Project Glosa is a working name. Public naming and trademark clearance are in progress.
Project Glosa · Science & claims

On the science behind Glosa
what we design around, what we are still testing, and what we refuse to claim.

Every educational, cognitive, psychological, focus, memory, accessibility, and AI claim Glosa makes is graded, sourced, dated, and open to revision. This page is that register in plain view. A source is not a permission slip for overstatement.

01
Claim states

Five states, one honest rule.

A claim in Glosa marketing, onboarding, in-product copy, App Store listing, press release, or investor deck may never exceed the state assigned to it here. States describe what the evidence supports for Glosa's implementation, not what a claim could theoretically achieve in some other product.

Established

Design requirement

Convergent evidence and mature standards support using the principle as a design requirement. Does not mean every implementation produces the effect.

Supported

Credible mechanism

Credible evidence supports the mechanism or outcome — with task, population, and implementation limits.

Context-dependent

Depends on setting

Direction or magnitude varies meaningfully with material, reader, timing, modality, or design.

Exploratory

Plausible; testing

Plausible enough to test, but not appropriate as an efficacy or superiority claim.

Not claimed

Insufficient or refused

Evidence is insufficient, the concept is misleading, or the claim would exceed what Glosa can establish.

A source is not a permission slip for overstatement.
02
What every claim review records

A single review page for every promise.

Before any claim reaches the public, a review page is filed for it. The review captures:

  • exact wording;
  • outcome and comparison;
  • population and material;
  • study design and sample;
  • timing of measurement;
  • effect size where available;
  • replication or meta-analysis status;
  • implementation similarity to Glosa;
  • conflicts and limitations;
  • revision date.

Marketing may never translate:

  • association into causation;
  • one study into a universal rule;
  • a preference into a learning advantage;
  • subjective calm into improved comprehension;
  • product completion into durable knowledge;
  • speed into understanding;
  • AI confidence into truth.
If a claim cannot answer the review questions, the copy uses descriptive product language, not efficacy language.
03
Master claims register

Thirty entries. Every one graded.

Each row lists what a marketing team might want Glosa to claim, the state assigned to that claim under this register, and the approved language Glosa may use in its place. If a claim you expect is not on this page, it has not been reviewed — and therefore has not been approved.

Not claimed

Fixed learning styles

Preferences configure the interface. Glosa does not diagnose a visual, auditory, or kinesthetic type. The idea failed the science.

“Choose how you want this work presented today.”
Context-dependent

Paper vs. screen

A small paper advantage appears in some settings — especially expository reading and self-calibration. Not a general rule; narrative differences are small.

“Glosa rebuilds spatial stability, visible extent, and quiet conditions often weakened on screens.”
Not claimed

Audio is inferior to reading

Higher-level comprehension can be similar in some contexts. Modality differences depend on task, material, and population.

“Read, listen, or move between both while keeping your place.”
Context-dependent

Dual text+audio retention

Cognitive load, learner, and material matter. No universal advantage established.

“Use text and audio together when it helps you access or track the work.”
Supported

Retrieval practice

Testing and free recall can outperform restudy for delayed retention in many contexts.

“Recall before review can strengthen later memory.”
Supported

Spacing

Distributed practice is broadly supported. Effective interval depends on retention horizon and material.

“Return to selected ideas over time.”
Supported

Self-explanation

Prompts can help; benefits and time costs vary.

“Explaining an idea in your own words can reveal what you understand.”
Not claimed

Dramatic speed reading

Speed–accuracy tradeoff. Very large gains usually sacrifice comprehension or change the task.

“Inspectional Mode helps you decide what deserves slow reading.”
Not claimed

Faster TTS improves learning

Speed is useful for access. Deeper learning depends on material and comprehension.

“Choose a listening speed that serves the task.”
Exploratory

Ambient sound & focus

Effects vary. Intelligible speech or lyrics often interfere. Self-selected sound may help some readers.

“Atmosphere is optional. Silence is always available.”
Not claimed

ASMR & comprehension

ASMR can affect responders' affect and physiology. Comprehension evidence is insufficient.

“ASMR options are experimental and voluntary.”
Not claimed

Binaural beats or isochronic tones

Product-category claims exceed current Glosa evidence review.

Excluded from first-party sound library and marketing.
Context-dependent

Synchronized highlighting

Strong practical and access value. Comprehension benefit depends on reader and material.

