patent.fyi
Applications  /  US 2026/0355636
United States Patent Application Publication

US 2026/0355636 A1

Published
Nov 12, 2026
Filed
May 7, 2025
Earliest date on the face
May 7, 2025
Classification
A63F 13/67
First published here
Nov 19, 2026

Generated automatically. Not reviewed by a person. Not legal advice or an opinion.

Section 1 · machine-generated

What if?

Questions about how individual parts of claim 1 might be changed. A question may describe something different from what the claim language recites, or a version of it; no view is offered on which.

1.1Live counters in memory

What ifWhat if the recap generator reads a snapshot of live counters from game memory when the session ends?
Claim language“a database storing video game log data”

The game updates a fixed-size memory buffer with the score, objective completion counts, and the elapsed time of the last lead change as play proceeds. At session end, the game passes the buffer to the recap generator, which turns those values into a recap.

1.2Complete log in ordered batches

What ifWhat if the system sends every stored session record through the summarization pipeline?
Claim language“selecting a subset of video game log data”

A database cursor reads every record in session sequence order and packs the serialized records into batches within the model's input limit. The model summarizes each batch, and a final pass combines the batch summaries into one recap.

1.3Statistics as model input

What ifWhat if the model receives computed session statistics?
Claim language“inputting the selected subset of video game log data”

An aggregation process computes total points, the number of lead changes, and the longest scoring streak from the event records. The model receives a fixed vector of named numeric fields and uses those values to write the recap.

1.4Recap built from templates

What ifWhat if a rule engine builds the recap from sentence templates and recorded event values?
Claim language“instructing the generative machine learning model to synthesize a summary”

The engine ranks event types using a priority table and breaks ties by timestamp. It fills templates with the final score and the times and outcomes of the three highest-ranked events, then displays the event sentences in chronological order.

1.5Event identifiers drive rendering

What ifWhat if the application receives event identifiers from the model and assembles the recap itself?
Claim language“receiving the synthesized summary”

The model returns an array of up to five event identifiers from the supplied records. The application validates each identifier against the session log, retrieves the matching records, and renders a timeline with timestamps and labels determined by event type.

Section 2 · machine-generated

Variations

Further arrangements of the same subject matter, described here as of the publication date at the foot of this page.

Other field of use

Operator recap for crane simulator training

A crane training simulator stores control positions, load swing measurements, proximity alerts, and instructor markers in a database keyed by training run. After a run, a selector extracts time windows around instructor markers and crossings of configured swing thresholds, then orders the records by timestamp. A generative language model receives the selected records and writes a chronology of operator actions and measured responses for the trainee's debrief screen.

Component substitution

Template engine for session recaps

A finite-state transducer takes the place of the generative model in the recap service. The selector retrieves records for a completed session and converts them into ordered symbols for events such as checkpoint completion, retries, and score changes, with values attached to each symbol. Transition rules group consecutive retries and emit sentence templates populated with checkpoint names, retry counts, and score deltas. The recap service joins the emitted sentences and displays them to the player.

Parameter or scale variation

Hierarchical recaps for long sessions

A recap worker divides completed sessions of up to eight hours into 15-minute intervals. For each interval, it selects up to 120 stored records using event priority and timestamp, then sends the records in chronological order to a generative language model with a 150-word output limit. A second model pass receives the interval summaries and their time ranges and generates a player recap capped at 400 words.

Combination with existing systems

Recaps linked to indexed match replays

A match replay service stores replay checkpoints and player inputs indexed by simulation tick, and the session log stores an event identifier and simulation tick for each recorded event. The recap service sends selected log records to a generative language model and requests a structured response with a source event identifier for each bullet. The client looks up each returned identifier in the selected records and uses its tick to create a replay link. When the player opens a link, the replay service loads the nearest checkpoint at or before the target tick and applies recorded inputs to reach that tick.

Session recaps in a coaching dashboard

A coaching dashboard stores player practice goals and their numerical targets by player identifier, and a session database stores timestamped drill measurements under the same identifier. When the player opens a completed drill in the dashboard, a recap service retrieves the active goal and selects the session records for the corresponding metric. The service calculates the session result and its difference from the target, then sends those values and selected records to a generative language model with instructions to describe the result and its progression. The dashboard displays the generated recap beside the measured result and target.

Automated or data-driven variant

Event selection from player feedback

The recap interface records whether the player marks a recap bullet as useful and associates each rating with the source event identifiers for that bullet. A gradient-boosted tree model trains on those ratings using event type, recurrence count, and elapsed time as features. Before generating the next recap, the service scores candidate records from the completed session and selects the highest-scoring records up to the input limit. The service orders the selected records by timestamp and sends them to a generative language model, then displays the resulting recap.

