v3 gave the pipeline a memory. v3.6 gives the user a relationship with it -- a coach before submission, a second score that separates the idea from the pitch, a plain-language explanation of every score, a fix-it plan after, and a way for real people and real partners to plug into the loop. This revision incorporates all nine conditions from the v3.5 critical review (4.1/10 consensus, Conditional Go): a beta-user validation gate before schema commitments, an honest financial model, a scoped 12-week MVP, a quality-scored Fix-It loop, an acknowledged consultant channel conflict, and a Minimum Viable Legal framework ahead of any White-Label pilot.
VerdictTank v3 solved memory: the pipeline now remembers every review it has ever run and checks whether its own predictions came true. That work stands. v3.6 does not replace it -- it wraps it in a product journey that starts before a user submits anything and continues after the verdict lands.
The trigger for v3.6 is a structured scan of the competitive landscape (Competitive Landscape Research Batch 001), covering adjacent categories: pre-submission coaching tools, URL-based intake products, community-scored idea platforms, audit-loop review patterns, no-code enterprise rules engines, and monitoring-plus-remediation SaaS. Seven concrete features and two cross-cutting strategic themes emerged as directly applicable to VerdictTank without diluting its core identity as a brutally honest, multi-judge critique engine.
Two themes recur across nearly every competitor studied and deserve top billing:
Theme 3, Transparency Sells: the products that explain why a score is what it is retain users at a materially higher rate than products that hand down a number and walk away.
Theme 5, The Fix-It Layer: the products that turn a one-shot assessment into a repeatable loop (assess, then help fix, then re-assess) convert a single transaction into a subscription relationship.
v3.6 integrates nine new capabilities directly into the v3 structure, spanning three product moments: before the review (Pre-Review Coach), during the review (Transparency Layer, Trust & Enterprise Controls), and after the review (Fix-It Layer, Distribution Expansion). None of these replace a v3 enhancement -- all nine sit alongside the existing corpus, prediction tracking, vertical templates, and API work, and several of them depend directly on v3 infrastructure already scoped (the sanitization gate now has more surfaces to cover; the corpus now stores two scores per review instead of one).
7 features from direct competitive research findings, 2 from cross-cutting strategic themes. All merged into existing sections, not appended.
The Fix-It Layer is the single highest-leverage change: reviewing stays paid, fixing is included, and users come back to re-score after acting on it.
Coach before, dual transparent score during, remediation plan and community signal after. VerdictTank stops being a single checkpoint and becomes a loop.
The Inner Circle beta cohort is the only population of real, paying VerdictTank users that currently exists. Before committing engineering time to any v3.6 capability, this cohort is surveyed and interviewed on four axes:
Baseline Net Promoter Score for the current v2 product, segmented by usage frequency, to establish a pre-v3.6 retention baseline against which any post-launch lift can actually be measured.
Open-ended and structured feature-request capture, coded against the nine v3.6 candidate capabilities to see how many were actually requested versus sourced purely from the competitive scan.
Exit and at-risk interviews with beta users who lapsed or downgraded, to identify whether the product's actual failure mode matches any of the nine proposed fixes or is something else entirely.
Product analytics on submission frequency, re-submission rate, and drop-off points in the existing single-score, single-shot flow, to identify where a Fix-It loop or dual score would actually intervene.
The survey and interview round is scheduled to complete before Phase 0 of the roadmap (Section 12) begins any schema work that would be costly to reverse. This is not a parallel-track nicety; it is a gate. Where the validation plan overlaps with an already-identified schema risk, most notably the dual-scoring bet described below, the same instrument and the same 25-user threshold serve both purposes rather than running two separate surveys.
Dual scoring (Section 6) is the one v3.6 feature that changes a permanent database schema before it ships, and it is explicitly gated behind this validation plan rather than assumed. The architecture (companion doc, Section 2.2) ships dual scoring first as a JSONB overlay, not as dedicated typed columns. The permanent-column migration only runs once a minimum of 25 beta users have used the dual-score overlay across at least two reviews each, and post-review survey signal shows a majority preference for two scores over the legacy single composite. If that bar is not cleared, the overlay is dropped and the corpus reverts to the legacy single score as the primary sortable column. See /tmp/verdicttank-v3.6-architecture.md, Section 2.2, for the full validation-gated migration plan.
