GazeStill · drawing set · v0.2

Part I · Cover Docket · Abstract · v0.1 skeleton

U.S. Provisional Patent Application · Drawing set

Docket GS-2026-A

Sheets 5

Rev. B · v0.2

Title · Method and system for structured re-orientation upon resumption of a long-form reading interface, comprising a confidence-banded paragraph range with last-attentive sentences and a next-sentence preview, derived from a webcam-inferred disengagement event

Inventor · J. H. Kim

Internal · GazeStill

Filing target · pending pilot gate (≥80% bookmark accuracy across 3 articles) + re-entry study

Classification (proposed) · G06F 3/01 · G06F 16/957

About this Personal project · v0.2 card-centered Pilot-gate pending

Draft cover sheet and drawings for GazeStill, the anchor of the attention subtree in the portfolio (PageHaptic is its haptic-output descendant). The prototype lives in a companion repo; this page advances toward a full provisional draft once the pilot gate clears — ≥ 80 % of deliberate mid-article attention breaks land the bookmark within ± 2 paragraphs across 3 articles. The disengagement-detection mechanic is squarely anticipated by US20060256083A1 (Rosenberg / Outland, 2006), its issued counterpart US7429108B2, and US8643680B2 (Amazon); the contribution claimed here is downstream of detection — the re-entry card shown on resume (FIG. 5). The patent-draft format forces that distinction to be made on the cover, not buried in the discussion.

Abstract of the disclosure Cover-sheet boilerplate v0.2

A system and method assist a reader in resuming a long-form document at the point of inferred attention collapse by rendering, on resumption, a structured re-orientation card rather than scrolling to a single anchor. A webcam integral to the host device drives an on-device gaze regressor; an off-text-dwell detector, normalized by per-paragraph difficulty, identifies a disengagement event and a back-dated bookmark paragraph — both mechanics being acknowledged building blocks from the prior-art gaze-bookmark literature (Rosenberg 2006; Amazon 2014). The disclosed contribution is the re-entry card dispatched at the next CLOSED→RESUMED transition: it presents (i) the last two sentences read attentively before the break, (ii) the first sentence after the bookmark as a forward preview, (iii) a confidence band visualised across the paragraph range over which attention was decaying, and (iv) an explicit "Resume here" affordance that commits the scroll. Prior gaze-bookmark systems place an anchor and return the reader to it; the disclosed system instead surfaces the context immediately surrounding the break so that re-engagement does not require the reader to mentally rebuild lost context. All inference is performed on the host device; raw video frames are discarded after gaze regression, and no per-user identifier, raw video, or gaze trace is transmitted off-host.

Field

G06F 3/01 · user input

Acknowledged prior art (detection)

US20060256083A1 — Rosenberg / Outland · look-away place-marker
US7429108B2 — issued; place-marker on look-away
US8643680B2 — Amazon · gaze-dwell bookmark

Distinguished from (re-entry UX)

Rosenberg / Outland — single-anchor place-marker, no re-entry context
Amazon dwell-bookmark — anchor only, no contextual restore
Kindle Page-Flip — scroll-position heuristic, no attention model
Pocket / Instapaper — explicit save, no resume scaffold

Index of sheets Tap a row to jump 5 sheets

Sheet 1 FIG. 1 Reader view · disengagement on ¶4 (input to FIG. 5) 100
Sheet 2 FIG. 2 Disengagement detector · off-text dwell vs threshold τ_d 200
Sheet 3 FIG. 3 Pipeline · webcam → fixations → disengagement → re-entry card 300
Sheet 4 FIG. 4 Session state machine · WRITE on disengage · READ on resume 400
Sheet 5 FIG. 5 Re-entry card anatomy · last attentive · next · band 500

Part II · Drawings FIG. 1 – 5 · Sheets 1 – 5

Sheet 1 / 5 FIG. 1 · Reader view · disengagement on ¶4 (input to the re-entry card) 100

FIG. 1

Sheet 2 / 5 FIG. 2 · Disengagement detector · off-text dwell vs threshold τ_d 200

FIG. 2

Sheet 3 / 5 FIG. 3 · Pipeline · webcam → fixations → disengagement → re-entry card 300

FIG. 3

Sheet 4 / 5 FIG. 4 · Bookmark resume state machine 400

FIG. 4

Sheet 5 / 5 Representative FIG. 5 · Re-entry card anatomy · last attentive · next sentence · confidence band 500

FIG. 5

Part III · Specification Background · Summary · Brief description

Background of invention Prior-art context

Webcam-derived gaze estimation is well-studied. WebGazer.js (Brown / Papoutsaki et al., IJCAI 2016) and other regression-based estimators achieve roughly 100–150 px error (~4° visual angle) under reasonable lighting, sufficient for paragraph-level but not line-level resolution at typical 15-inch laptop viewing distances. Dedicated eye-tracking hardware (Tobii, EyeTribe) achieves substantially better accuracy at the cost of additional hardware. Both classes are predominantly used for analytics — heatmaps, gaze plots, attention dwell-time — rather than as a control input to user-facing application logic.

