---
profile: elgora_markdown_bounty_challenge_v1
escrow_amount: "20000000"
submission_deadline: 1795219140
payout_policy: winner_take_all
---

# Science Spread Tracer

## Summary

I want to know how new scientific papers actually spread once they are published: which channels pick them up first, where they cascade next, and how far they go. This bounty buys an open-source tracker plus the dataset it produces for the ten papers listed below. The eligible Submission with the most verified mentions wins.

## Challenge details

### The papers

These ten papers are the only subjects of this bounty. A mention counts only if it refers to one of them.

| # | Title | DOI | Journal, published |
| --- | --- | --- | --- |
| 1 | Opportunistically Detecting Signs of Hypertension on a Consumer Smartwatch | 10.1038/s41586-026-11336-3 | Nature, 5 Nov 2026 |
| 2 | In Vivo CRISPR Base Editing of ANGPTL3 for Durable Lipid Lowering | 10.1038/s41586-026-11233-9 | Nature, 6 Nov 2026 |
| 3 | An AI-native experimental laboratory for autonomous biomolecular operations | 10.1038/s41586-026-11228-6 | Nature, 4 Nov 2026 |
| 4 | Observation of prethermal dynamical phase transitions and resonant melting | 10.1038/s41586-026-11225-9 | Nature, 4 Nov 2026 |
| 5 | Hydrogen Derivatives Enable Cost-Effective Carbon Neutrality in China | 10.1038/s41586-026-11351-4 | Nature, 5 Nov 2026 |
| 6 | Ecological and social cues orchestrate concerted courtship in Drosophila | 10.1038/s41586-026-11209-9 | Nature, 4 Nov 2026 |
| 7 | In vivo CAR-T double knockout screen maps synergistic antitumour gene pairs | 10.1038/s41586-026-10956-z | Nature, 4 Nov 2026 |
| 8 | A two-dimensional reticular triaxial molecularly woven fabric | 10.1038/s41586-026-11218-8 | Nature, 4 Nov 2026 |
| 9 | Accelerating commercialization and deployment of indoor photovoltaics | 10.1038/s41586-026-11133-y | Nature, 4 Nov 2026 |
| 10 | The true scale of global wildlife trade is obscured by data gaps | 10.1126/sciadv.aei3504 | Science Advances, 30 Oct 2026 |

### The window

Only mentions whose `timestamp_utc` falls from 30 October 2026 00:00:00 UTC up to the submission deadline count.

A mention's `timestamp_utc` is its original publication or posting time, as shown on the page or in the page's HTML metadata (such as `article:published_time`). For a Wikipedia mention it is the revision time, and for a YouTube mention the upload time. Update and edit times never count, and if a page shows more than one publication time, the earliest one counts. The time is written in ISO 8601 UTC (`YYYY-MM-DDTHH:MM:SSZ`), as follows:

- A time with a time zone or UTC offset is converted to UTC.
- A time without a time zone or offset is read as UTC.
- A date without a time is written as 00:00:00 UTC on that date.
- A mention with no publication date on the page or in its metadata is not a verified mention.

### The channels

A mention is a public post, comment, article, page revision or video on one of these channels:

| Channel key | Covers |
| --- | --- |
| `x` | X / Twitter |
| `bluesky` | Bluesky |
| `mastodon` | Any Mastodon instance |
| `reddit` | Reddit posts and comments |
| `hackernews` | Hacker News stories and comments |
| `wikipedia` | Wikipedia article revisions, in any language |
| `youtube` | YouTube video titles and descriptions |
| `news_blogs` | Article pages, as defined below |

An article page counts for `news_blogs` when it shows a headline, a publication date, and an author byline or publication name, and its text is a written article rather than a list of links, search results or feed items. Forum threads, comment sections, and pages on the platforms of the other seven channels are not `news_blogs` mentions.

Pages of the papers' own publishers and journals, and of research indexes and metrics services (such as PubMed, Europe PMC, Google Scholar, Semantic Scholar, ResearchGate, Crossref and Altmetric), are not mentions on any channel.

### Referring to a paper

A mention refers to a paper when its text contains the paper's DOI, a link that resolves to the paper (on the journal's site, doi.org or PubMed), or the paper's full title from the table, compared without regard to letter case or text formatting.