Per the review's Priority Fix #1, every capability under consideration, whether carried forward from v3.5 or newly proposed in this revision, is tagged as either user-validated (supported by beta cohort signal collected under this plan) or competitive-scan-only (sourced from Competitive Landscape Research Batch 001, with no beta user signal yet). None of the nine v3.5/v3.6 capabilities currently carry user-validated status; that is the point of running this plan before committing further build time.
| # | Capability | Origin Tag | Validation Status |
|---|---|---|---|
| 1 | Chat-to-Refine | Competitive-scan-only | No beta signal collected yet; survey includes a direct question on pre-submission coaching interest. |
| 2 | URL-to-Review | Competitive-scan-only | No beta signal collected yet; usage-pattern analytics will show whether file-upload friction is actually a drop-off point. |
| 3 | Dual Scoring (Idea + Proposal) | Competitive-scan-only | Explicit validation gate defined above; ships as a reversible overlay, not a schema commitment, until 25 users clear the bar. |
| 4 | "Explain the Low Score" transparency | Competitive-scan-only | Plausible by category pattern (transparency correlates with retention broadly) but not confirmed against this specific user base yet. |
| 5 | Fix-It Remediation (MVP) | Competitive-scan-only | Churn-reason interviews will confirm or reject "no path to improve a bad score" as an actual reason users lapsed. |
| 6 | Second-Opinion Audit Agent | Competitive-scan-only | Deferred to post-MVP (Section 12); no beta demand signal exists, and none is expected until Enterprise pilots exist. |
| 7 | Configurable Review Rules Engine | Competitive-scan-only | Deferred to post-MVP (Section 12); revisit only once a specific Enterprise deal requires it. |
| 8 | White-Label for Consultants | Competitive-scan-only | Deferred to post-MVP (Section 12); also blocked on the consultant channel conflict validation in Section 09 and the Minimum Viable Legal framework in Section 17. |
| 9 | Roast/Boost Community Peer Review | Competitive-scan-only | Deferred to post-MVP (Section 12); no beta demand signal exists. |
This table is a living artifact, not a one-time exercise: it is re-run and re-tagged after the survey and interview round closes, and again after each feature's exit criteria (Section 12) are actually measured against real usage. A capability does not graduate from competitive-scan-only to user-validated on the strength of the survey alone; it graduates once real usage data confirms the survey signal.
Two layers of baseline matter for this proposal: the live v2 platform, and the v3 upgrade already scoped and conditionally approved on top of it.
Four cross-vendor reviewers today, ~$0.47/review fully loaded, 3-9 minute turnaround, tiered $19-299/mo pricing. No memory of past reviews, no vertical specialization, no API, no public-facing output, and sanitization/PDF generation both handled manually.
v3 (Section 3) is the conditionally-approved fix for the memory and infrastructure gaps: a review corpus, prediction-vs-outcome tracking, reviewer accuracy scoring, vertical templates, adversarial red-teaming, market simulation, shareable reports, an API, and the sanitization/PDF automation everything else depends on.
What v2 and v3 both still lack, and what surfaced repeatedly in the Batch 001 competitive scan: nothing helps the user before they submit, nothing explains a score in plain language, nothing tells the user what to do about a bad score, and nothing lets outside partners (consultants, accelerators) run the product as their own without being folded into the same generic White-Label tier as everyone else. That is the gap v3.6 closes.
Everything approved in v3 remains in scope for v3.6. This section is a compressed recap so the rest of this document reads as an integration, not a rewrite. Full detail lives in the v3 proposal.
Every review result, searchable: scores, flaws, conditions, predictions.
T+90/180/365 cron checks whether flagged risks actually materialized.
Judges earn a track record; future weighting follows the data.
Auto-classified by vertical, reviewed against industry benchmarks.
Industry-specific attack vectors: regulatory, compliance, churn dynamics.
12-month simulated trajectory replaces the static financial table.
Sanitized, shareable score cards and summaries, opt-in.
Full pipeline as an endpoint. Upload a document, get a structured review.
Every new proposal benchmarked against the corpus, by vertical.
Pre-deploy gate strips model/architecture details before anything goes public.
Scripted end-to-end report generation, required for API and Enterprise volume.
