On this page
THE SHORT ANSWER
Check one important sentence at a time. Identify whether it reports a calculation, a tradition or a personal inference, then inspect the evidence for that exact claim. Finding a title, reading an excerpt and validating a prediction are different achievements. A long reference list cannot substitute for that distinction.
- Match the source to the claim: implementation for calculations, texts for traditions and real information for personal conclusions.
- Keep track of what was actually read. A summary is not a full-work review.
- Separate supported, unsupported and unresolved statements before saving a report.
This result counts zero visible Water characters.
A school associates this symbol with a particular theme.
Therefore, you should change industries.
A page was found, so the conclusion is verified.
These cards select four types of example: a calculation, a traditional interpretation, a personal inference and an unread citation. The statements are claims to examine, not established conclusions. The complete table below adds precise percentages and explains the checks and limits for all five types.
Identify what kind of claim you are reading
“Three different cards were drawn” is a result you can check. “This card is often associated with waiting” describes an interpretive tradition. “You may be waiting for someone’s permission” is an inference about your situation. Appearing in one report does not give these claims the same status.
For a birth chart, check the inputs and conventions behind the stems, branches and palaces first. Then examine where statements about personality, relationships or future events come from. A reproducible calculation does not validate every interpretation attached to it.
A citation should support the particular claim
Open the source and check whether its relevant section addresses the statement. Software documentation can explain an algorithm. A traditional text can document a symbolic association. Neither establishes a prediction about your future simply by being cited.
Reading a search summary, reading an excerpt and reading a whole book are different things. Ask the Agent to match conclusions to their supporting passages and identify any source it could not open or verify.
What kind of claim are you checking?
| Type | Example | Check and limit |
|---|---|---|
| Calculation | This result counts zero visible Water characters. | Compare the chart and counting method. This does not measure hidden stems or strength. |
| Tradition | A school associates this symbol with a particular theme. | Check the work and passage. One school is not every tradition. |
| Personal inference | Therefore, you should change industries. | Examine your goals and circumstances. The preceding sources do not establish this advice. |
| Unread citation | A page was found, so the conclusion is verified. | Check that relevant text was actually retrieved. A title or failed read is insufficient. |
| Precise percentage | Your job change has an 87% chance of success. | Ask for the event definition, data and validation. Precision alone is not measurement. |
On a small screen, scroll the table sideways.
Know the scope of Wenbu research
Research mode searches Wenbu’s guides, card and hexagram notes, and a curated reference catalogue. It can attempt to read public excerpts from external pages in that catalogue. It does not browse arbitrary URLs or search the entire web.
Access restrictions, page formats and reading limits can leave gaps. Finding a title is not the same as reading the text. A useful report should say what remains unchecked. If your question is outside the catalogue, ask what can currently be established and what additional source would be needed.
Check one claim that matters to you
Suppose a report says solar-time correction changes all four BaZi pillars. Ask: “For Wenbu’s current implementation, explain separately how the year, month, day and hour are determined. Show the calculation basis and do not generalize to every school.” This gives the reply a specific rule to verify.
Before saving a report, choose one important conclusion and check its basis. If the support is missing, mark it as unresolved. A confident sentence is not a reason to close the question.
- “Is this a calculated result, a traditional interpretation or an inference from my context?”
- “Which source directly supports this point? Separate anything the source does not establish.”
- “How would the conclusion change if this detail about my situation were wrong?”
Worked example: “No Water means you must change industries”
Treat this fictional sentence as a claim to review, not advice about your own chart.
Split it into three claims: the count is zero; zero means Water is favorable; a favorable element dictates a career change. Each needs separate support.
Check the chart and method. Wenbu counts visible stems and branches without weighting hidden stems, season or strength. A correct count does not establish the second claim.
Ask for a source and a read passage for each claim. Label a Wenbu guide as editorial material. If an external page could not be read, retain that gap.
Keep only the supported statement: “Under these inputs and counting rules, the visible Water count is zero. This is not a favorable-element assessment or a career recommendation.”
