Market dashboards assume the user is looking at a screen. But many research moments happen while walking, commuting, exercising, or doing another task. The obvious product response is audio. The less obvious engineering question is: where should the audio briefing be assembled?
For Ticker Watch’s single-ticker workflow, the research is already in the browser. Headlines, summaries, direction context, impact labels, SEC filing links, and the selected ticker are present. Sending that same material to another ShareCapsule service just to reconstruct a transcript would create an additional content pipeline without adding evidence.
From research cards to spoken segments
The audio layer does not scrape arbitrary page text. It extracts structured pieces from the rendered research interface and turns them into ordered segments.
This architecture keeps the transcript inspectable. The user can see exactly what the browser intends to say before or while listening, and source links remain available for anything that deserves deeper reading.
Why there are 3, 5, and 10-minute formats
Different research moments need different levels of compression. We introduced three target formats:
| Format | Goal | Evidence behavior |
|---|---|---|
| 3-minute quick scan | Fast orientation | Prioritize a few highest-impact or directional stories |
| 5-minute research brief | Default review | More stories, more context, selected SEC evidence |
| 10-minute deep dive | Longer research session | Broader evidence set, additional context and filings |
The word target is important. If only two meaningful sourced items exist, the 10-minute mode should not pad the briefing with invented commentary merely to hit a duration target. Research Radio estimates runtime from the actual generated words and selected playback speed.
This creates a useful product principle: duration is subordinate to evidence.
Prioritization before narration
Audio is sequential. A visual dashboard lets the eye jump around; a spoken briefing forces an order. That means prioritization becomes part of the audio architecture.
Ticker Watch sorts toward higher-impact evidence and gives additional priority to clearly directional news. Longer formats can include more lower-priority material and SEC filings. The purpose is not to predict price movement. It is to put the evidence most likely to deserve attention earlier in the listening sequence.
Every spoken story retains the same conceptual boundaries as the visual product:
- what happened;
- how the company-news direction is labeled;
- how significant the event appears for research prioritization;
- why it may matter;
- where to open the original source.
Changes-only audio
The most interesting audio mode appears when Research Radio is combined with encrypted research memory. Plus users can choose Changes only, which reads the newly surfaced items currently shown by “Since I last checked.”
That creates a compact workflow:
The checkpoint itself remains local to the device in the current design. The audio module consumes the already-rendered delta rather than sending the checkpoint to an audio backend.
Why browser speech synthesis?
Browser speech synthesis has attractive privacy and cost properties for this use case. It can speak a locally constructed transcript without requiring ShareCapsule to upload the transcript to a text-to-speech service merely for playback.
But browser speech is not perfectly uniform:
- voice availability differs by operating system and browser;
- some reported voices are local while platform defaults may use remote services;
- pause/resume behavior can vary;
- mobile browsers may impose user-interaction restrictions;
- voice quality is not consistent across devices.
For that reason, the product says it prefers a local system voice when the browser reports one. It does not claim that every possible browser default is guaranteed local.
The transcript is part of the trust model
A good market podcast can sound authoritative. That is precisely why the visible transcript matters. Ticker Watch keeps the spoken segments on the page, along with the source links. The user can inspect the wording instead of receiving a disembodied voice with no provenance.
The closing segment also repeats the product boundary: direction and impact organize research; they do not predict price movement or recommend a trade.
Free audio vs paid automation
We intentionally keep the core single-ticker Research Radio formats in the free experience. They operate on research the user already loaded and demonstrate the product’s information-compression value.
Plus focuses on continuity and automation:
- changes-only audio from encrypted research memory;
- whole-watchlist briefings;
- scheduled briefings;
- briefing history and advanced alert workflows.
That commercial boundary matches the technical one. Free answers, “Help me understand this ticker now.” Plus increasingly answers, “Keep track of my research workflow over time.”
Failure modes we care about
Audio overconfidence
Spoken language can make a weak interpretation sound certain. Wording must preserve uncertainty and label meanings.
Source drift
If the underlying rendered evidence changes, the transcript should be regenerated rather than continuing to speak stale segments.
Artificial padding
A target duration must never become a reason to fabricate extra market commentary.
Voice privacy ambiguity
The application cannot honestly describe every browser’s default speech path as local. The UI therefore reports what it knows about local voice availability and states the limitation.
What this means for investors
The value is not “AI radio” as a novelty. It is research compression with provenance. A user can choose a short scan, a deeper briefing, or a changes-only catch-up, then open the original sources when something deserves attention.
What comes next
The next engineering problems are larger-scale continuity: whole-watchlist “since last checked,” encrypted cross-device research-memory synchronization, more robust source identity, and eventually normalized research APIs that preserve the same minimum-data and provenance boundaries.