Nine new capabilities, sourced from Competitive Landscape Research Batch 001: seven direct feature findings (labeled by priority as scanned) and two cross-cutting strategic themes. Each is placed into the product moment it belongs to, not treated as a standalone bucket.
| # | Capability | Source Pattern | Priority | Product Moment | Timeline |
|---|---|---|---|---|---|
| 1 | Chat-to-Refine | Pre-submission AI coaching pattern | P1 | Before review | 1-2 wks |
| 2 | URL-to-Review | URL-based content intake pattern | P1 | Before review | 1-2 wks |
| 3 | Dual Scoring (Idea + Proposal) | Split-signal scoring pattern | P1 | During review | 2-3 wks |
| 4 | "Explain the Low Score" transparency | Strategic theme: Transparency Sells | P1 | During review | 2 wks |
| 5 | "Here's How to Fix It" remediation | Strategic theme: The Fix-It Layer | P2 | After review | 4-5 wks |
| 6 | Second-Opinion Audit Agent | Audit-loop review pattern | P2 | During review | 6-8 wks |
| 7 | Configurable Review Rules Engine | No-code enterprise rules pattern | P2 | Enterprise controls | 3-4 wks |
| 8 | White-Label for Consultants (dedicated track) | Branded-service reseller pattern | P2 | Distribution | 2-3 wks |
| 9 | Roast/Boost Community Peer Review | Community scoring pattern | P3 | After review | 3-4 wks |
Read together: items 1-2 lower the barrier to a good submission, items 3-4 make the resulting score legible and trustworthy, item 5 makes the score actionable and brings users back, items 6-7 harden the product for enterprises with real compliance and quality-control needs, and items 8-9 expand who VerdictTank reaches without diluting the core judgment engine.
Two new capabilities sit between the user and the pipeline, lowering the barrier to a submission worth reviewing at all.
Before submission, an AI chat helps the user structure their argument, surface gaps, and tighten clarity. Not a reviewer -- a coach. It never scores; it only helps the user get their best draft in front of the pipeline. Timeline: 1-2 weeks.
Paste a URL instead of uploading a file. The system extracts the content, structures it into the standard intake format, and runs it through the same pipeline. Removes friction for pitch decks published as web pages, Notion docs, or landing pages. Timeline: 1-2 weeks.
The chat-to-refine pattern (seen in pre-submission coaching tools in the competitive scan) works because it is decoupled from judgment. If the coach and the judge were the same conversation, users would learn to argue with the score instead of improving the proposal. Keeping the coach free-standing and pre-submission preserves the brutal-honesty brand of the actual review while still lowering the failure rate of first-time submissions.
Chat-to-Refine is offered as a free, unlimited-use funnel step available to every tier including Free -- it costs a fraction of a full review and its entire purpose is to produce more, and better, paid reviews downstream. URL-to-Review is available at every paid tier and functions as an alternate intake path alongside file upload, not a separate product.
Three capabilities make the verdict itself more legible: two new scores instead of one, and a plain-language explanation behind every dimension score.
Every review now produces two numbers, not one: an Idea Score (0-100, is the underlying concept sound) and a Proposal Score (0-100, how well was it argued). A brilliant idea poorly pitched and a weak idea beautifully argued now produce visibly different signals instead of collapsing into one composite number.
Every per-dimension score ships with a specific, evidence-backed explanation. Not "Market Analysis: 4/10" -- "Market Analysis: 4/10 -- no TAM calculation, no competitor pricing data, assumes zero competition." Direct application of the transparency pattern seen across the strongest-retaining products in the scan.
This is the feature most likely to be described by users, unprompted, as "the thing nobody else does." Single-score critique tools conflate concept quality with execution quality, which means a founder with a great idea and a mediocre deck gets the same verdict as a founder with a mediocre idea and a great deck. VerdictTank v3.6 tells them apart. It also changes what the corpus can say: "your Idea Score is in the 80th percentile but your Proposal Score is in the 30th" is a sharper, more actionable comparison than a single blended percentile.
The existing 10-dimension rubric splits cleanly: dimensions like market size, competitive moat, and unit economics feed the Idea Score; dimensions like clarity of argument, evidence quality, and internal consistency feed the Proposal Score. The corpus database (v3, Section 3) stores both scores per review from day one rather than retrofitting later -- this is the one place where a v3.6 feature changes a v3 schema before it ships. The per-dimension explanation text (item 4) is generated as a required field on every dimension score, not an optional add-on, and is what the sanitization gate (v3 Infrastructure Fix #10) must additionally scan before anything reaches a public report or API response, since explanation text is exactly the kind of freeform output most likely to leak internal architecture detail if left unchecked.
Dual scoring (Section 6) ships as a reversible overlay, not a schema commitment, and is explicitly gated behind the 25-user beta survey defined in Section 02a. This section states what happens on the branch where that gate is not cleared — if fewer than 65% of surveyed beta users prefer two scores over the legacy single composite, dual scoring does not ship as designed, and the differentiator falls back to a different, already-built asset.
