What Makes a Deliverable Board-Ready: The 18-check quality gate

The Board Meeting Test

Here's a scenario that plays out in boardrooms every quarter: a founder presents a market analysis slide showing a $50 billion TAM. A board member asks, "Where did that number come from?"

The founder pauses. "Our consulting firm provided it." The board member pushes: "What was their source? What methodology did they use? Can I see the underlying data?"

This is the moment most consulting deliverables fail. Not because the analysis was wrong, but because the evidence trail was never built.

The stakes are not hypothetical. In Mata v. Avianca, Inc., a federal judge sanctioned two attorneys $5,000 after they filed a legal brief citing six court decisions that ChatGPT had invented — authoritative-looking cases that did not exist (Mata v. Avianca, Inc., 2023). The document read as credible. It simply was not true. A market analysis with an unsourced TAM figure fails the same test in a boardroom, just more quietly.

At Sagentix, every deliverable passes through an 18-check automated quality gate before it reaches a client Sagentix 18-check quality gate, 2026. These aren't optional review steps — they're automated checks that block delivery until the standard is met. The gate comprises 6 core checks (detailed below — citation density, declarative titles, structural completeness, evidence citations, APA format, top-tier structure) plus 12 additional automated validations covering anti-hallucination source-integrity verification, claim provenance coverage, cross-document consistency, document architecture conformance, CI version history, knowledge-search validation, subscription data utilization, regulatory terminology, portal data embedding, unfilled-placeholder detection, industry access-motion conformance, and numeric consistency. Four further checks are phase-specific rather than universal — three that only run on LinkedIn-strategy work, a competitor price-availability verdict that only runs on market intelligence, and an AI-search measurement-discipline check that only runs on a digital audit — so 23 checks are implemented and 18 of them run on every deliverable. That distinction is deliberate: the number quoted here is the one you can hold me to on your engagement, whichever phase it is. The 6 core checks below explain the why behind the architecture.

There is a name for a document built this way. In safety-critical engineering, you don't get a system certified by asserting it is safe — you submit an assurance case: a structured, evidence-backed argument that every claim holds, where each sub-claim carries at least two independent evidence paths and the whole argument is mechanically checked for gaps, contradictions, and unsupported leaps (Muram & Javed, 2023). A board interrogating your TAM figure is running exactly that kind of audit. The 18-check quality gate exists so that every deliverable ships with its assurance case already built — and it has only grown since: Check 9 alone now runs twelve architecture sub-gates (G8–G19), including a hard-fail third-party data-source disclaimer and a rule that every external claim carry two independent, hyperlinked sources.

All 18 Checks

Every deliverable, in every phase, passes the same eighteen automated checks before a human reads it. The numbering below is the gate's own, so a finding in a report traces straight back to the check that raised it Sagentix 18-Check Quality Gate, 2026. Six of them are worth explaining properly, and the sections after this one do that.

Is it finished? — Check 1 no unfilled {{placeholder}} survives into a delivered document · Check 3 every section the phase specification requires is present, in order · Check 4 measured word count against a floor that scales with engagement complexity rather than a fixed target.

Is every claim sourced? — Check 2 any [PROOF NEEDED] marker left in the draft blocks delivery · Check 5 citation density against a threshold, so evidence is distributed through the argument rather than clustered in one well-sourced section · Check 6 APA 7th in-text citations and a complete References section · Check 13 each load-bearing claim traced to the specific source supporting it, recorded rather than asserted · Check 15 the licensed research actually pulled for the engagement has to appear in the analysis, because paying for a source and not reading it is a silent failure.

Does it read like a top-tier document? — Check 7 every section title states a finding, not a topic · Check 8 SCQA structure, scenario analysis and explicit risk treatment, phase-appropriate · Check 9 sixteen architecture sub-gates covering the three-layer reading model, formatting, and the third-party data disclaimer, scored as a conformance percentage.

Could any of it be invented? — Check 11 every claim extracted from research is verified against its source, and a claim marked CONTRADICTED blocks delivery outright · Check 12 the anti-hallucination gate: unresolved web-source tags, cited industry codes checked against the briefs actually held, and citations appearing in the text but not the References · Check 22 numeric consistency, because a stated count is a claim about the document itself — "four of six align" above a table showing three survives every fact-check, since each individual fact is correct.

Does the whole engagement still agree with itself? — Check 10 refreshed deliverables carry an accurate version table · Check 14 figures and conclusions reconciled across every phase delivered so far, not just within the document in hand · Check 16 the structured data behind the client dashboard is present and complete, so what the client reads and what the portal displays cannot drift apart · Check 21 the go-to-market motion recommended has to be one that works in the target industry's buying system.