Save the correction and why it was needed. A changed input or counting method requires a fresh check, rather than reuse of the old count.
Removing an unsupported inference is a useful research result, even when it produces no definitive career answer.
Key terms
- Direct support
- Evidence addressing the claim itself under matching conditions.
- Reading scope
- The text actually available: a summary, excerpt or fuller work.
- Unresolved
- Not established by the available evidence; neither automatically false nor true.
Common questions
Is a real link enough?
No. Its relevant passage must support the particular statement and its conditions, rather than merely discuss a related topic.
Do several agreeing sources validate a claim?
Check whether they repeat one another. Repetition can document a tradition’s spread without validating a personal prediction.
What does a Wenbu source label guarantee?
It helps identify material and reading status, not perfect output. Research remains limited to the collection. NIST also identifies confident false output as a generative-AI risk.
Sources and their scope
Directly supports the visible-character counting method and excluded factors. The career sentence is a fictional audit example.
Supports the internal collection, external allowlist and public-excerpt reading limits.
Supports product requirements for read sources, failure disclosure and labeled inference; requirements do not guarantee error-free output.
Section 2.2 describes confident false output; MS-2.5-003 calls for checking sources and citations. This is not an accuracy evaluation of Wenbu or divination.
Traditional interpretations are not established predictions. See the source notes for context.
Read this guide with your agent
Get the complete text, diagram descriptions and sources in an open format. These files contain public knowledge only.
ADDITIONAL PRACTICE · 2026-11-11
What to do when a pinned source returns source_superseded
When an agent says it read a source, check which part it read
Label each citation's scope before counting citations. According to the integration page, read_library reads guide or symbol entries returned by search and by default returns the full text with truncated:false; adding a section parameter reads only that section and marks scope as section. External source IDs are not supported by that tool, MCP has no read_reference tool, and the host must open the returned link itself; a search summary does not mean the text was read. So an answer's citations fall into at least three tiers: search summary only, one section read, or full text read. Write each sentence beside its citation's tier, and mark any sentence the tier cannot carry as scope unverified rather than treating a link as a read.
Wenbu · agent integrations ↗Wenbu · versioned public capabilities ↗
- Give each citation exactly one tier: search summary, section or full text; write unverified when unsure.
- Match tier to sentence: a section-level citation supports only what that section discusses, and a summary cannot carry a specific conclusion.
- External links are opened by the host, not by Wenbu tools; keep the gap when a page cannot be opened.
How to assign a tier to one citation
- 01Identify the entry type
Guide or symbol entries can go to read_library; external source IDs are not supported there and move to the host-read branch.
- 02Read the returned scope and truncated fields
Default full text is the full-text tier; a section request with scope marked as section is the section tier; search entry information alone is the summary tier.
- 03Compare back to the sentence
The sentence is supported only if its content falls inside the text that tier covers; generalizations, numbers or cross-topic conclusions are marked unverified.
Wenbu · agent integrations ↗Wenbu · versioned public capabilities ↗
Split one answer into sentence, citation and tier
Take a real agent answer you already have and copy each sentence that carries a citation onto its own row. The third column takes only three values: summary (you saw a search title or snippet), section (read_library was called with a section parameter and returned scope as section), or full text (read_library returned the default full text with truncated:false). Avoid vague entries such as probably read; the tier should point to an observable return state.
For an external source ID, write host-read or unread in the third column. The integration page says those IDs are not supported by read_library, MCP has no read_reference tool, and the host must open the returned link itself. If the answer shows only a title, treat it as a summary even when the link looks perfectly on topic.
Wenbu · agent integrations ↗Wenbu · versioned public capabilities ↗
Use the tier to test whether the sentence holds
A higher tier is not automatically better; the tier has to match the sentence. A section-level citation can support what that section explicitly discusses, but if the sentence turns one example from the section into a general rule, it has outgrown the scope. A summary-level citation can only establish that a related entry was found, so sentences with numbers, conditions or conclusions should be marked scope unverified.
