How ConCallIQ keeps generated earnings pages grounded
A short explanation of why generated pages must pass schema checks, filing-backed financial validation, and internal link validation before publishing.
Generated text is cheap. Trustworthy generated pages need constraints. ConCallIQ validates required fields, labels pending financial verification clearly, emits source links where available, and keeps the not-investment-advice boundary visible on every page.
The three-gate publishing model
Every generated earnings page on ConCallIQ has to pass three gates before it appears on the public site.
Gate 1: Schema validation
Each generated page must populate a fixed structured representation: a summary, a guidance block, a risks block, a key-numbers block, and a sentiment label. Pages with missing or malformed fields are rejected automatically.
Gate 2: Filing-backed financial check
Revenue, EBITDA, EBITDA margin, PAT, and YoY/QoQ deltas are cross-checked against the official BSE/NSE filing for the quarter. We source figures from:
| Source | Used for | Notes |
|---|---|---|
| Screener.in | Headline P&L numbers | Fast, structured, widely used |
| NSE XBRL | Segment-level breakdowns | Authoritative for listed entities |
| BSE corporate announcements | PDF of investor presentation | Catches recent restatements |
| Company investor PDFs | Management-provided slides | Last resort, manually parsed |
The tolerance is 1%. Anything outside that gets the page held back, the AI-extracted number replaced with the filing number, and a verification badge attached to the published version.
Gate 3: Internal link validation
Generated pages link out to related quarters, peer comparisons, and sector pages. The build process verifies every outbound internal link resolves to a real route before publishing.
What a published page tells you about its own trust level
Every ConCallIQ company page carries one of three financial-verification states:
- Verified. Numbers match filings within 1% tolerance.
- Pending verification. Numbers are AI-extracted from the transcript, awaiting filing reconciliation.
- Discrepancy resolved using filings. AI extraction disagreed with filings and the filing values were used.
The badge is visible at the top of every page. Users can click through to the filing source where one is linked.
Why we are this paranoid
Three reasons:
- AI extraction errors compound. A wrong revenue number on a single page is one mistake. A wrong revenue number on the latest page changes the YoY/QoQ deltas, the sector ranking, and the comparison page — five downstream errors from one upstream one.
- Trust is the product. A research site that occasionally publishes a wrong PAT is not a research site, even if 99% of its pages are correct.
- It is cheap to add and expensive to retrofit. Building the verification gate into the pipeline from day one is much cheaper than auditing thousands of generated pages later.
What we do not do
- We do not republish brokerage research.
- We do not present forward forecasts as facts.
- We do not include price targets.
- We do not collect or display portfolio data.
Nothing on ConCallIQ is investment advice. Everything is informational, sourced, and verifiable.
Related: Management commentary is data, Methodology.