Five further checks run only where they apply — competitor price availability in market intelligence, three LinkedIn-specific checks in the LinkedIn strategy phase, and AI-search measurement discipline in the digital audit — plus three sub-checks. That is 23 implemented in total. The 18 above are the ones every deliverable passes, which is why 18 is the number quoted.

In Depth: APA 7th Citations on Every Claim (Checks 5 and 6)

Every factual claim in a Sagentix deliverable carries an in-text citation in APA 7th edition format: (Author, Year) linked to a full reference entry with report numbers and publication details Sagentix 18-check quality gate, 2026.

Why it matters: The citation isn't decoration. It's a verification path. When a board member questions a market size figure, the founder can point to a specific NAICS-coded industry brief — for example, a management consulting industry profile (NAICS 541611) sourced from VerticalIQ — that can be independently retrieved and verified (VerticalIQ, 2026). No ambiguity. No "our consultant's estimates."

How it differs from typical output: Most consulting deliverables cite sources loosely — based on industry research, or according to market estimates. AI-generated analyses often cite nothing at all, or worse, cite sources that don't exist. In a peer-reviewed audit of GPT-generated bibliographies, 55% of GPT-3.5 and 18% of GPT-4 citations were outright fabricated — and among the real citations, 43% (GPT-3.5) and 24% (GPT-4) still contained substantive errors in volume, pages, or year (Walters & Wilder, 2023). It is not confined to bibliographies: when Stanford researchers asked leading models direct, verifiable questions about randomly selected federal court cases, the models hallucinated between 58% (GPT-4) and 88% (Llama 2) of the time (Dahl et al., 2024); in a clinical literature test, a chatbot produced 35 references of which only two were real (McGowan et al., 2023). Fabrication on checkable professional facts is a baseline behaviour, not an edge case.

That is why the gate's source-integrity check does not just count citations. It applies the leading peer-reviewed method for measuring factual precision: decompose the deliverable into atomic factual claims and verify the percentage supported by a reliable knowledge source (Min et al., 2023). When the method's authors ran it on ChatGPT's long-form biography generation, factual precision was only 58% — roughly four in ten confident "facts" unsupported by a reliable knowledge source (Min et al., 2023). A companion technique cross-checks the same claim across multiple samples, on the logic that a grounded model stays consistent while a hallucinating one contradicts itself (Manakul, Liusie, & Gales, 2023). Claims that cannot be traced are removed or re-anchored — never silently presented as fact Sagentix Phase 10 Evidence Discipline, 2026.

In Depth: Research Data with Page-Level Provenance (Checks 13 and 15)

Market data points sourced from premium industry research include the specific report code and, where possible, the section from which the data was extracted. This isn't a courtesy — it's an audit trail.

Why it matters: Premium industry research reports like VerticalIQ's NAICS-indexed briefs are the Tier A standard for industry analysis in management consulting. But citing a generic source for a market size figure is like citing "Google" for a search result. The NAICS code (e.g., 541611 for administrative and general management consulting) and the specific section (Canadian Market, Industry Risks, Current Conditions) allow anyone to verify the claim against the original source (VerticalIQ, 2026).

How it differs: Traditional consulting firms often have premium research access but rarely cite at this level of specificity. The data enters the analysis as "industry revenue is approximately $X billion" without provenance. When challenged, the analyst might remember which report they used — or might not Sagentix 18-check quality gate, 2026.

In Depth: Declarative Action Titles (Check 7)

Every H2 and H3 heading in a Sagentix deliverable is a complete declarative sentence that states a finding, not a topic label.

Instead of: "Market Overview" The H2 reads: "Canada's software publishing market reached $23.2B across roughly 3,300 publishers" — a finding anchored to a NAICS-coded brief (VerticalIQ, 2026), not a topic label

Why it matters: This is the Pyramid Principle, formalized by Barbara Minto at McKinsey in the 1960s–70s and still a widely adopted standard for management consulting communication (Minto, 2009). A reader should be able to read only the headings of a document and understand the complete argument. Topic labels ("Market Overview," "Competitive Analysis") tell the reader what category of information follows, but not what the analysis found.

How it differs: Most consulting deliverables — including those from major firms — default to topic labels because they're easier to write. Declarative titles require the analyst to commit to a specific finding in every section, which imposes discipline on the analysis itself (Minto, 2009).

In Depth: Cross-Phase Integration (Check 14)

Every Sagentix engagement produces 10 interconnected phase deliverables. The cross-phase check verifies that data, claims, and recommendations are consistent across all of them Sagentix GTM Methodology, 2026.