The detection mechanic of writing a bookmark on a gaze-derived look-away event is squarely anticipated by prior art: US20060256083A1 (Rosenberg / Outland, 2006) and its issued counterpart US7429108B2 disclose a gaze-responsive interface that determines a look-away event and displays a place-marker at or near the word the reader was viewing; US8643680B2 (Amazon, Selker & Baldwin, 2014) discloses gaze-dwell bookmarking and gaze-derived page-turn detection. The disclosed method does not claim the detection mechanic over these references and explicitly relies on them as enabling building blocks for the contribution that follows.

What the prior art does not address is what is rendered at the next CLOSED → RESUMED transition. Rosenberg / Outland places a single graphical marker (highlight, outline, underline) on the lost word, and Amazon scrolls to or annotates the dwell point. The reader, returning hours or days later, is given a location but no context: the surrounding text immediately preceding the break is no longer easily distinguishable from later text, and the reader must mentally rebuild the lost reading context from cold.

The disclosed contribution — the load-bearing framing of this disclosure — is a structured re-orientation card that the system constructs and renders on resume. The card presents (i) the last two sentences read attentively before the disengagement event, (ii) the first sentence after the bookmark as a forward preview, (iii) a confidence band visualised across a paragraph range over which attention was decaying, and (iv) an explicit commit affordance. The card converts a known-location bookmark into a re-engagement scaffold; it is independent of the specific detection mechanic used upstream.

The detection pipeline disclosed herein — webcam → on-device gaze regressor → off-text dwell over a paragraph hit-test, normalised by per-paragraph Flesch difficulty, with a precursor back-date to the latest paragraph receiving two or more fixations — is one practical realisation that produces the inputs the card requires. Any equivalent disengagement detector from the prior-art family above could be substituted without changing the card's structure or claims.

Summary of the invention per 37 CFR § 1.73

The disclosed system constructs and renders a structured re-orientation card (502) at the next CLOSED → RESUMED transition of a long-form reading interface, in response to a disengagement event recorded earlier in the session. The card comprises (a) a confidence-band note (504) reporting the system's estimated confidence and paragraph range over which attention was decaying; (b) a last-attentive blockquote (506) presenting the last two sentences read attentively prior to the disengagement event, extracted from the bookmarked paragraph; (c) a next-sentence blockquote (510) presenting the first sentence of the following paragraph as a forward preview; and (d) an explicit commit affordance (514) that, when actuated, scrolls the underlying document to the bookmarked paragraph and dismisses the card.

The disengagement event consumed by the card is produced by a webcam-derived pipeline running on the host device. The webcam (102) feeds an on-device gaze regressor (304); a dwell window (306) accumulates an off-text dwell signal whose threshold τ_d is normalised by per-paragraph Flesch readability; on first upward crossing of τ_d, a precursor back-date step selects as the bookmark the most recent paragraph that received at least two fixations within the active sliding window. The detection pipeline is acknowledged to overlap with prior-art gaze-bookmark systems; the card constructor (308) and the card structure itself (FIG. 5) are the disclosed contribution.

Brief description of drawings Sheets 1 – 4

FIG. 1Reader view: webcam 102, page 104, fixation cluster decaying from ¶1 → ¶4, gutter bookmark 106 at the disengagement paragraph. Shown as the system anatomy that feeds the re-entry card (FIG. 5).
FIG. 2Off-text dwell signal 202 over elapsed reading time, with threshold 204 τ_d = 3.0 s (Flesch-normalised across 1.8 s – 5.4 s). First upward crossing dispatches the disengagement event 208; precursor back-date 206 selects the bookmarked paragraph as the most recent paragraph receiving ≥ 2 fixations.
FIG. 3On-device pipeline blocks 302 webcam → 304 gaze regressor → 306 dwell window → 308 re-entry card constructor → 310 store. Frames are discarded after gaze regression.
FIG. 4Session state machine 400 with states READING 402, DISENGAGED 404, RESUMED 406, CLOSED 408. Card is constructed on entry into DISENGAGED and rendered on entry into RESUMED.
FIG. 5Re-entry card anatomy (the disclosed contribution). Card 502 with confidence-band note 504, last-attentive blockquote 506 (last two sentences of ¶bookmark), next-sentence blockquote 510 (first sentence of ¶bookmark+1), commit button 514. Confidence band 508 over range_paragraph_ids is rendered on the underlying article. Bookmark bracket 512 anchors the card to ¶bookmark.

