Meridian Transaction Advisory provides M&A due diligence and post-close integration support to corporate acquirers in life sciences and industrials. James Hartley sits at the intersection of financial due diligence and post-close integration accounting — bridging deal-team assumptions and audited post-close numbers. His deliverables feed directly into stakeholder briefings and gap analyses where accuracy and auditability are non-negotiable.
The amendment layer quietly invalidated year-over-year comparisons
Hartley needed five acquisition-related metric series from a large-cap pharmaceutical acquirer's SEC EDGAR US-GAAP company-facts JSON: goodwill, intangible assets excluding goodwill, integration-related costs, contingent consideration, and impairment charges — each spanning multiple annual reporting periods.
The source file carries a structural trap. Later-filed amendments restate prior-period comparative values, creating duplicate entries that share a filing-year label but carry different period-end dates. Keying an extraction to filing year rather than period-end date silently retains stale figures. The discrepancy only surfaces when a downstream reviewer ties the derived table back to the raw JSON.
A second problem compounded the deduplication issue: the SEC US-GAAP taxonomy has no standalone deal-cost tag. Integration-related costs serve as the closest proxy but conflate post-close integration expenses with restructuring charges. Any deliverable using this series as an acquisition-expense proxy must carry an explicit disclosure, or a compliance reviewer will flag the entire analysis at the worst possible moment.
The result was a 10-plus-pass manual workflow: edit a Python script, re-run, spot-check summary tables against the raw JSON, correct year keys, and restart when a new misalignment appeared. The process consumed most of the available prep time before a single deliverable sentence could be written.
Energent.ai became the extraction, verification, and disclosure engine
Hartley loaded the raw company-facts JSON directly into an Energent.ai session and described the five series. The agent worked end-to-end without a context switch:
- Inspected the full file schema and identified relevant US-GAAP tags across all five series
- Flagged upfront that no standalone acquisition-cost tag existed in the taxonomy, proposed integration-related costs as the available proxy, and noted the disclosure requirement before writing a single line of extraction code
- Wrote a Python normalization script keyed to period-end calendar year rather than filing year, with explicit deduplication logic retaining the most recently filed value per period
- Ran an independent verification pass against the source JSON — catching a year-key misalignment that had persisted silently through several incremental fixes
- Embedded the integration-related-cost limitation in both the analysis narrative and the dashboard labels before Hartley reviewed the draft
- Produced a structured analysis post and an interactive HTML dashboard from the validated series, covering the company's full acquisition and write-down history
No manual script debugging. No downstream verification cycle. No separate disclosure memo.
Period-end keying and in-session verification closed the loop
- Period-end date as the period anchor. Filing-year labels produce silent mismatches when amendments restate prior periods; the period-end date embedded in each fact entry is the only reliable deduplication key for this file structure.
- Verification against the same source file. The agent's check ran against the raw JSON the extraction pulled from — discrepancies surfaced inside the workflow, not in a downstream review cycle.
- Proactive taxonomy disclosure. The proxy limitation was embedded in the deliverable outputs before review, not added after a compliance flag.
- Single-session scope. Collapsing extraction, verification, and dashboard production into one run removed the multi-step handoff that had been the bottleneck.
Ten-plus passes replaced by one verified session
- Five acquisition-related series extracted, deduplicated, and year-aligned from a source that had previously required 10-plus manual passes to normalize
- One silent year-key misalignment caught and corrected during the in-session verification pass
- Both deliverables — structured analysis post and interactive HTML dashboard — ready for stakeholder review without a further normalization or annotation pass
- Analyst time shifted from debugging extraction scripts to interpreting results and stress-testing the proxy assumption
"That's the kind of thing a junior analyst would miss entirely — and it's the thing a reviewer catches at the worst possible moment. The agent surfaced it before I even asked." — James Hartley, Post-Close Integration Analyst at Meridian Transaction Advisory