“Follow the spoken passage and keep your place.”
Not claimed

Fonts that cure dyslexia

Typography preferences and accessibility matter. Universal treatment claims are inappropriate.

“Choose the typeface and spacing that feel most readable.”
Not claimed

AI summaries equal reading

Summary supports orientation but removes sequence, evidence, voice, and detail.

“A map before the territory.”
Not claimed

AI knows what a passage means

AI can offer attributed interpretations. Meaning remains contested and reader-authored.

“Here are sourced interpretations and the evidence behind them.”
Not claimed

Automatic sacredness detection

No validated method. High risk of cultural or spiritual overreach.

“Turn on Reverent Mode when you want the tool to withdraw.”
Exploratory

Reader-first AI preserves agency

Cognitive offloading and ownership research motivate the design. Glosa-specific effect requires testing.

“Glosa invites your response before comparison.”
Not claimed

Provenance guarantees truth

Provenance improves auditability, but sources can be wrong, incomplete, or misrepresented.

“See what supports the claim and where uncertainty remains.”
Not claimed

Local processing = safe

Locality reduces transmission but device, storage, and security risks remain.

“Processed on this device; see the data receipt.”
Context-dependent

Personalization improves learning

Goal, material, and prior-knowledge adaptation may help. Opaque profiling and style matching are not justified.

“Glosa adapts the interface to choices you can inspect and change.”
Exploratory

Interruption-free design

Attention fragmentation can impair encoding. Glosa-specific effect needs testing.

“Designed to reduce interruption and preserve the reading path.”
Exploratory

Confidence checks help metacomprehension

Calibration matters. Paper advantage suggests interfaces should not inflate confidence.

“Record confidence before seeing the comparison.”
Not claimed

Visual diagrams help all readers

Representations can help when matched to content — not to a fixed visual type.

“Open a diagram when structure is easier to see than describe.”
Not claimed

Public-domain classics are cognitively superior

They are rights-accessible, culturally important, and suitable for product testing. Superiority is not established.

“A curated constellation of works that shaped literature, philosophy, and science.”
Not claimed

Glosa increases session length

Session length may rise or fall. Longer is not inherently better.

Use purpose completion and closure, not time-on-app, as the outcome.
Not yet claimed

Glosa improves learning

Individual mechanisms have support. Integrated product effect requires trials.

“Built around retrieval, spacing, source comparison, and reader control.”
Not claimed

Glosa reduces anxiety

Calm design may be preferred. Mental-health treatment claims require clinical evidence and regulatory review.

“A quieter reading environment.”
Context-dependent

AI can convey lived experience

AI can retrieve and compare attributed testimony. It has no first-person experience.

“People have described this as…”
Not claimed

Model improvement requires private reading

Product evaluation can use synthetic fixtures, opt-in studies, and reviewed examples.

“Your reading is not training data.”
04
Mode-specific claim rules

What each reading mode may — and may not — say.

Glosa has multiple modes: Savor and Story for immersive reading, Inspectional for orientation, Analytical for study, Listen and Dual for audio, and Research for source-linked comparison. Each mode is allowed a different vocabulary.

Savor · Story

May say
  • spoiler-aware;
  • designed to preserve immersion;
  • optional contextual definitions and allusions;
  • text and audio continuity.
May not say
  • increases empathy;
  • produces flow;
  • improves mental health;
  • stimulates the right brain;
  • creates neural resonance;
  • soundtrack matches the true emotion of the book.

Inspectional

May say
  • maps structure and questions before close reading;
  • helps the reader choose what deserves attention;
  • preserves links to the source.
May not say
  • replaces the book;
  • conveys complete understanding;
  • makes reading unnecessary;
  • guarantees faster comprehension.

Analytical · Study

May say
  • uses retrieval, self-explanation, confidence, and spaced return;
  • invites active engagement;
  • helps expose gaps in understanding;
  • supports citation and source comparison.
May not say
  • scientifically optimized for every brain;
  • guarantees grades or retention;
  • diagnoses ability;
  • replaces instruction or professional expertise.

Listen · Dual

May say
  • makes content accessible in audio;
  • keeps audio and text position aligned;
  • allows speed, voice, and navigation control.
May not say
  • faster speed creates faster learning;
  • dual mode is always superior;
  • a premium voice changes comprehension.

