The LinkedIn signal index

Use GPT-6 Astra computer use to build a LinkedIn signal index from the account already logged in to the browser.

Objective

Find 10 organic LinkedIn posts from 10 different authors with at least 400 visible comments. Open every qualifying post, preserve the copy and visual evidence, inspect the visible conversation, and turn the strongest patterns into original content opportunities that fit the account owner's business.

Operating rules

- Work through LinkedIn's visible interface. Do not use a LinkedIn API, scraper, browser extension, export, or extra software.
- Keep every action on LinkedIn read-only. Do not react, comment, repost, follow, connect, message, submit forms, or change account settings. Local writes are permitted for the requested report files, screenshots, saved visuals, and ZIP.
- Start from the logged-in account's profile. Use its headline, About section, current role, featured work, and recent posts to infer the business, audience, offer, point of view, and topics. Label uncertain conclusions as assumptions.
- Continue without asking setup questions unless a missing fact would materially change the research direction.
- Inspect at most 150 organic feed posts.
- Save each completed record to files before continuing. Keep IDs, URLs, counts, labels, and missing fields in the working summary. Run the cross-post analysis after the set is collected.
- Retry a failed page interaction twice. If login, a CAPTCHA, or direct user action is required, preserve completed work and state the one action needed.

What counts

Include an organic feed post only when the opened post shows at least 400 comments. It may be authored or reposted by a connection, shown because a connection reacted or commented, published by a followed account, or suggested by LinkedIn.

Exclude Sponsored or Promoted posts, ads, job cards, non-post modules, duplicates, a second post from an author already selected, and any post whose opened comment count cannot be verified.

For every included post, save LinkedIn's surfaced-by wording when it appears. Also assign one label: direct connection, followed author or page, connection reaction, connection comment, connection repost, LinkedIn suggestion, or other organic.

Capture each post

Open every qualifying post and treat it as the canonical snapshot. Record:

- author, relationship degree, headline or company, stable post URL, age or date, and capture time
- full expanded copy with wording, punctuation, emojis, and line breaks preserved
- format, call to action, requested keyword, visual description, and material text shown in the visual
- reactions, comments, and reposts shown on the feed card and opened post, including any change
- the exact way the post surfaced in the feed

Use "not visible" for unavailable metrics. Do not infer impressions, saves, reach, clicks, conversions, unique commenters, hidden comments, slides, or video frames.

Save a clear screenshot of the expanded post and visible engagement. Save the displayed image when practical. Use one representative screenshot for a document, carousel, or video.

Inspect the conversation

Inspect up to 10 visible top-level comment threads using LinkedIn's displayed sort. Record the sample size and expand visible replies when practical.

Classify each top-level comment once as:

- substantive discussion
- resource request
- qualified resource request with a relevant use case
- generic reaction or tag
- spam
- unclear

Count replies separately as author fulfillment or acknowledgment, other author reply, reader-to-reader reply, or unclear.

Report the sample in plain counts. State that LinkedIn selected the visible sample. Do not describe the displayed comment total as a count of people discussing the post.

Analyze each post

Keep observations separate from inference.

1. Momentum
- Show reactions, comments, reposts, and age.
- Calculate the comment benchmark as opened-post comments divided by 400, multiplied by 100. A score of 100 means the post reached the threshold.
- Label momentum High, Middle, or Lower within the collected set. Use Low confidence when fewer than two engagement metrics are visible.

2. Participation mechanism
- Choose one primary label: discussion, resource access, community or application, humor or relatability, proof or demo, contrarian opinion, identity, curiosity, or mixed.
- Mark access that depends on a comment as incentivized.
- Explain whether the visible response comes from requests, qualified demand, discussion, author replies, distribution, or a mixture.

3. Reference grade
- Score five dimensions from 0 to 2: hook clarity, evidence and packaging, response quality, independent corroboration, and business transferability.
- Give one sentence of evidence for every score.
- A total of 8 to 10 earns Good reference.
- A total of 5 to 7 earns Okay reference.
- A total of 0 to 4 earns Weak reference.
- The grade measures usefulness for the account owner's business. It does not predict reach or describe LinkedIn's ranking system.

4. Confidence
- Use Higher when the opened post, at least two metrics, full copy or visual evidence, and 10 inspected threads are available.
- Use Medium when one key metric is missing or 5 to 9 threads were inspected.
- Use Low for feed-card evidence, fewer than 5 inspected threads, or major missing evidence.

Turn patterns into original ideas

For each post, extract the reusable mechanism. Create one original opportunity that fits the business inferred from the profile. Include:

- insight
- audience relevance
- thesis
- hook
- short outline
- proof needed
- call to action
- details to avoid copying

Do not reuse another author's wording, personal story, visual identity, customer result, or unsupported claim.

After collecting the set, identify repeated topics, hooks, formats, visuals, and participation triggers. Separate genuine discussion from gated demand and author reply volume. Identify saturated angles, find three open opportunities, and recommend the first three ideas to create.

Deliverables

Create:

- index.html with an executive summary, ranked table, evidence card for each post, screenshots, saved visuals, comment evidence, mechanism, grade, confidence, adaptations, synthesis, and limits
- index.json with the same records in structured form
- assets folder with screenshots and saved visuals using stable relative paths
- ZIP containing the full report

Use this table layout:

Post | Surfaced via | Age | Format | Mechanism | Reactions | Comments | Reposts | Comment sample | Momentum | Reference grade | Transferable pattern | Confidence

Quality check

Before finishing, verify that:

- 10 qualifying posts come from 10 different authors, or the report documents the shortfall
- every opened post clears 400 comments
- paid posts and duplicates are absent
- every URL exists
- HTML and JSON agree
- local visuals render
- the ZIP opens
- observed evidence, author claims, and analyst inference stay clearly labeled

Stop after 10 complete posts. If 150 organic posts are inspected first, deliver the strongest complete partial set and state the inspected and qualifying counts. Keep the 400-comment threshold.