-
Notifications
You must be signed in to change notification settings - Fork 0
Showcase Workflows
Three end-to-end workflows that demonstrate the full GATHER → VERIFY → INTERPRET pipeline. Each workflow uses the same gather and verify steps and differs only in the st-verdict lens at the interpret step.
If you've never run Cross before, do the Onboarding first to set up your API keys.
| Workflow | Question | Lens |
|---|---|---|
| A — "Is this fake news?" | What's wrong with this report? | --what-is-false |
| B — "What's missing?" | What did the report fail to cover? | --what-is-missing |
| C — "What can I trust here?" | Which claims are well-supported? | --what-is-true |
After running any of the three, ask st-verdict --how-to-fix for a concrete next-action recommendation.
Use when: you've stumbled on a news article, a forum post, or an AI-generated report and you want to know which specific claims are wrong before citing or sharing it.
# 1. GATHER — pull the article into a container, extract the prompt
st-fetch https://example.com/article-about-rv-batteries --prep
# 2. VERIFY — fact-check it with every configured AI provider
st-cross article.json
# 3. INTERPRET — focused breakdown of what is false / disputed
st-verdict -s 1 --what-is-false --ai-summary article.json$ st-verdict -s 1 --what-is-false --ai-summary article.json
Generating Summary [false-lens] with xai…
The most critical inaccuracy in the report is the claim that annual maintenance
costs for lithium RV batteries range from $100 to $200, marked partially_false
by three of five fact-checkers (xai:grok-4-1, anthropic:claude-opus-4-5,
perplexity:sonar). This figure understates typical costs, which manufacturer
data and dealer service plans place between $300 and $700 depending on system
size and inverter integration.
Remaining false-lens claims cluster into two themes: (1) Lifecycle ratings —
the assertion that a 100 Ah lithium pack delivers "10,000+ cycles at 80%
depth-of-discharge" was rated partially_false (xai, gemini); manufacturer
specs cite 3,500–6,000 cycles for that threshold. (2) Charge-controller
compatibility — the article claimed any MPPT controller works with lithium,
which the fact-checkers flagged as misleading because absorption-voltage
profiles must be lithium-specific to avoid undercharging.
Authors and readers should verify all maintenance-cost and lifecycle figures
against current manufacturer datasheets and at least one independent installer
quote before citing this article.
The lens names specific claims, names which fact-checkers flagged each one (signal strength comes from agreement), groups them into themes, and closes with actionable advice. Compare with the chart-only output of plain st-verdict article.json, which only ranks providers.
Use when: the report seems plausible but you suspect it has gaps — or you're using Cross to evaluate AI authors and want to know what each one failed to address relative to the prompt's intent.
# 1. GATHER (same as Workflow A)
st-fetch https://example.com/article-about-rv-batteries --prep
# 2. VERIFY (same as Workflow A)
st-cross article.json
# 3. INTERPRET — coverage-gap analysis
st-verdict -s 1 --what-is-missing --ai-summary article.json$ st-verdict -s 1 --what-is-missing --ai-summary article.json
Generating Summary [missing-lens] with xai…
The most important omission in the report is the absence of any discussion of
warranty terms — a knowledgeable reader of the prompt "compare RV battery
options for a six-month trip" would expect to see warranty length and coverage
scope addressed because batteries are the single highest-cost failure point on
long-haul RVs and warranty claims are the primary risk-mitigation lever.
Two secondary gaps stand out. First, there is no mention of cold-weather
performance — lithium iron phosphate (LiFePO4) packs lose 20–40% of usable
capacity below 0°C and most ship without integrated heaters; for a six-month
trip this is critical. Second, the report omits any installation-cost figures
or licensed-installer requirements; readers are left to assume DIY but state
electrical codes vary on whether high-amperage inverter wiring is permitted.
To complete the report, the author should add a warranty-comparison subsection,
a cold-weather performance table, and an installation-cost / regulatory note.
The missing lens reads data[0].prompt and the report markdown together so the AI can reason about what should be there relative to the topic, not just what is there. Use this when the chart says "scores are fine" but you have a nagging sense that the report is incomplete.
Use when: you want to extract the trustworthy core of a report — the claims that multiple fact-checkers verified — so you can confidently cite, quote, or re-publish them. The positive counterpart of Workflow A.
# 1. GATHER (same)
st-fetch https://example.com/article-about-rv-batteries --prep
# 2. VERIFY (same)
st-cross article.json
# 3. INTERPRET — focused breakdown of verified claims
st-verdict -s 1 --what-is-true --ai-caption article.json$ st-verdict -s 1 --what-is-true --ai-caption article.json
Generating Caption [true-lens] with xai…
The article's strongest verified thread is its energy-density comparison: the
claim that LiFePO4 packs deliver roughly 90–110 Wh/kg versus 30–40 Wh/kg for
flooded lead-acid was rated true by all five fact-checkers, with three citing
the same independent industry datasheet. This is the single most citation-ready
fact in the report.
Supporting verified claims cluster around two themes: depth-of-discharge
(LiFePO4's safe ~80–95% versus lead-acid's ~50%, marked true by four checkers)
and charge efficiency (LiFePO4's ~99% Coulombic versus lead-acid's ~85%, marked
true by three). For a buyer's-guide repost, lead with the energy-density
figure and use the depth-of-discharge claim as the supporting headline; both
clear the multi-checker confidence threshold.
A 100–160-word caption is usually the right detail level for this lens — long enough to cite specific evidence, short enough to drop into a Discourse post or a forum reply. Use --ai-summary if you want a 3-paragraph technical summary instead.
| You want to… | Lens | Typical detail flag |
|---|---|---|
| Find errors before citing | --what-is-false |
--ai-summary |
| Find coverage gaps | --what-is-missing |
--ai-summary |
| Extract the trustworthy core | --what-is-true |
--ai-caption |
| Decide what to do next | --how-to-fix |
--ai-short (default) |
| Just see the chart | (none — default behaviour) |
--ai-short (default) |
The four lenses are mutually exclusive — pick one per invocation. Run multiple in sequence to get a full picture; chaining is cheap because st-cross's fact-check results are already in the container and st-verdict only re-reads them.
Related: Three Stages · st-verdict · st-fetch · st-cross · Container Format