superiority research.

Other

• qReading adaptive measurement method.
• Current adaptive typography HCI systems.

Appendix B: Completion Checklist

• Executive Summary
• Original Objective
• Scope
• Repository Context
• Current Understanding
• Key Discoveries
• Evidence Registry
• Hypothesis Registry
• Failed Assumptions
• Open Questions
• Recommended Next Research
• Research Backlog
• Suggested Specialized Research Agents
• Parallel Research Opportunities
• Risks
• Cross-Discipline Opportunities
• Knowledge Relationships
• Repository Updates
• Website Updates
• AI Consumption Notes
• Handoff Instructions
• Research Journal
• Appendix
• Theory Impact Assessment
• Research Quality Metrics
• Research Debt
• Completion Checklist