Trace a manual opportunity-task summary: sentence-to-field provenance

Shorter wording must not expand evidence; trace each sentence to its source

Teams supplying existing inquiry/follow-up tasks as PM Radar JSON may write manual summaries; this article adds no opportunity-summary engine. Establish source/scope, then map sentences to fields and retain unknowns. Waiting for acceptance cannot become accepted, and a link cannot turn a proposal into approval or an independently verified external fact.

Different from second checks and batch comparison

Second checks assess evidence for one anomaly claim; batch comparison establishes comparability of exports. Here compressing several fields into a sentence must preserve conditions, attribution and scope. Sentence numbers, source labels and citation tables are editorial aids, not app-generated reference IDs or an audit feature.

Sentence-to-field provenance table

Summary content Source field/evidence Supported attribution Do not rewrite as
Project/item identity project, existing id and reviewed source Sample item in this source Same title proves the same customer/order
Item title Original title/context Source says review is needed Wording proves an actual urgent event
Current state Explicit status if supplied Source labels waiting for acceptance Accepted, approved or complete
Proposed next step Conditions, negations and proposals in next Source suggests review first Execution authorized or sending scheduled
Priority priority and source explanation Supplied P0/P1 or unknown Specific loss/deadline or unknown means unimportant
Source time date_str plus read/cache state Within that source-read scope A current live fact or refresh tap means fresh
Evidence reference evidence and authorized original record A reference exists; content needs review HTTPS means true, non-private or approved
Unknowns/conclusion Absent fields and unmapped claims Pending review or remove unsupported sentence Invented details to complete a summary

A synthetic sentence map

Editorial source S-A: Sample item; status=waiting for acceptance; next=suggest review next time, not approved; priority absent.
Candidate sentence 1: The sample was accepted and execution authorized.
Mapping: Status does not support accepted; next explicitly lacks approval. Reject sentence 1.
Supported rewrite: Source marks the sample pending acceptance; next is an unapproved review suggestion, priority unknown.

S-A and sentence 1 are manual examples, not real task IDs, JSON fields or executed detector output. No real customer text/private evidence address or model call is involved. An authorized external summarizer still requires the same review; fluent prose is not evidence.

Attribution, scope and reference locations

Retain original wording/location before writing attributed text such as source labels rather than a global assertion. Identify whether material is full tasks, a decisions subset, current filters or a Watch summary. Subset evidence cannot establish a whole-ledger conclusion.

Review private originals within existing permissions rather than copying full tasks or parameterized links into public summaries. A supplied field establishes a source value, not authority, truth or write permission. Unmapped sentences remain unreviewed or unused; do not fabricate decorative references.

FAQ

Does Yihu automatically generate opportunity summaries/citations?

No engine is asserted or added. Existing source reads task JSON and summary carriers; sentence writing, numbering and mapping here are manual guidance.

Can waiting for acceptance be shortened to accepted?

No. Waiting and completion are opposite states. Preserve status/attribution without turning a subset label into approval.

Can a customer-like title establish the same customer record?

Not by title similarity. Verify source/stable item identity, and do not copy personal information into a public example.

Is an HTTPS citation per sentence enough?

No. Review which claim the reference supports, permission, version and scope. Link shape alone validates neither truth nor approval.

Can missing priority/time simply disappear from a summary?

Do not let omission imply verification. Mark unknown or omit conclusions that depend on absent evidence, without invented current time/low priority.

Can a reviewed summary automatically be sent to a customer?

No automated sending/write authority is supplied. Text review, handoff approval and customer communication are different operations under original permissions.

Reading and evidence

Second-check evidence · Batch comparison · Format compatibility · Follow-up minimization · Start · Privacy

Evidence: Radar fields/link parsing, README collection/reply limits.