One case is easy to miss: a calculation sentence and an interpretation sentence sharing a single citation. Calculation conventions let you check inputs and rules, but they do not vouch for a traditional reading or a personal inference placed after them. Tagging the two sentence types separately is more useful than writing cited throughout.
Wenbu · agent integrations ↗Wenbu · calculation conventions ↗
Write the result as a re-checkable list
Keep at least four fields: original sentence, citation ID, tier and verdict (supported, beyond scope or unverified). For a beyond-scope sentence, ask the agent for the specific section or rewrite the sentence to match what was actually read. For an unverified sentence, do not delete it yet; keep it to show where the gap sits.
On a second pass, read only the list, not the tone of the original answer. Confidence does not change a tier; the tier comes from the return state and the text actually opened. Reuse the same three-column list on another answer to compare the two checks.
Wenbu · agent integrations ↗Wenbu · versioned public capabilities ↗
| Tier | Return state it rests on | Sentences it can support | Sentences it cannot support |
|---|---|---|---|
| Summary | Only search_library entry information; body text not read | Locating statements such as a related entry exists | Specific section content, numbers, conditions and conclusions |
| Section | read_library called with a section parameter; scope marked as section | What that section explicitly states, with its limits | Topics the section does not discuss, or an example turned into a rule |
| Full text | read_library default full text with truncated:false | Statements locatable within that document | Inferences beyond the document's topic, and personal future predictions |
| Host-read | External source ID opened by the host | Passages the host actually opened and read | Cases where only a title or link was seen and the body was not opened |
Wenbu · agent integrations ↗Wenbu · versioned public capabilities ↗
Hypothetical example: tagging three citations
The following is a fictional agent answer excerpt used to demonstrate the labelling method. It is not a real user record and says nothing about Wenbu's view of any person.
- Sentence one: the current implementation counts only visible stems and branches in the element chart. Its citation points to a guide entry, and read_library returned the default full text with truncated:false. Tier: full text. The sentence matches that document's counting convention, so the verdict is supported.
- Sentence two: therefore your day master is weak and you should add Water. The citation is the same guide, but read_library was called with a section parameter and scope was marked as section. Tier: section. The sentence makes a strength judgement the section does not cover, so the verdict is beyond scope; rewrite it as the section does not discuss strength.
- Sentence three: separate external research supports this conclusion. The citation is an external source ID. The integration page says such IDs are not supported by read_library, MCP has no read_reference tool, and the host must open the link. Tier: host-read. If the link was never opened, the verdict is unverified, and the gap stays open rather than being filled in.
- Write the three rows side by side: sentence, citation ID, tier, verdict. Keep only the first row as written; rewrite the second; mark the third unverified and note which link must be opened to continue.
The observable result is that only one of three citations had a tier that carried its sentence. The list proves no divination conclusion, but it shows plainly which sentence lacks support and which passage to read next.
Wenbu · agent integrations ↗Wenbu · versioned public capabilities ↗Wenbu · calculation conventions ↗
Questions you may have
Does truncated:false mean the whole book was read?
No. The integration page describes the knowledge guides returning full text by default, and truncated:false describes that this return was not cut off, not a claim about reading any external work. External sources do not go through this tool at all.
Can a search summary count as having read the text?
No. The integration page states that a search summary does not mean the body was read, and the capabilities contract calls search entries metadata rather than fetched source text. A summary supports only locating statements.
Wenbu · agent integrations ↗Wenbu · versioned public capabilities ↗
How should I record a link the host opened itself?
Record host-read and note which passage was actually opened. If you saw only a title or the page failed to load, record unread or unverified instead of upgrading it to full text because the link exists.
Terms in this guide
- Tier
- This page's label for a citation's reading state: summary, section or full text, with host-read or unread for external sources.
- Scope unverified
- The sentence goes beyond the text its citation tier covers, or the body was never opened; it signals insufficient support, not that the sentence is false.