What it catches:

• A TAM figure in Phase 01 (Market Intelligence) that doesn't match the TAM cited in Phase 04 (Pitch Deck)
• Pricing tiers in Phase 06 that contradict the positioning established in Phase 02 (Value Proposition Design)
• Competitor names or market share figures that differ between phases
• Strategic recommendations in Phase 08 (Strategy Execution) that reference capabilities not established in earlier phases

Why it matters: In traditional consulting engagements, different phases are often written by different team members at different times. Data drift is common — a market size gets rounded differently, a competitor gets added in one phase but not another, pricing math doesn't reconcile. Cross-phase inconsistencies destroy credibility because they signal that the analysis is assembled, not integrated Sagentix 18-check quality gate, 2026.

How it differs: Most consulting firms review individual deliverables in isolation. Cross-phase consistency checks are rare because they require comparing every factual claim across hundreds of pages. Automated assurance-case review frameworks treat exactly this — consistency management across evolving artifacts, with defeaters flagged where a later claim undercuts an earlier one — as a first-class check (Muram & Javed, 2023). Automated tooling makes this feasible at a level that manual review cannot match Sagentix Phase 10 Evidence Discipline, 2026.

In Depth: Evidence Level Tagging, L1–L5 (Check 12)

Every claim in a Sagentix deliverable is tagged with its evidence level Sagentix Phase 10 Evidence Discipline, 2026:

• L1 — Primary research: Direct data from premium industry research (VerticalIQ), government databases (BLS, StatCan, Census), or regulatory bodies
• L2 — Validated secondary: Claims verified against at least one authoritative source (Gartner, Forrester, McKinsey, HBR)
• L3 — Corroborated estimate: Data triangulated from multiple secondary sources
• L4 — Analytical inference: Logical conclusion drawn from L1-L3 data, clearly marked as inference
• L5 — Assumption: Stated assumption with rationale, flagged for client validation

Why it matters: Not all evidence is created equal. A market size from a NAICS-coded industry profile (L1) carries more weight than a competitor revenue estimate triangulated from press releases (L3), which carries more weight than a growth rate assumption (L5). Evidence tagging lets the reader assess confidence levels without having to evaluate each source independently Sagentix Phase 10 Evidence Discipline, 2026.

How it differs: This concept exists in academic research (peer review, meta-analysis frameworks) but is virtually absent from management consulting. Most strategy deliverables present all claims at the same confidence level, leaving the reader to guess which numbers are solid and which are estimates Sagentix 18-check quality gate, 2026.

In Depth: Top-Tier Formatting Standards (Check 9)

The final core check validates structural and formatting elements against the standards used by top-tier strategy firms Sagentix 18-check quality gate, 2026:

• 3-Layer Reading Model: Every deliverable supports three reading depths — 5-minute executive scan, 15-minute management review, and 45-minute deep read
• SCQA Executive Summary: Situation-Complication-Question-Answer structure, the consulting opening pattern codified by Barbara Minto (Minto, 2009)
• Callout boxes: Eight standardized callout types — Implication, So What, Strategic Insight, Data Callout, Cross-Reference, Methodology Note, Vertical Takeaway, and Key Takeaway
• Part dividers with 3-5 sentence summaries for each major section
• Version History table tracking all changes and their rationale

Why it matters: Formatting isn't aesthetic — it's cognitive. The 3-Layer Reading Model exists because a CEO, a VP of Strategy, and a market analyst all need to use the same document differently. The SCQA framework exists because it forces the executive summary to tell a story, not list findings (Minto, 2009). These structures have been refined over five decades of management consulting practice because they work.

The Compound Effect

Any one of these 6 core checks improves deliverable quality. Combined with the additional automated validations — which catch anti-hallucination failures, missing claim provenance, cross-document inconsistencies, regulatory terminology errors, unfilled placeholders, and packaging issues — they create the full 18-check automated quality gate: a system that doesn't just present conclusions, but makes its entire reasoning process transparent and verifiable Sagentix 18-check quality gate, 2026.

The design principle behind it comes from published research. The most instructive recent work pairs a generative model with a deterministic verifier: a large language model proposes the structure and the claims, but a formal method checks each one and rejects the unsound — the formal layer explicitly exists to "constrain LLM hallucinations," because the model's raw output "contains hallucinations" that must be filtered (Chen, Deng, & Du, 2025). That is the Sagentix architecture in one sentence: the model drafts, the deterministic gate verifies, and nothing ships on the generative layer's word alone. And the gate is only the last automated line — it runs after a client due-diligence pass, a mandatory dual web-and-knowledge search, a knowledge-validation stage that blocks contradicted claims, and a nine-stage anti-hallucination orchestrator, four of whose stages can halt delivery outright.

