BitShovel.One real problem at a time
Open the current workflowCurrent Workflow中文

One real problem at a time

中文

HISTORICAL DIG · JUL—AUG 2026

How do I keep a source behind every claim?

Research products can let people choose sources, review a plan, and receive citations, but more citations do not automatically make scope, dates, original claims, or uncertainty clear.

First recorded · Last public revision · Revision 2
Question scopeSource tableClaim checkingStop conditions

HOW IT RELATES TO WORKFLOW 001

Relation to the current subject

It supplies the source layer beneath a Dig—not another research-tool recommendation.

Workflow 001 separates personal history, maker descriptions, current reproduction, and unknowns. The recoverable-automation record preserves inputs, runs, and retry boundaries. AI-output verification returns to the actual material. This historical record states the step all three depend on: scope the question and sources first, then make every critical claim traceable to the original.

Record state at the time

  • Information research
  • Recoverable automation
  • Standard record
  • Developing
  • Historical E2 · three maker sources · no BitShovel same-question run

THE GUIDANCE AT THE TIME

Judgment at the time

A summary is an entry point to the source table—not a replacement for it.

Open full rationale and return rules
Define the question, date range, and allowed sources first. Require every critical claim to carry a direct link, date, and a plain statement of what the source supports. The summary is only an entry point to the source table; consequential claims still need the original material checked.
Who it was for
People who repeatedly research markets, products, or editorial topics and cannot deliver an unsourced summary
Why it mattered then
Open note

The smoother an automated summary reads, the easier it is to hide search boundaries, source freshness, and evidence gaps inside apparently complete prose that is hard to update or correct later.

Where the judgment stops
Open note

This historical record checks public maker descriptions, not research accuracy. A citation's presence does not mean it supports a claim, and search, connector, or export capability does not establish a correct conclusion for a topic. BitShovel has not run a standardized same-question comparison or connected private drives, email, or accounts.

CHECK BEFORE READING ON

Fit check

Use it only when the question is narrow, the sources can stay public, and a person can still spot-check the result.

This record may help

  • The question is narrow enough to limit by date and source type, and you already know at least part of the answer.
  • You need a source table whose items can still be opened, updated, and corrected later.

Do not use it yet when

  • The first trial already requires email, a private drive, customer material, or another sensitive source.
  • You need these three maker sources to establish factual accuracy, independence, or current general performance.
  • You plan to connect one result directly to publishing, payment, or another irreversible action.

HISTORICAL TRIAL · UNRUN AND UNFROZEN

Historical trial

Run one capped public-source trial and inspect only the three claims most likely to change the decision.

Time needed
20–30 minutes (earlier trial cap)
Likely cost
One capped trial only; do not schedule a recurring job first
Permission boundary
Public sources only; do not connect private drives, email, or accounts
  1. 01

    First step

    Open this step

    Choose one narrow question whose answer you know, limit it to the last 30 days and three source types, and require claims, direct links, dates, support scope, and conflicts.

  2. 02

    What should exist

    Open this step

    A source table that opens item by item, plus a short summary that states unknowns.

  3. 03

    How to tell it worked

    Open this step

    Spot-check the three claims most likely to affect the decision; the original text, publication date, and context support the wording.

  4. 04

    Stop when

    Open this step

    Stop when links fail, dates fall outside scope, citations do not support claims, conflicting sources are left unmarked, or the task requests private-data access.

  5. 05

    How to back out

    Open this step

    Discard the automated summary, keep the question and checked sources, narrow the scope, and then decide whether to rerun.

METHOD INTERFACES · NOT BITSHOVEL RESEARCH RESULTS

Method and interfaces

Narrow scope first, then keep sources, notes, and answers traceable.

Open method boundary

The two pinned open-source project images show “clarify before research” and “sources, notes, and answers in one workspace.” They explain a work structure; they do not establish correct citations, working links, or a BitShovel run of either project.

Pinned Open Deep Research workflow diagramMAKER DOCUMENTATION · NOT A BITSHOVEL RUN
LangChain Open Deep Research workflow diagram showing a request moving through clarification and brief generation

This Scope stage shows one ordering: clarify the request, then form a research brief. It does not establish a checked date range, source scope, citation support, or accuracy.

About this image and its source

Open Deep Research repository diagram, © 2025 LangChain, MIT. BitShovel only cropped the Scope stage and converted it to WebP without redrawing text, arrows, or relationships. It is not a BitShovel run. LangChain does not participate in or endorse BitShovel.

Open source page ↗Read use and attribution guidelines ↗
Pinned Open Notebook demo interfaceMAKER DEMO · NOT A RESEARCH RESULT
Open Notebook project demo with sources on the left, generated notes in the middle, and source identifiers in the chat answer on the right

Sources, generated notes, and answers with source identifiers can share one workspace. This does not establish that a citation supports a claim, a link opens, a conclusion is correct, or BitShovel ran Open Notebook.

About this image and its source

Open Notebook repository demo, © 2024 Luis Novo, MIT. BitShovel only removed metadata and converted it to WebP without cropping, translating, or redrawing the interface. The project author does not participate in or endorse BitShovel.

Open source page ↗Read use and attribution guidelines ↗

SOURCES AND EVIDENCE

Sources and evidence

As of August 15, 2026, all three records are maker sources. They describe features and launch-era risks—not independent accuracy or user outcomes.

Open evidence boundary

The historical E2 label came from reading two maker descriptions alongside an OpenAI launch-era risk document. It supports the need for a checking method—not an accuracy rate, saved time, or a claim that automated research is reliable.

  1. 01
    Current maker guidance · Rechecked

    OpenAI deep research: define the outcome, sources, and plan before reviewing citations and activity

    Open source boundary

    The current help page describes public web, uploaded files, or connected apps, plus plan review, citations, a sources section, and activity history. This earlier trial remains limited to public sources with no private material connected.

    Open original source ↗
  2. 02
    Launch-era risk document · Rechecked

    OpenAI deep research system card: launch-era prompt-injection, privacy, and hallucination boundaries

    Open source boundary

    This is a 2025 launch-era risk document for the early model at the time. It defines checks to consider, not a current error rate for the product or all research tools.

    Open original source ↗
  3. 03
    Maker documentation · Rechecked

    Perplexity Sonar Deep Research documentation

    Open source boundary

    The page now appears under Legacy API navigation. It is another maker's description of multi-source search and citation output—not proof of accuracy.

    Open original source ↗

REVISION RECORD

Revision record

Revision 2 separates current source status, product capability, and factual accuracy again.

Open 2 revisions
  1. Revision 1

    Changed the research output to a source table readers can reopen, rather than only automated prose.

  2. Revision 2

    Rechecked current maker guidance: kept connectors and private material out of the first trial, returned the launch-era system card and Legacy API page to their time contexts, and removed a third product image without an open reuse license.

    What the record said before
    Three sources together showed that research products support source selection, plan review, and citations.
    What changed
    The three maker sources only help define steps and risks to check. Whether a source supports a claim still requires opening the original, and the first trial uses public sources only.

WHAT THIS RECORD LEAVES

What this record leaves

Keep every important claim connected to its source—and keep unknowns visible.

The current Workflow continues separating personal history, project records, and maker material. This historical method only protects source, date, support scope, and stop conditions.

Back to the Dig index