If dual scoring is rejected, the primary differentiator becomes the Fix-It Layer combined with a prediction-verified track record. Every Fix-It session already produces a before/after delta that feeds the corpus (Section 07). Over time that corpus builds a database of what actually improved a score on re-review, not just what the panel predicted would improve it. That is a data moat — competitors can copy a feature list, but they cannot fabricate a corpus of real before/after outcomes overnight.
This is a feature swap, not a project reset. MVP scope, timeline (Section 12), pricing (Section 11), and team stay exactly as committed. Weeks 7-9 of the roadmap, freed by dropping the dual-score evaluator build, are reallocated per Section 08a rather than left idle.
After the verdict, the system generates a structured action plan tied to the low-scoring dimensions: "To improve your Market Analysis score: (1) calculate TAM using this formula, (2) add a competitor pricing table (template provided), (3) cite 3 industry reports." The submitter can then revise and re-submit.
Every AI critique product on the market, VerdictTank v2 and v3 included, is structurally a one-shot transaction: pay, get scored, done. The Fix-It Layer converts that into a loop. The competitive pattern here (seen in monitoring-plus-remediation SaaS in the scan) inverts the usual model where monitoring/detection is free and fixing is the paid upsell -- VerdictTank flips it deliberately: reviewing is paid, fixing is included. That is the differentiator, and it is the reason a Pro or Enterprise subscriber comes back inside the same billing month instead of churning after one review.
Combined with dual scoring (Section 6) and re-review, this also produces the cleanest possible proof point for reviewer accuracy: a user who follows the fix-it plan and re-submits generates a natural before/after score delta that the corpus can track and eventually surface as a product stat ("proposals that follow the Fix-It plan improve their Proposal Score by a median of X points on re-review").
Available on Pro and above. Re-review after applying a Fix-It plan is billed as a normal review against the plan's monthly allotment (or overage rate) -- it is not a free re-run, since the AI cost of a full pipeline pass is the same either way. What is "included" is the remediation plan itself, not unlimited re-scoring. Free tier gets a lightweight, single-paragraph fix-it summary as a taste of the feature; the full structured, dimension-by-dimension action plan with templates is a paid-tier feature.
Priority Fix #7 (Fix-It Quality Loophole), Judge 3 "What Both Missed" finding #1: the original v3.5 design tracked "re-review score deltas" to distinguish superficial template-filling from genuine improvement, but never actually specified an evaluation step that did the distinguishing. Checking whether a field got filled in is not the same as checking whether the gap it was meant to close is actually closed. A user who pastes one sentence into a TAM field to satisfy a presence-check would have scored identically to a user who did the work, undermining the entire fix-it-loop value proposition this section exists to deliver.
Every resubmission is now evaluated against a 3-tier quality rubric per action item, not a binary presence check:
| Tier | Score | What it means |
|---|---|---|
| Superficial | 1 | The section changed, but the specific gap named in the fix-it plan is still unaddressed, for example a TAM number appears with no methodology, source, or calculation. Text presence alone does not clear this tier. |
| Minimal | 2 | The gap is nominally addressed but shallow, for example a TAM figure with a one-line note but no breakdown, citation, or sensitivity range. Passes a "did they try" bar but not a "would this survive scrutiny" bar. |
| Substantive | 3 | The gap is closed with the rigor the action item called for, for example a TAM calculated via the recommended method, source-cited, with stated methodology. Would plausibly change an informed reader's assessment. |
The mean tier score across all action items with a prior plan becomes fix_completion_score (0-3), surfaced back to the user separately from the re-scored idea_score/proposal_score so a user can see both "did your score go up" and "did you actually fix what we flagged." This is a defined, auditable computation, not the vague "score deltas" framing this section carried in the prior revision. Full evaluator specification lives in the architecture companion doc (Section 5.5); this proposal states the product-facing behavior only.
Two capabilities target Enterprise and White-Label customers who need the panel's judgment checked and its criteria tailored to their own standards.
After the primary multi-judge review completes, a separate audit agent reviews the review itself -- checking for blind spots, groupthink, or dimensions the panel underweighted. Applies an audit-loop pattern seen in adjacent review-automation tooling directly to proposal review. Timeline: 6-8 weeks.
Enterprise and White-Label customers define custom review rules through a no-code builder: industry-specific compliance checks, internal evaluation criteria, brand voice guidelines. The panel's standard rubric still runs; custom rules layer on top rather than replacing it. Timeline: 3-4 weeks.