See the claim and source notes
read_library returns the full text with truncated:false by default, and a section request returns only that section with scope marked as section.
scope 会标为 section
External source IDs cannot be passed to read_library; MCP has no read_reference tool, the host opens links, and a search summary is not a read.
MCP 没有 read_reference 工具
The capabilities contract asks readers to read the full document or relevant section before citing and calls search entries metadata rather than fetched text.
reference search entries are metadata, not fetched source text
Sources and review scope
Wenbu drafted and checked this article in Codex. The source list records the scope of the review.
- Current complete zh canonical article ↗Current public original and earlier additions; add substantial coverage without repeating or replacing them. · 2026-10-29 · Fetched text
- Current complete en canonical article ↗Current public original and earlier additions; add substantial coverage without repeating or replacing them. · 2026-10-29 · Fetched text
- Wenbu · agent integrations ↗Current first-party product documentation · 2026-10-29 · Fetched text
- Wenbu · calculation conventions ↗First-party implementation; not evidence of predictive accuracy · 2026-10-29 · Fetched text
- Wenbu · versioned public capabilities ↗First-party current tool, convention and privacy contract; not external endorsement or predictive validation · 2026-10-29 · Fetched text
Fill the three-column list on one real answer
Pick the cited sentence you care about most, write down its citation ID and tier, and judge whether the tier carries it; this is a concrete check the current integration supports.
Fill the three-column list on one real answer ↗Treat source_superseded as a failed read of that pinned section. The current and nextAction fields tell you where to look; they are not source text. Read the current section successfully before replacing the citation. Check the original sentence against the new passage and narrow or remove it if the passage no longer supports its full scope.
Current complete zh canonical article ↗Current complete en canonical article ↗Wenbu · agent protocol ↗Wenbu · agent integrations ↗Wenbu · reviewing a reading ↗Wenbu · versioned public capabilities ↗
- Keep the failed version in your audit notes. Do not file the new text under the old section ID.
- Follow nextAction and obtain the current passage before using it as evidence.
- Check what the new passage supports and change the sentence where needed.
From a withdrawn reference to fresh evidence
- 01Flag the old citation
Keep the revision and section ID in the change notes.
- 02Read the current passage
Follow nextAction and check the text and revision.
- 03Revise the claim
Match the scope of the sentence to the evidence you actually read.
Current complete zh canonical article ↗Current complete en canonical article ↗Wenbu · agent protocol ↗Wenbu · agent integrations ↗Wenbu · reviewing a reading ↗Wenbu · versioned public capabilities ↗
Read the error before changing the citation
Suppose a report cited an older section of a chart guide and a correction later withdrew that section. The pinned read returns isError:true with source_superseded, current and nextAction. This response does not count as reading that section. Keep the old section ID and revision in your notes, and flag the sentence that relied on it for rechecking.
Follow nextAction to read the current source. Open the actual passage and note the link, section and revision that support the relevant sentence. If you have only the heading or an abstract, continue to the passage itself. Do not paste current wording under the withdrawn section ID or describe the presence of current metadata as a successful read.
Current complete zh canonical article ↗Current complete en canonical article ↗Wenbu · agent protocol ↗Wenbu · agent integrations ↗Wenbu · reviewing a reading ↗Wenbu · versioned public capabilities ↗
Revise the sentence as well as the link
Try a fictional example. The old report says “this value stays the same in every time zone.” The current passage only reports the same value for two particular inputs. Change the sentence to “these two example inputs produced the same value,” retain the input conditions and cite the new passage you read. Changing the citation alone would leave the unsupported universal claim in place.
Ordinary additions preserve revision pins that are still public. Use this procedure when the service reports a withdrawal. If the new read fails or the passage does not support the statement, leave that gap visible. A title or search snippet does not replace the passage.
