Verify Claim & Cited Report

What It Does

Every source in a Research Cafe has a Verify claim button in the reading pane's toolbar. Clicking it runs a background AI research loop that cross-references evidence and writes a verdict back to the source — and automatically produces a cited research report in the Reports group of your Cafe's sources rail.

• Verify claim — the agent extracts the source's main checkable claims, you pick which to verify and set a token ceiling, and the agent finds corroborating and opposing evidence from authoritative sources. A per-claim verdict badge appears on the source's rail row (as a status dot) and in the reading pane. When the loop finishes, a cited report appears in the Reports group with one section per claim, a verdict, inline [N] citations, and a References list containing only the cited sources.

This feature is available to all tiers (Free, Pro, Creator) and costs tokens.

You can also start a verify loop by highlighting a passage (or right-clicking a source) and choosing Verify claim from the Ask-AI popover — see Research Mode. A text-selection Verify checks the enclosing source's main claim, not the isolated sentence.

---

Verify Claim

How It Works

1. Open a Research Cafe and click the source you want to verify — it opens in the reading pane.
2. Click Verify claim in the pane's toolbar. The dialog fetches the source's main checkable claims in a short preview pass — this is cheap and happens automatically when the dialog opens.
3. The dialog shows a selectable list of claims — the most checkable one is pre-checked. You can:
   • Leave the pre-checked claim selected
   • Select more claims ("verify all")
   • Type your own specific claim in the text field
4. Set a token-limit ceiling — the pre-filled default is sensible; raise it if you want a deeper search. The estimate updates as you change it.
5. Click Verify. The dialog closes and you return to the reading pane — the source shows a "Verifying…" state (also visible as an in-flight spinner on its rail row).
6. When the loop finishes: the verdict badge resolves to one of four states on both the rail row and the pane, and a cited report is automatically created in the Reports group.

The Verdicts

What "Couldn't Verify" Means

"Couldn't verify" is an honest answer, not a failure. There are several legitimate reasons the loop may reach this verdict:

• The claim is an opinion or subjective statement — these are not falsifiable and cannot be verified by evidence.
• The claim is very niche or recent — peer-reviewed literature may not exist yet.
• The agent searched within the token budget and found no strong evidence either way — the claim is underdocumented.

A "Couldn't verify" verdict shows everything the agent checked — the underlying sources it found, what it searched, and a summary of why verification was inconclusive — in the reading pane's evidence panel. This gives you all the material to check the claim yourself.

"Budget reached" is not a verdict. If a claim couldn't be checked because the run hit your token budget before reaching it, the claim shows an amber "Budget reached" marker (not "Couldn't verify"). That means "we ran out of budget," not "we found nothing." Raise the token budget (or verify fewer claims at once) and re-verify.

Important: verification is best-effort

Verification is AI-assisted and best-effort. The agent cross-references your claim against other, independent, higher-tier sources and weighs the evidence — but it can be wrong, miss context, or be misled by weak sources. A green "Verified" badge is not a guarantee of truth. Always open the evidence panel and judge the sources yourself. (This is also covered in our Terms; the in-app evidence panel repeats the caveat at the point of use.)

"Verified" vs "Credibility" — two different things

These two badges measure different things and can legitimately disagree:

• Credibility (High / Medium / Low) rates the source itself — its provenance (academic, government, news, blog, social).
• Verified rates a specific claim — whether it's corroborated by other, independent sources.

So a Low-credibility source can carry a Verified claim (the claim checks out elsewhere), and a High-credibility source can carry a Contradicted one. Verifying a claim does not raise the source's credibility — it tells you the claim is backed up elsewhere. When a source is verified, the Verified badge is the signal to trust for that claim; the credibility badge still describes the source.

Viewing the Evidence Panel

Click View evidence in the reading pane on a resolved source to expand the verdict panel. For each claim you verified, the panel shows:

• The underlying source — the origin the claim traces back to (if found), as a clickable link
• Supporting evidence — authoritative sources that corroborate the claim, with the relevant passage quoted
• Opposing evidence — credible sources that contradict the claim, with context
• A summary of the verdict

Every piece of evidence is a real source saved to your Cafe's Sources rail, so you can read, enrich, and cite them. Click any evidence item to open it in the reading pane.

Fair Per-Claim Budget

When you verify multiple claims at once, the token budget is allocated fairly across all selected claims — each claim receives a guaranteed minimum slice so no single claim can starve the rest. The last claim may use slightly more if budget remains. If the budget runs out before every claim gets a full pass, the remaining claims resolve to "Couldn't verify — budget reached."