Both features exist to answer the same institutional-buyer objection: "how do I trust an AI panel's judgment enough to put my firm's name behind it, or bend it to our own criteria?" The audit agent answers the trust question with a second independent check baked into the pipeline. The rules engine answers the customization question without opening the door to arbitrary prompt injection into the panel -- custom rules are additive checks evaluated alongside the fixed rubric, never a replacement for it, which keeps VerdictTank's brutal-honesty core intact even when White-Label and Enterprise customers bring their own criteria.
The audit agent's own findings are subject to the same reviewer-accuracy tracking (v3, Tier 1) as the primary panel -- an audit agent that never finds anything, or that only echoes the primary panel, is itself a signal the accuracy scoring system should catch.
Section 02a defines three specific numeric gates the beta survey must clear. This section states, in advance and in specific terms, what happens on each failure branch — so that a miss is a pre-planned pivot, not an improvised one.
| Failure Mode | Trigger | Pivot Action |
|---|---|---|
| 1. Dual scoring rejected | Fewer than 65% of surveyed beta users prefer two scores over one | Fallback differentiator activates (Section 06a). Dual-score overlay dropped from Phase 1. Timeline shortens by ~2 weeks (Section 12 weeks 7-9 are freed). No GTM impact — Section 14a channels are unaffected. |
| 2. Free→Pro conversion below 2% | Actual conversion measured below the Validated Floor's 2% assumption (Section 14b) | Pro tier price drops to $49/mo (from $79) until conversion recovers to 3%. A "Founder Plan" at $29/mo is added specifically for beta-to-Pro migrators, as a retention bridge rather than a permanent tier. The Validated Floor in Section 14b already models at 2% conversion — this failure mode is survivable by construction, not by improvisation. |
| 3. Monthly churn above 10% | Actual churn measured above the Validated Floor's worst-case 10% assumption (Section 14b) | Free tier gate added: after 5 free reviews, a satisfaction survey must be completed to unlock further reviews. Pro tier adds a "review streak" discount — 3+ consecutive paid months earns 15% off. Both are cheap builds (a survey endpoint and a coupon code), chosen deliberately because a churn crisis is not the moment to fund an expensive retention feature. |
The v3 White-Label tier (Section 11) already covers accelerators, VCs, and consultancies at the pricing level. v3.6 makes it a dedicated product track rather than a checkbox on the Enterprise tier: custom domains, custom logos, branded email templates, and a consultant-facing admin console for managing client portfolios. Consulting firms and accelerators can now offer VerdictTank as their own branded review service end to end. Timeline: 2-3 weeks.
An optional, togglable community layer: peers can leave "roast" (critical) or "boost" (supportive) signals on a proposal. This is a supplementary human signal, not a replacement for the AI panel, and is off by default -- the submitter opts in per-proposal. Timeline: 3-4 weeks.
Item 8 does not create a new pricing tier; it defines what "White-Label" already means in v3 pricing more precisely, with concrete deliverables (custom domain, custom branding, portfolio console) instead of a general promise. Item 8 is the reason the White-Label tier's feature list is expanded in Section 11 without changing its price.
Item 9 is intentionally P3 and opt-in. Community layers carry real risk (brigading, low-signal noise, moderation overhead) that does not belong anywhere near the core judgment product by default. It ships last, behind a feature flag, and is evaluated after one full quarter of opt-in data before any decision is made about turning it on by default for any tier.
Priority Fix #8, Judge 3 "What Both Missed" finding #2: the White-Label track, as scoped above, assumes consultants and accelerators want to resell VerdictTank. The review points out a structural problem this proposal did not previously name: VerdictTank's core product, an automated review plus a remediation plan, directly competes with the billable review and advisory services those same consultants sell today. A consultant asked to White-Label VerdictTank is being asked to plug in a tool that can substitute for the hours they currently invoice. Framed only as a distribution channel, White-Label risks being read by the very partners it targets as a cannibalizing competitor wearing their logo, not a resale opportunity. This is a positioning problem, not a pricing problem, and pricing adjustments alone do not fix it.
The mitigation this proposal adopts is a repositioning of what White-Label sells. The consultant does not resell "an AI that does your job." The consultant sells the judgment of what the AI verdict means for this specific client, in this specific situation, delivered by someone the client already trusts. VerdictTank produces the score, the explanation, and the fix-it plan; the consultant's paid value shifts to interpreting that output, prioritizing which fixes matter most for this client's specific fundraising or procurement context, and standing behind the recommendation with their own professional relationship and reputation. The AI becomes the consultant's leverage, doing the first-pass critique work at near-zero marginal cost, freeing the consultant's billable hours for the judgment-and-relationship work that a tool cannot replace. This reframes White-Label from "give consultants a competitor" to "give consultants a research assistant that makes their existing billable hours go further."
