A completed research and product-integration proposal, offered as advice — not authorization.
Project Glosa has completed a new research and product-integration proposal covering Reader Calibration, import and listening parity, verifiable claims, the Reader Data Covenant, exact-edition source identity, and a staged path into Study and Research. The proposal is advisory. It does not modify the accepted Alpha baseline or authorize implementation.
Rev3 is a research pack and integration proposal. Nine documents (GL10 through GL18), one full product spec proposal, a competitive matrix, a claims register, and this website copy. It updates how Project Glosa describes itself; it does not change what has been built.
Rev3 is:
Rev3 is not:
The merge changed only documentation. Constitution: unchanged. First Principles: unchanged. Production code: unchanged. Alpha 0.1 and 0.2 history: unchanged.
Verification is available in the merge record. Every hunk applied cleanly against the verified baseline. Every SHA-256 in the file manifest reconciled.
Rev3 is layered on top of what has been accepted. It does not overwrite it. The Alpha's promises about silence, sovereignty, and source-first reading remain in force.
The author's authoritative text — exact edition, stable location, original record. Untouched and recoverable.
Your notes, questions, highlights, recollections, purpose, and interpretation. Private by default.
Generated context, inquiry, analysis, synthesis — typed, sourced, and withdrawable. Never silently merged.
The Assistance Dial still runs from Silent through Synthesis, and silence is still a complete product state. The reader-owned instrument is still the anchor of the design. What Rev3 changes is how those commitments are documented, evidenced, and staged — not what they are.
Rev3 proposes a small, transparent onboarding that asks the reader about the present session — the work in front of them, the time they have, whether they want to read or listen, and how much assistance may appear. Those answers become a reversible Reader Compass.
The Reader Compass is a session setting, not a personality profile. It is visible to the reader, editable at any time, and never used to infer identity, cognitive type, or long-term traits.
Reader Calibration chooses conditions for this work and this session. It does not claim to discover who the reader is.
Onboarding is asked to earn its attention:
Status: Prototype
Rev3 hardens a preference that has always been implicit in Project Glosa: fix the ordinary reading surface before adding another AI feature to it. Typography, position memory, listening continuity, reliable highlights, and stable margins are the ground on which everything else is judged.
A summary that arrives faster than the reader would like, over a page that scrolls unpredictably, is a worse product than one page-turn done well. Rev3 sets acceptance criteria that treat reader quality as the first-order metric — before any dial-3-and-above feature is considered ready.
A great mastering engineer does not flatten every moment into maximum brightness. Glosa treats AI the same way.
Concretely, this means AI does not: summarize every chapter without asking; paint automatic key points; impose emotion; invent certainty; take a persona pretending to be the author; lecture when a sourced whisper will do; or produce output when silence would serve the work better.
Rev3 formalizes what has been the working default. Glosa does not use your books, documents, notes, highlights, questions, voice, reading history, or private embeddings to train shared models unless you choose a specific contribution, review what will be sent, and consent for that exact purpose.
Contribution is separated into distinct, revocable choices — every one off by default:
Every cloud action shows what leaves the device, why, to whom, and for how long. Every session can produce a data receipt. Nothing is bundled into general consent.
No ads. No behavioral data sale. No shared-model training on private reading by default — subject to published feature-specific data receipts.
Rev3 publishes a claims register. Every cognitive or educational claim is graded — Established, Supported, Context-dependent, Exploratory, or Not claimed — with a source, a limitation, and a revision date.
Recalling for yourself beats re-reading for long-term memory, within task and interval limits. No universal guarantee.
A small paper advantage appears in some settings — especially expository reading and self-calibration. It is not a general rule.
Modality effects depend on material, task, reader, navigation, and measurement. Listening is fine for comprehension; self-paced reading tends to keep more.
Optional, silence-first. No comprehension, nervous-system, or focus-frequency claim is made.
Preferences configure the interface; Glosa does not diagnose a visual, auditory, or kinesthetic type. The idea failed the science.
Inspectional Mode accelerates selection and orientation — not impossible eye-speed promises.
AI offers attributed interpretations and alternatives. The reader keeps authorship.
Reverent Mode stays reader-controlled. Glosa does not infer spiritual states.
Rev3 sharpens a distinction that public-domain classics tend to blur: a title in the public domain does not make every translation, introduction, image, or recording public domain. Each Glosa Classic is tied to a specific approved edition, and the page shows the difference plainly.
For your own material, the architecture is the same: import preserves an immutable source record, creates stable reading locations, and tells you whether processing stayed local or used a cloud service.
Glosa will not bypass DRM or pretend uncertain OCR is authoritative text.
Import surfaces Rev3 addresses (as proposals, not commitments):
Status of import parity: Proposed for Personal
Rev3 arranges what Project Glosa may become as a series of gates — each independently operable, each entered on evidence, none required to reach the one before it. The public reader never has to become a research cockpit.
Approved public-domain editions, a serene page, margin context, basic listening, a local guest Vault.
EPUB, PDF, documents, web, scans, OCR, organization, export, optional encrypted sync.
Inspectional and analytical reading, recall, confidence, self-explanation, spacing, citations.
Syntopical comparison, terminology, claims, evidence, timelines, contradictions, dossiers.
Corpus lineage, authority tiers, interdisciplinary concept graph, experience testimony, research agents.
Voice, oral reflection, and shared attention — accessibility without impersonating lived experience.
Chronoaesthesis may be vast beneath the surface. Glosa remains quiet above it.
Publishing sequence, per the brief:
A preliminary naming and functional-collision scan flagged real risk. The GLOSE mark — a registered US trademark for a digital reading platform with functionally overlapping services — is the closest. A newly active Swedish App Store education app named Glosa, an earlier AI language-learning product at glosa.app, and a scattering of descriptive and common uses complete the picture.
The scan is preliminary and not legal advice. Its conclusion is careful: Glosa should remain the internal working name, and it is not cleared for public launch. A qualified trademark attorney is needed before broad public branding, store entry, or heavy investment.
Per founder Amendment 1 to the Rev3 approval: if marketplace prominence, audience overlap, or professional-clearance evidence indicates material risk to Glosa, the founder will review proportionate remedies before public adoption. Internal contingency names exist as private examples only; none is selected, cleared, or approved for publication.
What appears publicly today:
What does not appear publicly:
Status: Available in accepted Alpha for the disclosures shown here. Everything downstream awaits founder review.
A separate decision follows this checkpoint. The Rev3 canonical merge is complete. Wave-A architecture spikes remain unauthorized. The Classics beta implementation, staged website copy integration, and a naming-clearance pause are the bounded next options on the table — no option is selected by this document.