ololand-forensic-qoe-ebitda-bridge

Codex wrapper for /ebitda-bridge

Follow the OloLand command instructions below when the user asks for /ebitda-bridge or the equivalent workflow in Codex.

EBITDA Bridge

Help the user review management's add-backs without treating computed scenario totals or heuristic classifications as accepted normalized earnings. Confirm the connected tool's current live schema, availability, authorization, and execution requirements before invoking it. Package instructions do not prove that a backend contract has been deployed or admitted to the current host.

Usage

/ebitda-bridge <deal_id>

Arguments

• deal_id (required) — An accessible deal selected by the user.
• On OpenAI, financial_data is also required: a finite numerical reported_ebitda and an explicit adjustments list. Each supplied row needs label and numerical amount; use only optional fields accepted by the live schema. Do not invent missing values, coerce text or booleans to numbers, or supply classification, confidence, approval, or canonical-basis overrides.
• An explicitly supplied empty list is the caller's assertion, not evidence that no adjustments exist. Missing or null inputs remain a gap.

Execution

1. On OpenAI, ask for missing explicit values and the adjustment schedule; call build_ebitda_bridge(deal_id, financial_data=financial_data) only after the live schema and the user's authorization are confirmed. Never make a deal-only call or infer the inputs from a stored financial snapshot, CIM, workbook, or data-room extraction on this path. On other hosts, permit build_ebitda_bridge(deal_id) without requiring explicit supplied values when the current live schema supports native snapshot fallback. Preserve the response's actual basis and evidence limitations.
2. The OpenAI bridge is a pure caller-supplied scenario. It does not retrieve source documents, verify supplied source claims, approve adjustments, or establish a canonical earnings basis. Any recurrence, categorization, documentation, tie-out, or owner-compensation checks are bounded heuristics over supplied inputs. A supplied documentation label does not prove support.
3. Report actual returned arithmetic and flags. In the OpenAI output, base_ebitda names the supplied base; the input key reported_ebitda does not prove the company reported it. Preserve ebitda_basis_label, basis_is_canonical=false, input_provenance=caller_supplied_unverified, and evidence_verification=not_performed. Explain any returned heuristic_quality_score or heuristic_risk_level as a heuristic of the supplied scenario, not company earnings quality or approval.
4. Preserve returned totals, deductions, tie-out gaps, projection gaps, and partial states. Do not claim the engine excluded questionable add-backs: OpenAI totals have totals_basis=all_supplied_adjustments. Never turn a returned classification into analyst acceptance or remove rows to manufacture a defensible total. Zero-base undefined ratios/scores remain gaps even when totals are computable.
5. Distinguish supplied source/documentation labels and returned source_claim fields from real citations. A caller claim is not an authorized File/page identity, a verified source quote, or an evidence receipt. List the evidence still needed to assess recurrence, supporting amounts, replacement compensation, and adjustment eligibility. Independent source verification and authorized human judgment remain necessary for accepted normalized earnings; this bridge alone cannot establish them.