Which WeChat article and account page fields can I retrieve?
Each WeChat page field comes from one specific action. For an article review sheet, take titles and text from wechat_mp.v2.fetch_article_detail_h5, reads and engagement from the standard wechat_mp.v2.fetch_article_stats, placement within a publication batch from the account list wechat_mp.v2.fetch_account_articles, and account fields from fetch_account_profile. Record each column’s source action and collection time, and leave missing values blank.
For example, a campaign review reads what each article covers, compares engagement over a comparable period and checks which article led its publication batch. Account information is a separate request with the verified account ID.
Choose required columns before requesting everything
If you only need title and body, skip the profile, comment and metric queries. Add requests only for newly required columns, and reuse existing records where they fit. Account material can be reused for the verified account only, never copied to another account with the same name.
Make every row traceable to its evidence
Keep the requested URL, the returned article link, the account identity, and separate request IDs and collection times for the body and the statistics. Matching titles do not make two records a pair, and short and long aliases are not assumed equivalent. When an association is unresolved, stop; do not place numbers next to the most similar title.
join_article_review.py (the last section of this page) rechecks raw responses, hashes and fields instead of trusting an editable metrics CSV. join_article_review.py combines validated saved body records with standard statistics only: headline position stays unjoined, and join_article_review.py collects no comments or profile information. A hash shows that archived bytes are unchanged; a hash does not show that an article is true or independently authenticate the article’s network origin.
Deliver a review table with states, not guessed completeness
Each requested column needs an actual source, an observation time and a state, or a reason the column remains unavailable or unsupported. Keep candidate judgments, excerpts, interpretations and follow-up steps as human review material, separate from API facts. Keep an abnormal visibility state next to an otherwise normal-looking number. Do not fill position 1 without headline evidence.
join_article_review.py accepts 1–5 articles per batch. Twenty articles need four explicit pair manifests and four new output directories, plus a separate cross-batch duplicate check; join_article_review.py does not retrieve every field for twenty articles automatically. When a pair fails, keep the batch inputs and investigate; do not drop the article to make the delivery look complete.
The field map and the offline body/statistics join follow. Sources: the published output contracts of each action listed. Account-private data and complete-page coverage are outside the fields listed.
Article text, identity and publication metadata
Use data:wechat_mp.v2.fetch_article_detail_h5. The optional published fields below sit inside data.content in both raw modes; raw=true adds metadata and does not move the body.
The current fixed output schema does not declare _content_type. Do not require _content_type, and do not infer a classic article from a missing _content_type. Local export and joining accept a missing _content_type or an explicit article marker, and stop for review on any other explicit value; this is a conservative local rule, not a guaranteed taxonomy returned by the action. Image pages and extra metadata do not mean the whole visible page was read. The article reading workflow records the returned-text scope and checks exact quotes against the saved source text.
Headline or secondary article comes from the account list
For wechat_mp.v2.fetch_account_articles with raw:false and article item_show_type:0, read data.articles[].idx: 1 is the headline, and larger values are secondary positions within the batch. Keep account ID, app_msg_id and idx together, and use the position inside the publication batch, not the row on the page.
Filter headline and secondary articles provides an offline exporter, unknown-position handling and raw-source evidence. For identity derived from links, use the link audit; a short link alone carries no idx.
Engagement requires the correct statistics action
Standard raw:true includes show_* visibility flags. Keep a hidden metric, an absent value, null, a display threshold and an actual zero distinct. An H5 fallback cannot recover reads. Export a metric review sheet keeps the action, URL, collection time, original value and state for each metric.
Comments, replies and account homepage
Comments and replies are separate requests. First-level fetch_article_comments pages with buffer; fetch_comment_replies targets one content_id and pages with an integer offset. To get more replies, continue with the offset of fetch_comment_replies. A comment count is not a collected comment, one page is not the whole discussion, and hidden or unpublished comments are not promised. ID and precision rules explain why an unsafe unquoted numeric ID must stop a reply workflow that depends on the ID; converting the ID to a string afterward does not restore the digits.