Part IV · Claims 7 total · 1 indep · 5 dep · 1 apparatus

Claims 1 independent · 5 dependent · 1 apparatus Draft v0.2

What is claimed is:

1. A method for restoring a reader's engagement upon resumption of a long-form reading interface following an inferred attention break, comprising:

(a)receiving, at a host computing device, a disengagement event identifying a bookmark paragraph and a confidence-banded paragraph range over which a reader's attention was decaying prior to the event;
(b)upon a CLOSED → RESUMED transition of said reading interface, constructing a re-orientation card (502) comprising at least: (i) a last-attentive blockquote (506) presenting the final one or more sentences of the bookmark paragraph; (ii) a next-sentence blockquote (510) presenting at least the first sentence following the bookmark paragraph; (iii) a confidence-band note (504) reporting the said confidence and range; and (iv) a commit affordance (514);
(c)rendering said re-orientation card above the bookmarked paragraph in the document layout, and visually distinguishing the confidence-banded paragraph range (508) within the rendered document;
(d)upon actuation of the commit affordance, dismissing the re-orientation card and scrolling the underlying document to the bookmark paragraph; and
(e)wherein the re-orientation card surfaces both the reading context immediately preceding the disengagement event and a forward preview, such that re-engagement does not require the reader to rebuild context from a single anchor.

2. The method of claim 1, wherein the disengagement event of (a) is produced by an on-device pipeline comprising a webcam-fed gaze regressor (304), an off-text dwell window (306) whose threshold τ_d is normalised by per-paragraph readability, and a precursor back-date step selecting as the bookmark paragraph the most recent paragraph receiving at least two fixations within the active window.

3. The method of claim 2, wherein per-paragraph readability is computed from a Flesch reading-ease score and the multiplier applied to τ_d is bounded between 0.6 and 1.8.

4. The method of claim 2, wherein each captured video frame is discarded after gaze regression, and no raw video frame, no gaze-location stream, and no per-user identifier is persisted on, or transmitted from, the host device.

5. The method of claim 1, wherein the disengagement event of (a) is received from any disengagement detector, including the place-marker-on-look-away detector taught by US7429108B2 or the gaze-dwell detector taught by US8643680B2; and wherein the contribution of the present claim resides in the construction and rendering of the re-orientation card at (b) and (c).

6. The method of claim 1, wherein the bookmark resolution is paragraph-level and the confidence band of (b)(iii) spans one to four contiguous paragraphs.

7. A host computing device comprising a webcam, a display, one or more processors, and non-transitory memory storing instructions which, when executed, perform the method of any one of claims 1 – 6.

Claims · 7 total · 1 independent · 5 dependent · 1 apparatus

Part V · Appendices Rejection forecast · Bibliography

Prior-art rejection forecast Element-by-element novelty argument v0.2

The disclosed claims are drafted on the assumption that an examiner would cite US7429108B2 (Rosenberg / Outland; gaze-responsive place-marker on look-away) and US8643680B2 (Amazon; gaze-dwell bookmark and gaze-derived page turn) as the closest art. Below, claim 1 is read element-by-element against each reference and the surviving limitations are identified.

Claim 1 element	US7429108B2 (Rosenberg)	US8643680B2 (Amazon)	Surviving
(a) receive disengagement event	taught — look-away event is the trigger	taught — gaze-dwell and absence-of-gaze events	no — preamble overlap
(b)(i) last-attentive blockquote · last 1+ sentences of bookmark ¶	not taught — Rosenberg places a graphical marker on or near the lost word; no sentence extraction or quoted display	not taught — Amazon highlights or scrolls to the dwell location; no sentence-level context replay	yes — primary novelty
(b)(ii) next-sentence blockquote · first sentence after bookmark	not taught	not taught	yes — primary novelty
(b)(iii) confidence-band note · confidence + paragraph-range disclosure	not taught — Rosenberg's marker is a single point with no decay-range representation	not taught — Amazon offers no confidence/range UI surface	yes — novel
(b)(iv) explicit commit affordance	not taught — Rosenberg's marker is passive; no commit step	partial — Amazon allows gaze gestures (wink, dwell) to commit; the disclosed claim requires an explicit pointer-actuated affordance, not a gaze gesture	yes — narrowed against Amazon by requiring non-gaze actuation
(c) visually distinguish range within underlying document	not taught — marker is at a point	not taught — Amazon scrolls but does not paint a band	yes
(d) commit dismisses card and scrolls to bookmark	n/a (no card)	n/a (no card)	yes — depends on card existing