Specific version of a part

Indexed session event lookup

A relational database stores each event with a session identifier, player identifier, sequence number, and event payload. The recap service uses a composite index on session identifier and sequence number to retrieve a completed session's events in order.

Event windows around score changes

A query locates each change in the leading side and selects records within five seconds before and after each change. The selection process removes duplicate records from overlapping windows and caps the result at 200 records ordered by timestamp.

Compact records in timestamp order

The service serializes each selected event as one line with elapsed milliseconds, event type, actor, and outcome. It concatenates the lines in timestamp order and sends them in one model request.

Player recap in match history

The game service stores the recap under the completed match entry in the player's match history. When the player opens that entry, the client retrieves the recap and renders it below the final score.

Section 3 · quoted from USPTO records

Other work

Passages reproduced word for word from two earlier-published U.S. patent documents, placed beside the language of claim 1. No characterization is offered. Each passage is marked A or B for the document it comes from; ¶ numbers are the paragraph numbers in the USPTO’s published full text of that document.

Document AUS 11,003,704 B2
Published
May 11, 2021
Filed
Nov 26, 2019
Class
G06F 16/345
Document BUS 2024/0152689 A1
Published
May 9, 2024
Filed
Jan 11, 2024
Class
G06F 40/166
Claim 1
Document A
Document B
1.1accessing a database storing video game log data relating to a previous video gaming session involving a video game player playing a video game;
During the conversation, a conversation manager 112 records conversation data (e.g., voice data, video data) transmitted between the client devices of the customer 118 and agent 114 as part of the communication session and stores the conversation data in a conversations database 110 of the communications service provider 102.
B · US 2024/0152689 A1 · ¶ 26
1.2selecting a subset of video game log data from the database for summarization by a generative machine learning model;
Extractive models form summaries by copying parts of the input without any modification, selecting relevant phrases of the input document, and concatenating them to form a summary.
A · US 11,003,704 B2 · ¶ 7
Extractive summarization seeks to extract grammatically valid subsets of the source document such that, when combined, the subsets produce a coherent, shorter text.
B · US 2024/0152689 A1 · ¶ 3
1.3inputting the selected subset of video game log data to the generative machine learning model;
Each iteration comprises providing, to the machine-learning model, the source text and the running summary, and adding, using the machine-learning model, a new word to the running summary.
B · US 2024/0152689 A1 · abstract
1.4instructing the generative machine learning model to synthesize a summary of the previous video gaming session based at least on the selected subset of video game log data;
A system for text summarization includes an encoder for encoding input tokens of a document and a decoder for emitting summary tokens which summarize the document based on the encoded input tokens.
A · US 11,003,704 B2 · abstract
1.5receiving the synthesized summary of the previous video gaming session from the generative machine learning model; and
Architecture 100 also includes summary word emitter 192, which provides pointer data for output summary tokens 194 and document summary data store 198 for storing results for abstractive text summarization architecture 100.
A · US 11,003,704 B2 · ¶ 39
A method comprising: by one or more processors, accessing text; inputting, by the one or more processors, the text into a machine-learning model trained to output a word to be added to a summary based on the inputted text; and adding, by the one or more processors, the outputted word to the summary.
B · US 2024/0152689 A1 · claim 1
1.6outputting the synthesized summary of the previous video gaming session to the video game player.
In one example, the method 600 further comprises causing presentation of the summary on a display.
B · US 2024/0152689 A1 · ¶ 108

An empty cell means only that no passage was reproduced there.

Statements in these documents
Training a model that can generate long, coherent, and meaningful summaries remains an open research problem.
A · US 11,003,704 B2 · ¶ 5
By condensing large quantities of information into short, informative summaries, summarization can aid many downstream applications such as creating news digests, search, and report generation.
A · US 11,003,704 B2 · ¶ 6
A weakness of traditional sequence-to-sequence learning, when applied to summarization, is the lack of a direct copy mechanism, leading to missing or misrepresented details in decoded summaries.
B · US 2024/0152689 A1 · ¶ 4
An attention mechanism helps ameliorate this issue by directly learning to focus on specific words or phrases in a source document.
B · US 2024/0152689 A1 · ¶ 4
Cite aspatent.fyi, US 2026/0355636 A1, claim 1, first published 2026-11-19T13:37:15-05:00, https://patent.fyi/US20260355636/, SHA-256 76e42fb578c923e54efabfe983d99428a8dbed8841e66fa4ba4c3b530431561e
First published
Permanent addresshttps://patent.fyi/US20260355636/
Content fingerprint (SHA-256)76e42fb578c923e54efabfe983d99428a8dbed8841e66fa4ba4c3b530431561e
Methodv2026-11-18.2 · automated, not reviewed by a person
Correctionscorrections@patent.fyi