For fetch_account_profile with raw:false, the documented fields under data include user_name, nick_name, ip_wording, original_article_count, original_content_str, finder_username, finder_nickname and ban_type. Read the fields as account identity, displayed region, original-content information, linked Channels identity and raw status code. The fields are not a legal owner, a registered address, exact followers or complete history. signature and head_url are explicitly unavailable on the fetch_account_profile route; an empty value is not a profile choice. See account fields and ownership boundaries.
The account homepage list uses item_show_type: 0 articles, 5 videos, 7 audio, 8 image-text posts. Keep a separate cursor for each account and tab; the homepage list is not a Channels query. See content tabs and pagination for the saved-page exporter and the actual next_offset rules.
The published WeChat catalog has no dedicated corporate-owner action: check the registered entity against public registration records, and check followers and donation revenue in a backend you have access to. Do not fill registration numbers from nicknames, and do not infer followers from reads. Donation amounts, private audience analytics and unselected comments are not part of the field map.
Deliver a five-article review without buying every endpoint
- Start with five exact public article URLs and save each requested URL; do not silently merge short and long aliases. Choose the required columns before calling anything.
- Plan five H5 detail requests for titles and text. Add five standard statistics requests if you also need engagement counts. Check the rate of each action on BeatAPI’s pricing page.
- If headline position is required, reuse a saved article list that contains the article identity, or budget the list pages separately. The ten detail and statistics requests do not return the five positions. Profile, comment pages and selected reply pages add separate requests, only when needed.
- Use the linked text and statistics exporters to keep raw responses and hashes. Join by your saved request-to-article mapping, not by title. Keep each request ID with the request’s own collection time; sequential calls are not an atomic snapshot.
- Join the detail and standard-statistics exports with
join_article_review.pybelow, using the saved request mapping. Deliver one row per article with the requested URL, title, account ID, publication time, text and metric source files, metric observation time, and a state next to every requested metric. Add headline position only with list evidence. A missing field stays missing; annotate unavailable, hidden and threshold values instead of filling zero. Do not copy the full body into a summary cell and call the copy analysis.
For a synthetic example: a source with title=“Example”, read_num:null, show_read:false and no headline evidence yields the title Example, reads null with hidden visibility, and position unknown. The row does not show zero reads, first position or a failed article.
Check two rows before sharing
Pick one complete row and one row with a missing value. Open both articles and compare the sheet with the saved details, statistics and list records. Ask a teammate to say when each count was collected and whether each blank means missing, hidden or failed.
Build the review table from saved detail and standard-statistics exports
First run the article source exporter and the metrics exporter. The join reuses the two exporters’ output and makes no new query. Create pairs.json with, for each article, an explicit source record, a metrics directory and a one-based index into the evidence.json requests:
Keep the original wechat-article.json that each source record references. Pair paths are relative to pairs.json; source_file inside a record is relative to the record’s directory unless the path is already absolute. Select up to five pairs explicitly, not by title or by guessed short/long URL equivalence. The recorded request URL must match byte for byte between detail and statistics; do not substitute another returned link as the join key.
Save the code below as join_article_review.py and run python3 join_article_review.py pairs.json article-review:
The new article-review.csv has one row per article, separate detail and statistics collection times and request IDs, and value/state/raw/visibility_raw columns for each metric. review-evidence.json and the archived originals keep file hashes. CSV safety prefixes do not change the archived text. join_article_review.py reclassifies the original statistics instead of trusting editable metrics.csv values, and checks the source record’s title, account, text and publication time against the original detail.
position_state is always not_joined, because join_article_review.py does not join headline-position evidence. join_article_review.py does not collect comments, replies or account profiles. Only a validated batch produces the table: fix invalid source pairs instead of removing validation or silently dropping failed articles. Treat output interrupted by a filesystem write error as incomplete; the original inputs stay unchanged, and a partial table does not pass acceptance.
Hashes show matching archived bytes; hashes do not establish factual truth or the network authenticity of editable request metadata. Original bodies, metrics and evidence may contain private working information: keep the files local by default, and review and redact a separate copy for sharing. Local generation makes no new BeatAPI call.