What It Costs

The verify loop is a mini research run: it uses your token balance to run web and academic searches. The cost estimate is shown in the confirm dialog before you commit — no tokens are charged until you click Verify.

Relative cost is similar to a Quick Auto-Research run on the same claim. Verifying a single focused claim is cheap; verifying many claims with a large token ceiling will cost proportionally more. The cited report is generated as part of the same run at no extra charge.

---

The Verify-Produced Cited Report

When a verify loop completes, CoffeeScribe automatically generates a cited report from the verified evidence and saves it in the Reports group of your Cafe's sources rail (the same location as Auto-Write reports). You do not need to click anything extra — the report appears as soon as verification is complete.

What Is in the Report

The verify report has one section per verified claim:

• A ## {Claim} heading
• A verdict line (Verified / Contradicted / Mixed / Couldn't verify)
• 1–2 paragraphs answering the claim with inline [N] citations drawn from the supporting and opposing evidence the agent found
• A References list at the end containing only the sources actually cited in the body — not every source the agent touched, just the ones that made it into the answer

Cited-Only References

The References list in a verify report only includes sources that a [N] marker in the body actually points to. Sources the agent checked but did not end up citing (gathered-but-unreferenced) remain in your Cafe's Sources rail but are excluded from the report's bibliography. This keeps the reference list clean and directly accountable.

Opening and Navigating the Report

• Click View report → on the source (in the reading pane) to open the report in the full-screen viewer, or find it directly in the sources rail's Reports group.
• In the full-screen viewer, click any [N] marker in the body to open the corresponding source in the reading pane (in-Cafe view). On shared/standalone report pages, [N] markers scroll down to the bibliography entry instead.
• Click any bibliography entry to open the original source URL.

Converting the Citation Style

The report is generated in the Cafe's current citation style. To change it:

1. Open the full-screen report.
2. Use the Citation style dropdown to pick a style: APA 7, APA 6, MLA 9, Chicago Notes-Bibliography, Chicago Author-Date, Harvard, or IEEE.
3. Click Apply. The bibliography re-renders in the new style and the in-text markers convert for author-date styles (APA, Harvard, MLA, Chicago-AD → (Author, Year) format; numbered styles like IEEE keep [N]).
4. The new style is saved to the report — it persists on reload and in the shared view.

This is idempotent — the raw [N] markers in the stored body are never overwritten; the style is applied at read time, so you can switch styles freely.

Copying and Downloading

The report viewer has Copy and Download (.md) buttons. Copy puts the full report text (body + bibliography, plain text) on the clipboard. Download saves a .md file you can open in any Markdown editor or paste into Notion, Obsidian, Word, etc.

Deleting a Verify Report

Open the report (in the reading pane) and click the Delete button (trash icon). A confirmation prompt appears. Deleting a report does not delete the underlying sources — they remain in your Cafe's Sources rail.

---

Durability — Runs in the Background

The verify loop runs as a background job — the same as Auto-Research. Once you click Verify:

• You can close the dialog, switch tabs, navigate away, or close the browser — the run continues server-side.
• The verdict and cited report are saved directly to the database. When you come back, the source's rail row shows the resolved badge and the report appears in the Reports group.
• You do not need to keep the dialog or the Cafe open while the loop runs.

---

Tips

• Verify scientific or health claims first — these are highest-stakes and best served by cross-referencing peer-reviewed literature.
• One claim at a time is usually better than "verify all" — a focused single-claim verify costs less and produces a sharper verdict than a multi-claim run with the same budget.
• Raise the token ceiling for contested topics — if a topic has genuine scientific disagreement, the agent needs enough budget to find both sides.
• Check the evidence cards — every piece of evidence the loop finds is a real source added to your Cafe. These are genuinely useful for writing — cite them directly, add them to your Tray, or use them in an Auto-Write report.
• "Couldn't verify" is useful — it tells you the claim is opinion-only or underdocumented, which is important context before citing it.
• Use the cited report for writing — the report is a ready-made, cited answer to the claim. Copy it, download it, or share it directly. Citation style can be changed at any time.
• The evidence panel and the report are complementary — the evidence panel (click "View evidence") shows the raw agent output; the cited report is the synthesised, readable version. Keep both open for deep investigation.

---

Availability

All features are gated by your token balance — if your balance is empty, the verify button will prompt you to top up.