Current complete zh canonical article ↗Current complete en canonical article ↗Wenbu · agent protocol ↗Wenbu · agent integrations ↗Wenbu · reviewing a reading ↗Wenbu · versioned public capabilities ↗
| Response | What it establishes | Action |
|---|---|---|
| source_superseded | The pinned section was withdrawn; no passage was read | Record the old reference and request the current section |
| A successful current read | You have a passage to examine | Check the claim and keep the new section and revision |
| Failed read or no support | No adequate replacement evidence yet | Mark the gap, narrow the claim or remove it |
Current complete zh canonical article ↗Current complete en canonical article ↗Wenbu · agent protocol ↗Wenbu · agent integrations ↗Wenbu · reviewing a reading ↗Wenbu · versioned public capabilities ↗
Recheck a claim that was too broad
Use this fictional receipt. The old report says a value stays the same in every time zone.
- Flag the old citation after seeing source_superseded.
- Assume the new passage reports matching values for just two example inputs. Save the new section ID.
- Revise the sentence to refer to those two inputs and include their conditions.
Keep the old citation in the change notes. The report should use only the new, narrower claim and the passage you read.
Current complete zh canonical article ↗Current complete en canonical article ↗Wenbu · agent protocol ↗Wenbu · agent integrations ↗Wenbu · reviewing a reading ↗Wenbu · versioned public capabilities ↗
Questions you may have
Can I cite the current metadata while I wait for the passage?
It is useful as a pointer for another read. It cannot support the substance of the sentence. Read the passage and check the claim before citing it.
Current complete zh canonical article ↗Current complete en canonical article ↗Wenbu · agent protocol ↗Wenbu · agent integrations ↗Wenbu · reviewing a reading ↗Wenbu · versioned public capabilities ↗
Does a new section invalidate every older citation?
No. An ordinary addition preserves still-public revision pins. Recheck the claims tied to a particular section when the service says that section has been withdrawn.
Current complete zh canonical article ↗Current complete en canonical article ↗Wenbu · agent protocol ↗Wenbu · agent integrations ↗Wenbu · reviewing a reading ↗Wenbu · versioned public capabilities ↗
Terms in this guide
- Pinned read
- A request tied to a source revision and section so that the cited passage can be checked.
Current complete zh canonical article ↗Current complete en canonical article ↗Wenbu · agent protocol ↗Wenbu · agent integrations ↗Wenbu · reviewing a reading ↗Wenbu · versioned public capabilities ↗
- Replacement evidence
- A successfully read current passage that supports the revised sentence.
Current complete zh canonical article ↗Current complete en canonical article ↗Wenbu · agent protocol ↗Wenbu · agent integrations ↗Wenbu · reviewing a reading ↗Wenbu · versioned public capabilities ↗
See the claim and source notes
A withdrawn pinned section returns a specific error.
error.code:"source_superseded"
The error returns no withdrawn text or successful read receipt.
No withdrawn text or successful source receipt is returned
Only a later successful read can supply replacement citation evidence.
Only a subsequent successful read can supply new citation evidence
Sources and review scope
Wenbu drafted and checked this article in Codex. The source list records the scope of the review.
- Current complete zh canonical article ↗Current public original and earlier additions; add substantial coverage without repeating or replacing them. · 2026-11-11 · Fetched text
- Current complete en canonical article ↗Current public original and earlier additions; add substantial coverage without repeating or replacing them. · 2026-11-11 · Fetched text
- Wenbu · agent protocol ↗Current first-party tool, citation recovery and completion protocol · 2026-11-11 · Fetched text
- Wenbu · agent integrations ↗Current first-party product documentation · 2026-11-11 · Fetched text
- Wenbu · reviewing a reading ↗Existing editorial practice · 2026-11-11 · Read excerpt
- Wenbu · versioned public capabilities ↗First-party current tool, convention and privacy contract; not external endorsement or predictive validation · 2026-11-11 · Fetched text
Practise checking a citation in Wenbu Agent
Use the fictional receipt to describe the next step. When checking your own report, read the current passage before citing it.
Practise checking a citation in Wenbu Agent ↗