The mitigation above is a positioning hypothesis, not a confirmed answer. Per the cut list in Section 12, the White-Label track is deferred out of the 12-week MVP specifically because this channel-conflict question, together with the Minimum Viable Legal blocker in Section 17, has not been tested with actual consultants. Before the White-Label track resumes at any point post-MVP, the "trusted interpreter" positioning must be validated directly with a small number of target consultants, structured as a straightforward question: does a consultant see this as something that replaces their billable review work, or something that makes their existing billable relationship more valuable? If the honest answer trends toward "replaces," the White-Label track needs a different structure (for example, revenue share tied to the consultant's own advisory fee, rather than a flat reseller license) before it launches, not after.
The v3 panel expansion from 4 to 7 reviewers (Reasoning-Verification, Execution-Feasibility, Market-Reality judges) is unchanged and carried forward in full -- see the v3 proposal for the complete rationale and the deferred-candidate table. v3.6 adds one new role to the panel's output contract rather than its roster:
No new judge is added for dual scoring (Section 6). Instead, each existing judge's dimension scores are tagged at generation time as contributing to the Idea Score or the Proposal Score, and the Second-Opinion Audit Agent (Section 8) additionally checks that this tagging is applied consistently across judges before a verdict is finalized. This keeps the panel at 7 reviewers plus the audit agent, rather than growing it further, consistent with the v3 condition that panel size is a tunable parameter, not a one-way ratchet.
Per house sanitization policy, exact vendor and model identities remain internal and are stripped from all public-facing documentation and API responses by the automated sanitization gate, exactly as under v3.
v3.7 narrows the priced tier structure to three: Free, Pro, and Enterprise. White-Label for Consultants is removed from the pricing table entirely — not repriced, not bundled, removed — until row-level security and org-level tenant isolation are actually built and verified (see Section 08 and Section 09). Publishing a price against an isolation model that does not exist yet is the kind of claim this revision exists to stop making. The current verdicttank.com beta pricing ($29/mo "Inner Circle") remains a temporary bridge that sunsets at v3.7 General Availability, grandfathered per Section 12.
Custom domains, branded reports, a portfolio console for managing client proposals, and dedicated tenant isolation are all still on the roadmap. They are not for sale yet. No White-Label price is published anywhere, in this proposal or on verdicttank.com, until row-level security (RLS) and org-level tenant isolation are built and independently verified — publishing a number against infrastructure that does not exist is exactly the kind of premature commitment this revision is correcting for. Consultants and accelerators interested in this track can join the waitlist below; joining implies no commitment and no pricing is disclosed at signup.
Pay-per-review outside a subscription: $19.99, full pipeline, one-time. Annual billing on Pro and Enterprise: 20% discount. All tiers retain the degraded-mode fallback from v2/v3 — if a judge is unavailable, the pipeline runs with fewer judges rather than failing.
| Weeks | Phase | Scope | Concurrency |
|---|---|---|---|
| 0-3 | Beta User Validation Survey | Close the Inner Circle beta survey (Section 02a), analyze NPS/feature-request/churn/usage results, apply the 65% dual-scoring gate and the 25-user threshold before any schema commitment. | Sequential — gates everything after it |
| 4-6 | Schema & Core Pipeline | Database schema finalization, corpus structure, core review pipeline build-out. | Concurrent with survey analysis close-out (weeks 3-4 overlap) |
| 7-9 | Dual Scoring System | Second-dimension evaluator, Idea Score / Proposal Score split, side-by-side comparison view. | Contingent — only proceeds if the Week 0-3 survey clears the 65% preference gate (Section 02a). If it does not clear, see Section 06a. |
| 8-11 | Fix-It Layer | Single-tier binary evaluator (not yet the full 3-tier rubric), before/after delta comparison, re-score action wired to billing. | Concurrent with Dual Scoring, weeks 8-9 overlap |
| 10-12 | Chat-to-Refine & Polish | Coach-mode chat (structuring questions only, no ghostwriting), URL-to-Review extraction, end-to-end integration testing. | Sequential — depends on Fix-It and Dual Scoring being feature-complete |
Read as a Gantt in prose: weeks 0-3 are a hard gate no other work can jump ahead of; weeks 4-6 (schema) start once survey analysis is far enough along to lock the dual-score column decision, not once every survey response is in; weeks 7-9 and 8-11 genuinely overlap because Dual Scoring and Fix-It touch different parts of the pipeline; weeks 10-12 close out sequentially because integration testing cannot start until both feature branches are stable.