### Verified mention

A row of `data/mentions.csv` is a verified mention when all of these hold:

- its `captured_text` appears on the page at its `url`, or on the archived copy at its `archive_url`;
- its `captured_text` refers to the paper given in `paper_number`;
- the mention is on the channel given in `channel`;
- its `timestamp_utc` is the value the timestamp rule in The window gives for it, and falls inside the window.

Rows that point to the same post, comment, article, revision or video count as one mention.

Any data source, API or method is allowed.

## Deliverables

The Submission contains:

- `source/`: the tracker's source code, with `source/README.md` giving the runtime version, installation steps, pinned dependency versions, and commands for collection and replay, and `source/LICENSE` containing the text of a license approved by the Open Source Initiative (for example MIT or Apache-2.0).
- `data/mentions.csv`: one row per mention, with these columns:
  - `paper_number` (1 to 10)
  - `channel` (a channel key from the table)
  - `url`
  - `archive_url` (a Wayback Machine or archive.today copy of the page, or empty)
  - `author_handle`
  - `timestamp_utc` (as defined in The window)
  - `captured_text` (the text of the mention that refers to the paper, at most 2,000 characters)
  - `engagement` (the like, upvote or view count the platform shows, or empty)
- `data/first_mentions.csv`: one row for each paper and channel with at least one row in `data/mentions.csv`, with these columns:
  - `paper_number` (1 to 10)
  - `channel` (a channel key from the table)
  - `first_timestamp_utc` (the earliest `timestamp_utc` among that paper's rows on that channel)
  - `url` (the `url` of a row with that earliest `timestamp_utc`)
- `evidence/`: the saved source responses or page extracts used to produce the submitted dataset, together with a manifest linking each evidence file to its source URL, retrieval time in UTC, and SHA-256 digest. Include the source content needed to derive each mention's paper, channel, publication time and captured text. Do not include credentials or private content.
- `source/config.example.json`: an editable example configuration for the paper DOI/title list, date window and enabled channels. Document any API credentials, permissions, paid services or rate limits required for fresh collection; do not embed secrets.
- Optionally, `report.md`: for each paper, the order in which channels picked it up, and any patterns across papers. It is not judged.

## Acceptance Criteria

A Submission is eligible when all of these hold:

1. `source/README.md`, `source/LICENSE` (with an OSI-approved license), `data/mentions.csv` and `data/first_mentions.csv` are present, and both CSV files have the columns listed in Deliverables.
2. At least 90% of the rows in `data/mentions.csv` are verified mentions.
3. Verified mentions in `data/mentions.csv` cover at least 4 different channels.
4. `data/first_mentions.csv` has exactly one row for each paper and channel pair that appears in `data/mentions.csv`, and in each row `first_timestamp_utc` is the earliest `timestamp_utc` among that pair's rows in `data/mentions.csv` and `url` is the `url` of a row with that timestamp.

5. The tracker has two documented modes: collection retrieves source records for configurable papers and a date window, saves their evidence, and produces both CSVs; replay produces both CSVs from the supplied evidence without using network access, credentials, the submitted CSVs, or a hard-coded copy of their rows. Collection source code must cover at least the four channels that satisfy criterion 3; additional channels are optional. Availability of a third-party service at judging time is not an acceptance condition.
6. With the documented runtime and dependencies, the replay command completes successfully and reproduces the submitted CSVs from the supplied evidence. Equality means the same columns and the same multiset of parsed field values, ignoring row order, CSV quoting, line endings, and a UTF-8 byte-order mark. The output location is configurable so the original CSVs are not inputs.
7. Replay uses the editable paper and channel configuration: removing one paper from that configuration removes all of its mentions and first-mention rows from the replay output, while leaving the other configured papers' results unchanged. Likewise, disabling one channel removes only that channel's rows. Source code must derive rows from evidence, not from a prewritten output table.

A dataset-only Submission or nonfunctional placeholder tracker is ineligible. Replay demonstrates reproducibility of the captured dataset; it does not guarantee that future platform access, APIs or search coverage remain unchanged.

## How is the winner selected?

Among eligible Submissions, the one with the most verified mentions in `data/mentions.csv` wins.

Ties break in this order:

1. More channels with at least one verified mention.
2. More of the ten papers with at least one verified mention.
3. The Submission received first.