Verifier Stack

OloLand's verifier stack mechanically checks that every number in an artifact is supported by the deal's source documents, that every citation marker points at a real retrievable chunk, and that the deal's stored DCF identities hold. It is the "defend every number, cite every source" guarantee in code, not in marketing copy.

What the verifiers actually do

verify_inline_numbers

Extracts every dollar amount and every percentage from the artifact text. For each number, checks whether it appears (within 5% relative / 0.5pt absolute tolerance) in at least one chunk retrieved from the deal corpus.

• Catches: fabricated revenue figures, transposed digits (e.g. $34.2M vs $32.4M), unit drift ($304K reported as $304M).
• Misses: numbers that happen to appear in the corpus by coincidence (e.g. a generic "$1M" mentioned in a risk factor section); numbers that are derived but never raw in the source (e.g. growth rates the analyst computed correctly but the chunk only shows the underlying revenues).

verify_claim_chunks

Stricter than verify_inline_numbers. Splits the text into sentences containing financial numbers. A sentence passes only when all of its numbers can be matched in a single chunk.

• Catches: sentences that stitch together unrelated numbers from different sources (a tell for hallucinated synthesis).
• Misses: the same coincidence cases as above.

verify_citation_existence

For artifacts that use the legacy [Source: file=..., p.N] block format: extracts every block and validates that the filename exists in the deal's document set (with fuzzy matching) and the cited page is in range.

• Catches: invented filenames, wrong page numbers.
• Skipped: when the artifact uses the modern [N] marker format only — that's audited by check_citation_coverage.

verify_dcf_arithmetic

Replays the two DCF identities against the deal's most recent DCFRun:

1. enterprise_value == pv_fcf_sum + pv_terminal_value
2. equity_value == enterprise_value - net_debt

Tolerance: 0.1% relative.

• Catches: internally inconsistent DCF models — a load-bearing check, because every number cited from a broken DCF is suspect even if it appears verbatim in the source.
• Status `no_data` when the deal has no DCFRun — pass include_dcf_arithmetic=false if the artifact does not reference a DCF.

check_citation_coverage ([N] marker audit)

Counts [N] markers in the artifact, retrieves the same chunk set the verifier uses, and reports markers that fall outside the available range. The orchestrator drops out-of-range markers at turn end; this lets external callers see the same view.

• Catches: markers the model invented (e.g. [12] when only 8 chunks were retrieved).
• Misses: markers that are in-range but cite the wrong chunk (use verify_claim_chunks for that).

reconcile_documents

Walks the CPA > tax > management > AI source hierarchy across FinancialDataSnapshot and AIExtractedFinancialMetric observations. Returns per-metric reconciliation status (reconciled / discrepancy / unverified), recommended values, discrepancy spreads, and overall confidence.

• Catches: silent disagreements between CPA-audited financials and management-supplied numbers; AI-extracted figures that contradict primary sources.
• Use when: the artifact is going to IC or a regulator-facing deliverable. Skip for back-of-envelope work — it's heavier than the per-number verifiers.

Interpreting the verdict

What the verifier is NOT

• Not a fact-checker for prose. It does not judge whether a qualitative claim ("the company has strong moats") is true. It only checks numbers and citations.
• Not a grader. It is a precondition for grading. A high grader score on an artifact with verifier-FAIL is structurally suspect.
• Not a replacement for human IC review. It catches the mechanical failure modes (fabricated numbers, dangling citations, broken DCF identities). Strategic, judgment-based review still belongs with the partner.

Methodology references

• Atomic-claim verifiers: backend/services/ai/scorers/inline_number_verifier.py, claim_chunk_verifier.py, citation_existence_checker.py
• DCF identity check: backend/services/harness/dd_tools.py::_verify_dcf_arithmetic
• Source hierarchy + reconciliation: backend/services/financial/cross_doc_reconciler.py
• Citation-marker validation: backend/services/agent_runtime/orchestrator.py::validate_citation_markers

Companion command: /verify.