Research (later-stage)

May say
  • aligns terms, claims, sources, dates, and evidence;
  • preserves source lineage;
  • exposes uncertainty and disagreement;
  • generates source-linked dossiers.
May not say
  • discovers truth autonomously;
  • resolves open scientific questions by authority;
  • turns correlation into causation;
  • treats speculative interdisciplinary analogy as established fact.
05
Approved & prohibited copy

Two lists, kept in public.

Any external Glosa copy — website, App Store listing, onboarding step, paywall, press release, investor deck, or in-product tip — draws only from the approved list. The prohibited list is the specific language a claims review must clear before it can be used at all.

Approved language
  • Designed to reduce interruption and keep the original in view.
  • A map before the territory.
  • Recall before review.
  • Choose how the work meets you today.
  • Read, listen, or move between both.
  • Every factual gloss shows its source and uncertainty.
  • Atmosphere is optional; silence is a complete setting.
  • Built around evidence-supported learning practices, with claims and limitations published.
  • Your reading is not training data.
  • Glosa is testing what it has not yet earned the right to claim.
Prohibited without a new evidence review
  • brain-based;
  • neuroscience-backed (as a blanket product claim);
  • rewires your brain;
  • neural resonance;
  • left-brain / right-brain reading;
  • learning-style optimized;
  • photoreading;
  • speed-read with full comprehension;
  • learn 3×/4×/10× faster;
  • remember everything;
  • smarter in five/fifteen minutes;
  • focus frequencies;
  • binaural learning;
  • anxiety reduction;
  • ADHD treatment;
  • dyslexia cure;
  • AI knows you;
  • AI knows what the author meant;
  • scientifically proven, without exact nearby evidence and scope;
  • guaranteed comprehension;
  • objective key points;
  • unbiased summary;
  • citations guarantee truth.
06
The research program

What Glosa proposes to test.

Rev3 outlines seven proposed studies to answer the questions the register leaves open. All are advisory. None has been authorized. Each would use synthetic or public-domain fixtures before private content, distinguish exploratory from confirmatory work, and report null and negative findings alongside positive.

Experiment A

Spatial stability

Compare standard scroll, distraction-free scroll, and stable paginated or landmarked reading. Measure comprehension, delayed recall, source-location memory, orientation, and preference.

Experiment B

Reader-first AI

Compare no AI, answer-first AI, reader-first explanation, and Socratic inquiry. Measure recall, calibration, authorship, assistance dependency, and performance after AI is removed.

Experiment C

Atmosphere

Compare silence, self-selected low-information sound, familiar instrumental audio, and ASMR among self-identified responders. Report subgroup effects and harms; do not average away opposite responses.

Experiment D

Inspectional map

Compare source only, conventional summary, and source-linked inspectional map. Measure whether readers enter the original, source memory, structural understanding, and false confidence.

Experiment E

Modality

Compare text, audio, and dual by material type and task. Include regression and backtracking behavior and delayed outcomes.

Experiment F

Reader Calibration

Compare editable contextual calibration with quiet default against an opaque “personalized plan.” Measure trust, fit, understanding of adaptation, and data expectations.

Experiment G

Source and provenance interface

Test whether statement typing and source spans improve error detection without overloading the reading experience.

07
Marketing copy gate

Nine questions before any claim goes public.

Every proposed public claim — a headline, a feature card, an App Store bullet, a tweet — must answer these nine questions in writing before it can be approved.

  1. What exact outcome is being claimed?
  2. Compared with what?
  3. For whom and on what material?
  4. Over what time period?
  5. Which evidence supports it?
  6. Is the implementation sufficiently similar to the research?
  7. What qualification must be adjacent?
  8. Can the product measure the outcome without exploiting the reader?
  9. Has legal and regulatory review been triggered?
If these questions cannot be answered, use descriptive product language rather than efficacy language.
08
Claims review workflow

One path. No shortcuts.

Feature or marketing proposal
Exact claim wording
Evidence and implementation match
Claim state + required qualification
Product / research / legal / privacy review
Approved copy and revision date
Monitor, retest, revise, or withdraw

Copy that survives this workflow appears on Glosa surfaces with a claim state next to it and a revision date. Copy that does not survive it does not appear at all.

09
Research ethics

Rules the studies must live under.

Claims-safe closing

Built around what the evidence supports. Honest about the rest.

Glosa is built around evidence-supported reading and learning practices while publishing the limits of every claim. It does not diagnose learning styles, promise impossible speed gains, or replace the work with a summary.

Return to Glosa  ·  Rev3 research update  ·  Interveil Labs home