> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.beatapi.io/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.beatapi.io/_mcp/server.

# How do I get the next search page?

> Paginate twitter.web.fetch_search_timeline by keeping keyword and search_type unchanged and passing data.next_cursor from the previous response as params.cursor; stop on absent, repeated, or empty-page cursors.

To fetch the next X search page, call `twitter.web.fetch_search_timeline` again with the same `keyword` and `search_type`, passing `data.next_cursor` from the previous response unchanged as `params.cursor`. Omit `cursor` on page one, and start a fresh page-one search whenever you change `keyword` or switch between `Top` and `Latest`.

For example, when page one queries a brand name with `search_type:"Latest"`, page two sends the exact same brand name, `search_type:"Latest"`, and page one's `data.next_cursor`. See [Search and CSV export](/guides/x-search-posts).

## When do I stop, and what can I report?

Read the stop state recorded in `coverage.json` and act accordingly:

| `coverage.json` state | Next step                                                                                            | What you can report                                                                                            |
| --------------------- | ---------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------- |
| `saved_batch_ended`   | The saved batch reached its last saved page. Send `next_params` only if budget permits another page. | The saved pages were processed; the sample is not exhaustive.                                                  |
| `cursor_absent`       | Stop and send no further requests.                                                                   | The response returned no continuation cursor. An absent cursor does not prove full historical coverage.        |
| `repeated_cursor`     | Stop and inspect the raw response files.                                                             | The cursor repeated, so continuing could loop.                                                                 |
| `empty_with_cursor`   | Stop and diagnose the empty page.                                                                    | The page succeeded with zero posts while still returning a cursor; do not purchase another page automatically. |

HTTP errors, business failures, wrong actions, missing arrays, and malformed JSON are failures rather than successful empty pages. Keep the attempt logs and raw response bytes, then follow the [empty-result diagnosis](/faq/x-empty-search-results). This workflow never retries automatically. If you receive approval to investigate after an `empty_with_cursor` stop, record that investigation as a new batch instead of appending the request to the stopped cursor chain.

## Does a two-page budget mean forty posts?

No. A two-page budget limits the number of requests rather than guaranteeing a fixed post count per page, unique post count, or date range. Page two can overlap with page one, so fewer rows after deduplication does not mean posts were lost.

At the current Inspect rate of `$0.003` per request, two successful pages cost an estimated `$0.006`, excluding retries or extra pages. Stop early on an empty page or an absent cursor instead of spending the remaining budget just to complete two pages.

## How do I continue saved JSON without paying for it again?

Export saved responses locally using `export_saved_x.py` from the search guide; fetching responses and exporting saved responses are separate workflows.

1. Record each page's actual parameters, query time, and original response filename in `pages.json`.
2. Run `export_saved_x.py`. `export_saved_x.py` verifies that every page shares the same `keyword` and `search_type`, that each cursor continues the previous page, that request IDs are distinct, and that collection timestamps are ordered, then writes `posts.csv`, `coverage.json`, `sources.json`, and byte-identical copies of the source files without network access.
3. Inspect `stop_reason` before fetching another page. Only `saved_batch_ended` with valid `next_params` prepares a continuation, and sending that continuation still requires budget approval.
4. Save the new response, append the page's actual metadata to `pages.json`, and export into a new output directory.

Never rerun page-one collection solely to add CSV columns or adjust local parsing.

## How should overlaps, timing and evidence appear in the report?

`posts.csv` retains the first observation of each original `tweet_id`. Later observations of the same post remain in the raw JSON files, and `coverage.json` records the duplicate count. Do not overwrite the first observation's like count with a later value.

State the actual earliest and latest `created_at` timestamps and every page's collection time in the deliverable. A two-page sample count is not total mention volume. File hashes identify the saved bytes without proving server authenticity, timestamp accuracy, or completeness.

Contract reference: `data:twitter.web.fetch_search_timeline`.