Anticipated 35 U.S.C. § 102 / § 103 posture. No single reference discloses elements (b)(i)–(iv) together; the closest combination would require importing the “card-on-resume” concept from a non-gaze reading-aware system (e.g., an e-reader summary widget) and substituting Rosenberg's look-away anchor for the reader's manual cue. That combination is not motivated by the cited art: Rosenberg's place-marker is itself the disclosed assist (the reader finds the lost word from the marker alone), and Amazon's bookmarks favor gaze-gesture commit (wink, dwell, nod) rather than scaffolded re-orientation rendered at a CLOSED→RESUMED transition. The disclosed claim therefore reads as non-obvious under KSR over the cited combinations available at filing. Confidence in this analysis is bounded by a v0.2 reading of the two references; a search-report-quality novelty study has not been commissioned.

Acknowledged exposure. Amazon describes "highlighting" and "annotation" surfaces around a dwell point in several embodiments, and an examiner could press one of those embodiments against claim element (c). The mitigation is the band-over-range structure of (b)(iii) and (c) — Amazon's annotation is point-anchored at the dwell location; the disclosed band is range-anchored over range_paragraph_ids derived from a per-paragraph decay model. Should the examiner press the point, dependent claim 6 (band spans 1–4 paragraphs) and claim 1(c) ("visually distinguishing" the range, not annotating a point) carry the narrowing.

Prior-art bibliography Selected; not exhaustive

Patents (detection mechanic, acknowledged prior art)

Rosenberg, L. B. Gaze-responsive interface to enhance on-screen user reading tasks. US20060256083A1 (Outland Research, 2006). Place-marker displayed on look-away event.
Rosenberg, L. B. Gaze-responsive interface to enhance on-screen user reading tasks. US7429108B2 (issued 30 Sep 2008; assigned to Outland Research; later assigned to Google, 2011). Issued counterpart of the above.
Selker, E. J. and Baldwin, L. B. Gaze-based content display. US8643680B2 (Amazon Technologies, 2014). Gaze-dwell bookmark and gaze-derived page-turn detection.

Gaze regression and fixation identification

Papoutsaki, A., Sangkloy, P., Laskey, J., Daskalova, N., Huang, J., and Hays, J. WebGazer: Scalable Webcam Eye Tracking Using User Interactions. IJCAI 2016. ~4° / ~130 px visual-angle error.
Salvucci, D. D. and Goldberg, J. H. Identifying fixations and saccades in eye-tracking protocols. ETRA 2000. I-DT (dispersion-threshold) fixation identification, as used in src/fixation.ts.

Reading-attention and mind-wandering signals

Bixler, R. and D'Mello, S. Automatic Gaze-Based User-Independent Detection of Mind Wandering During Computerized Reading. User Model. User-Adapt. Interact. 2016. Source of the fixation-duration precursor signal targeted by src/disengagement.ts.
Buscher, G., Cutrell, E., and Morris, M. R. What Do You See When You're Surfing? Using Eye Tracking to Predict Salient Regions of Web Pages. CHI 2009.
Hyrskykari, A. Eyes in Attentive Interfaces: Experiences from Creating iDict, a Gaze-Aware Reading Aid. PhD dissertation, University of Tampere, 2006. Earliest forms of gaze-derived reading support.

Reading-aware UI (non-gaze)

Amazon Kindle Page Flip. 2016–. Scroll-position heuristic; no attention model.
Pocket / Instapaper. Explicit save-for-later; no attention model.
Tobii. Eye tracker product line. 2001– (Tobii Pro research division spun off internally in 2014). Dedicated hardware.

Part VI · Execution Version · v0.2 · Card-centered

Version history Draft · not filed

v0.12026-01-30 · Skeleton draft. Detection-centered claims; prior-art appendix omitted the anticipating gaze-bookmark patents.
v0.22026-05-18 · Pivoted to re-entry card as the disclosed contribution. Added FIG. 5 (card anatomy, representative). Reconciled FIG. 2 numerics with src/disengagement.ts (off-text dwell, τ_d = 3.0 s Flesch-normalised). Added US20060256083A1, US7429108B2, US8643680B2 to background and bibliography. Claim 1 rewritten around card construction; claim 5 explicitly accepts any prior-art detector. Bibliography reorganised into labelled subsections.

Promotion to a full provisional draft remains conditioned on the pilot gate. The card's value — not the detector's accuracy — is the falsifiable claim, and the planned within-subjects study (SCROLL / LOOKAWAY / RE-ENTRY) at experiment-re-entry.md is the next concrete step after the pilot.

/gazestill · v0.2 · card-centered draft

Index