Unlisted · noindex · design plan

Roxy — Weekly Research Digest Plan

One concise, source-cited weekly summary · built & scheduled August 16, 2026 · updated August 16, 2026

This page details the "Weekly research digest" tentative card on Roxy's status page. Once a week — emailed to Bob at 3:00 AM Monday — Roxy summarizes what entered his library that week, every line cited to a real document, no new claims, honest when the week was quiet. It is a report of the library, not commentary on the market.

Status (August 16, 2026): built and scheduled. Bob opted in; the weekly job’s first real send is Monday August 17 at 03:00 ET to rbcorman@gmail.com, and a 65-line cited dry-run already went to James. Pause anytime with “pause digest.” Roxy’s first scheduled push — the on-demand digest proved out first, then the schedule.

What Bob gets

Every Monday at 3:00 AM, one short email to rbcorman@gmail.com: the research that landed in Bob's library since the last digest, grouped by publisher, each item a one-line gist with author, date, and a link straight to the source. It answers "what came in this week?" without Bob having to ask — a table of contents for the week's reading, waiting in his inbox before Monday starts, not a rewrite of it.

Roxy — Weekly Research Digest · Aug 10–16, 2026

6 new documents from 4 publishers this week.

Doomberg — "Title of the piece" · Aug 12 · natural-gas storage and the winter setup. source

Doomberg — "Another piece" · Aug 14 · follow-up on LNG export capacity. source

Publisher B — "Title" · Aug 11 · datacenter power and grid interconnect queues. source

Nothing new from three of your usual publishers this week.

The one real design choice: this is a push, not a pull

Every Roxy feature so far answers when Bob asks. The digest speaks on its own, on a schedule — that is the real complexity here, not the code. A pull product can never nag; a push product can. So the digest needs three things the pull features never did: Bob's explicit opt-in, a standing weekly job (Roxy's first — Monday 3:00 AM, emailed to rbcorman@gmail.com), and an easy pause. The build treats the summary itself as the first, cheapest piece and the schedule as a deliberate second step — matching the rule that a feature needing a new store, a new job, and new behavior all at once should ship them one at a time.

How Roxy builds the digest

Bounded window. Everything ingested since the last digest (default: the trailing 7 days), read from the library's dates — no open-ended lookback.
Grouped, then listed. New documents grouped by publisher/author, each a single cited line: title, author, date, one-line gist, link. This reads far better with the author/date metadata from the smarter-retrieval plan in place — the digest is a natural consumer of that work.
Gist, not analysis. The one-line gist describes what a document is about, drawn from the document itself. It is not a new claim, a rating, or a market call.
Cited or it doesn't ship. Every line links to the document it names. A line with no source does not go in the digest.
Honest about quiet weeks. A slow week says "3 new documents" or "nothing new this week" — it is never padded with old material or outside news to look busy.

The build plan (how Roxy builds and proves it)

Built in the same receipts-first spirit as the other plans: Roxy generates real digests over the real corpus and shows the evidence before anything gets scheduled or sent unprompted.

Phase	What Roxy builds	How Roxy proves it (test)	Gate to advance
1. On-demand digest (pull first)	Generate "digest of the last 7 days" when Bob asks — grouped, cited, one line per document. No schedule yet.	Roxy produces a digest over a real past week and, for each line, shows the document it came from; the document count matches a direct query of what was ingested in that window.	Every line traces to a real document; the count matches the library. No line without a source.
2. Quiet-week & gap honesty	Handle empty and broken weeks correctly.	Roxy runs the digest over a week with zero new documents (returns "nothing new this week," no padding) and over a week where ingest was down (says coverage may be incomplete rather than implying a quiet week).	Empty week is honest and unpadded; a known ingest gap is disclosed, not hidden.
3. Schedule it (Roxy's first standing job)	Add the weekly job (cron: 3:00 AM Monday) that generates the digest and emails it to `rbcorman@gmail.com` from Roxy's Gmail; opt-in, with a one-word pause.	Roxy shows the job fires at the scheduled Monday 3:00 AM slot (a dry-run email to James first, before it goes to Bob), produces the same cited digest as the on-demand path, and can be paused and resumed on command.	Scheduled digest matches the on-demand one; fires at Monday 3:00 AM (Bob's local time); pause/resume works; Bob has opted in.
4. Delivery & rails check	Confirm the email lands at `rbcorman@gmail.com` and that the digest stays within research-only rails.	Roxy sends a real digest email to `rbcorman@gmail.com`, containing only library-sourced, cited lines — no new claims, no sentiment, no buy/sell language, no outside-news items — Bob confirms receipt, and Roxy confirms nothing new is written to the library by the digest itself.	Digest is read-only over the library and free of claims/commentary. Build complete.

What this deliberately does not include

No new claims, ratings, or investment conclusions. The digest reports what arrived and what each item is about; it never tells Bob what to think or do about it.
No sentiment scoring. No "bullish/bearish" tags — that is an unsourced judgment dressed as data.
No outside-news firehose. The digest covers Bob's own library only, never a scrape of the wider internet. A news firehose before the library is trusted is on the kill list.
No re-summarizing at ingest. The one-line gist is generated at digest time from the stored document, not baked into the library where it could drift from the source. (Same reason smarter retrieval rejects summarization-at-ingest.)
No daily or real-time cadence. Weekly, one message. Not a feed.

When to do it

After the library proves useful to Bob — which is the same gate the status card already names. Cited answers are proven; the missing signal is Bob actually relying on them over time. The digest is worth far more once the author/date metadata is populated (it groups and dates the week cleanly), so if smarter retrieval is on the horizon, its Phase 1 metadata is a natural predecessor. Order among the remaining tentative features is still Bob's call — this stays tentative until he wants a weekly report.