| Category | Players | What They Do | VerdictTank v3.6 Difference |
|---|---|---|---|
| AI proposal writers | Template-based generation tools | Generate proposals from prompts | We critique them; we still don't write them -- but Chat-to-Refine now helps a user structure their own argument before submission, closing the gap without becoming a generation tool ourselves. |
| Pre-submission coaching tools | AI chat-based drafting assistants | Help structure an argument before submission, no independent judgment afterward | Chat-to-Refine matches this pattern for intake, then hands off to an independent, brutally honest multi-judge panel these tools don't have. |
| Single-score idea-validation platforms | Community or single-model scoring tools | One blended score, idea quality and pitch quality conflated | Dual Scoring (Idea Score + Proposal Score) separates the two signals -- the single differentiator most competitors in this category lack entirely. |
| Human consultants | Independent reviewers, boutique firms | Manual review, $300-800/hr, days of turnaround | Under $1 and minutes, now with a Fix-It plan included -- consultants charge extra for that follow-up work; we do not. |
| Monitoring/audit SaaS with paid remediation | Detect-then-upsell-the-fix platforms | Monitoring or detection free, remediation is the paid tier | VerdictTank inverts this: reviewing is paid, fixing is included. Removes the friction point that causes users to detect a problem and never pay to fix it. |
| Accelerator/VC diligence tools | Internal scoring rubrics, deal-flow CRMs | Human-scored, not benchmarked against a broader corpus | White-Label for Consultants gives accelerators a benchmarked, auditable, corpus-backed second opinion with their own branding, domain, and portfolio console -- not just a shared login. |
| Cumulative-memory, full-journey proposal review | No direct competitor exists as of August 2026. A platform that remembers its own track record, tells two scores instead of one, explains every dimension in plain language, and hands the user a fix-it plan is a genuinely new category composition, not an incremental feature. | We are still the category, now with the full loop: coach, judge, explain, fix, re-score. |
No analyst category exists for "AI business proposal review." Every prior version of this proposal implied comparability to adjacent categories without stating the comparison, or the caveat, explicitly. This section builds a bottom-up TAM/SAM/SOM from named user segments rather than borrowing a market-size figure from a neighboring category.
| Layer | Size | Basis |
|---|---|---|
| TAM | $1.12B | 350K fundraising startups worldwide × $400/yr, plus 1.5M SMB/RFP writers × $500/yr, plus 15K consultants/VCs × $1,200/yr. |
| SAM | $168M | 15% of TAM — English-language, digital-native, actively seeking funding or actively writing proposals right now (not a dormant/aspirational subset of the TAM population). |
| SOM | $1.47M | 0.9% of SAM — 2,450 paying users at Year 3, reality-checked directly against the Year 3 paying-user figure in Section 14b rather than derived independently of it. |
No analyst tracks "AI business proposal review" as a category. The closest proxies are proposal management SaaS (Qvidian, Loopio), growing at a 10-12% CAGR, and general-purpose GenAI tooling, growing at 30%+ CAGR. Both are cited here as directional signal about market appetite for AI-assisted proposal work generally — they are not the same category as VerdictTank, and their growth rates are not applied to the TAM/SAM/SOM figures above. The bottom-up build above stands on its own segment counts and per-user pricing, not on a borrowed category multiple.
Three organic channels carry the majority of acquisition. Paid ads are an amplifier on top of them, not the primary engine, and are capped deliberately so the financial model in Section 14b is not quietly dependent on an acquisition cost nobody has tested.
| Channel | Mechanism | CAC Range |
|---|---|---|
| 1. Product-Led Growth | Free tier usage drives word of mouth; word of mouth drives Pro conversion. No ad spend required for this loop to function. | $8-25 |
| 2. SEO / Content Marketing | Proposal-writing guides, startup fundraising content, indexed and organic. | $15-40 |
| 3. Partner / Consultant Referrals | Accelerators and pitch coaches recommending VerdictTank to their own client base. | $5-15 |
| Paid ads (amplifier only) | Retargeting against warm traffic, capped at ~20% of new signups. Not used for cold acquisition. | $60-120 |
Target mix: 80% organic (channels 1-3), 20% paid. Paid CAC is 3-8x organic CAC — the cap exists specifically so paid spend cannot become the default lever the moment organic growth slows, which is the failure pattern this section is designed to rule out in advance.