One of those four is worth naming on its own, because it covers ground a fact-check structurally cannot: citation integrity. Verifying that a claim is true and verifying that its source exists are two different retrievals over two different indexes, and they fail independently — the reference benchmark for cited AI text scores correctness and citation quality as separate dimensions, and even the strongest systems lack complete citation support 50% of the time on one of its datasets (Gao et al., 2023). So every in-text citation must resolve to exactly one entry in the reference list, by exact-identity match rather than similarity, or the deliverable does not ship Sagentix 18-check quality gate, 2026. Identity is doing real work in that sentence: a fabricated citation is typically assembled from real components — a real author, a real journal, a plausible year — so anything scored by resemblance rates it highly. In one clinical test, a model produced 35 references of which 2 were real, 12 were near-misses of genuine manuscripts, and 21 were a "pastiche of multiple existent manuscripts" (McGowan et al., 2023). Resemblance is the property a good fake already has, which is why the check cannot be a similarity score.

This is the standard that boards expect from top-tier strategy firms. It's the standard that survives investor due diligence — and due diligence is where outcomes are actually decided: in a survey of 885 venture capitalists, investors rated deal selection, the screening-and-diligence step, as the single most important of their activities, ahead of sourcing and post-investment support (Gompers et al., 2020). A board interrogating your evidence is running that same screen. It's the standard that separates strategy from opinion.

A deliverable is board-ready when the board can interrogate any claim in the document and find a verifiable source, a clear evidence level, and a logical chain connecting the data to the recommendation. Anything less is a briefing note, not a strategy.

Where This Leaves You

The Sagentix delivery model runs every phase through the 18-check quality gate before it leaves my desk — automated checks on citation density, declarative titles, evidence provenance, regulatory terminology, and anti-hallucination traces, followed by a founder-signed final review Sagentix 18-check quality gate, 2026. The platform sits on 1,412 curated artifacts, 10 phases end-to-end, 6–8 week delivery at CA$4.5K–$45K, with Phase 1 shipping under a money-back guarantee (subject to terms).

Which board question hits hardest in your case right now — TAM defensibility, pricing rationale, or competitive moat? Most deliverables get one of those three right and stumble on the other two.

References

• Chen, Z., Deng, Y., & Du, W. (2025). Trusta: Reasoning about assurance cases with formal methods and large language models. Science of Computer Programming, 244, 103288.
• Dahl, M., Magesh, V., Suzgun, M., & Ho, D. E. (2024). Large legal fictions: Profiling legal hallucinations in large language models. Journal of Legal Analysis, 16(1), 64–93.
• Gao, T., Yen, H., Yu, J., & Chen, D. (2023). Enabling large language models to generate text with citations. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (pp. 6465–6488). Association for Computational Linguistics.
• Gompers, P. A., Gornall, W., Kaplan, S. N., & Strebulaev, I. A. (2020). How do venture capitalists make decisions? Journal of Financial Economics, 135(1), 169–190.
• Manakul, P., Liusie, A., & Gales, M. J. F. (2023). SelfCheckGPT: Zero-resource black-box hallucination detection for generative large language models. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (pp. 9004–9017). Association for Computational Linguistics.
• Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023). Justia.
• McGowan, A., Gui, Y., Dobbs, M., Shuster, S., Cotter, M., Selloni, A., Goodman, M., Srivastava, A., Cecchi, G. A., & Corcoran, C. M. (2023). ChatGPT and Bard exhibit spontaneous citation fabrication during psychiatry literature search. Psychiatry Research, 326, 115334.
• Min, S., Krishna, K., Lyu, X., Lewis, M., Yih, W., Koh, P. W., Iyyer, M., Zettlemoyer, L., & Hajishirzi, H. (2023). FActScore: Fine-grained atomic evaluation of factual precision in long form text generation. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (pp. 12076–12100). Association for Computational Linguistics.
• Minto, B. (2009). The Pyramid Principle: Logic in writing and thinking (3rd ed.). Pearson Education.
• Muram, F. U., & Javed, M. A. (2023). ATTEST: Automating the review and update of assurance case arguments. Journal of Systems Architecture, 134, 102781.
• Sagentix Advisors Inc. (2026). Sagentix 18-check quality gate — citation density, declarative titles, and evidence provenance. Sagentix Advisors Inc.
• Sagentix Advisors Inc. (2026). Sagentix GTM Methodology — 10-phase engagement architecture and cross-phase integration. Sagentix Advisors Inc.
• Sagentix Advisors Inc. (2026). Phase 10 Evidence Discipline — L1–L5 tagging and anti-hallucination verification. Sagentix Advisors Inc.
• VerticalIQ. (2026). Management consulting services industry profile (NAICS 541611). VerticalIQ.
• VerticalIQ. (2026). Software publishers industry profile (NAICS 511210). VerticalIQ.
• Walters, W. H., & Wilder, E. I. (2023). Fabrication and errors in the bibliographic citations generated by ChatGPT. Scientific Reports, 13, 14045.