| Stage | Conversion |
|---|---|
| Website visitor → Signup | 8% |
| Signup → Activated (completes first review) | 45% |
| Activated → Free active user | 70% |
| Free → Pro (paid) | 3% |
195,000 visitors/yr × 8% signup × 45% activation × 70% free-active × 3% conversion ≈ 210 Year-1 paying users. This is not a separate estimate presented alongside the financial model — it is the same 210-payer figure that anchors the Validated Floor in Section 14b. A GTM section that produces a different user count than the financial model would be a proposal quietly running two sets of books; this one does not.
| Cost Component | Per Review |
|---|---|
| Base pipeline (research + primary + 3 cross-checks) | $0.47 |
| 3 new specialist reviewer additions (v3, Section 10) | $0.21 |
| Market simulation engine | $0.06 |
| Vertical red-team pass | $0.05 |
| Corpus write, embedding, and similarity search (dual-score schema) | $0.02 |
| Prediction tracking cron (amortized) | $0.01 |
| Infrastructure (sanitization gate, PDF automation, storage) | $0.04 |
| New: Dual scoring + per-dimension explanation generation | $0.04 |
| New: Fix-It action plan generation (Pro+ only) | $0.05 |
| New: Second-Opinion Audit Agent (Enterprise only, amortized across all reviews) | $0.03 |
| Fully loaded cost per full v3.6 review | $0.98 |
Chat-to-Refine and URL-to-Review add negligible marginal cost (~$0.02-0.03 per session) and are treated as funnel/acquisition cost, not review COGS, consistent with the free-tier loss-leader model already established in v3. Free-tier full reviews stay near $0.07/review; the lightweight fix-it summary on Free adds under a cent.
| Tier | Included Reviews/mo | Blended Revenue/Review | Cost/Review | Gross Margin |
|---|---|---|---|---|
| Pro ($79/mo) | 20 | $3.95 | $0.98 | 75% |
| Enterprise ($299/mo) | 100 | $4.99 | ~$1.20 (incl. corpus/API/audit-agent infra) | 76% |
| Year 1 | Year 2 | Year 3 | |
|---|---|---|---|
| Paying users (Free excluded) | 210 | 780 | 2,450 |
| White-label / consultant-track accounts | 5 | 18 | 48 |
| MRR (year-end) | $13,100 | $54,000 | $176,000 |
| Annual revenue | $88,000 | $412,000 | $1,470,000 |
| AI + infra costs | $15,200 | $59,000 | $188,000 |
| Net | $72,800 | $353,000 | $1,282,000 |
Uplift over v3's original forecast ($52.2K / $277K / $1.098M net) comes primarily from three effects: Chat-to-Refine and URL-to-Review lowering intake friction and increasing free-to-paid conversion; the Fix-It Layer converting one-shot reviews into repeat-use within the same billing month, raising blended reviews-per-paying-user; and the dedicated White-Label-for-Consultants track making that tier easier to sell against accelerator-specific budget lines than a generic enterprise pitch. Assumptions carried forward from v3: AI cost deflation of 15-20% annually, 5-7% monthly Pro churn, materially lower Enterprise/White-Label churn given contractual terms. The beta-to-Pro migration (Section 11) is modeled as net-neutral to Year 1 revenue -- beta accounts are assumed to already be counted in the Year 1 paying-user base at their eventual Pro-equivalent value once the grandfather window closes.
The base case above assumes 5-7% monthly Pro churn, in line with v3. Per the critical review, that assumption is untested against actual VerdictTank usage and deserves a stated downside case. The row below shows what happens if monthly Pro churn runs above 30%, a level consistent with a product that has not found real retention traction rather than one experiencing normal SaaS attrition.
| Scenario | Monthly Pro Churn | Year 1 Net | Year 2 Net | Year 3 Net |
|---|---|---|---|---|
| Base case (modeled above) | 5-7% | $72,800 | $353,000 | $1,282,000 |
| Downside: churn >30%/mo | >30% | ~$8,000-15,000 | Roughly flat to negative once fixed infra and support costs are counted | Product does not reach the compounding growth the base case assumes; Pro tier functions as a leaky bucket rather than a subscription base |
At sustained >30% monthly churn, the Pro tier's blended-revenue-per-review assumption in the margin table above no longer holds, because the "20 reviews/month included, converts to a subscription relationship" logic that makes Pro attractive to model breaks down when most subscribers cancel before using more than one or two reviews. This scenario is the reason the Beta User Validation Plan (Section 02a) treats churn-reason interviews as a first-class input rather than an afterthought: this proposal would rather learn the actual churn driver from 40 real beta users before committing further engineering time than discover it from a live Pro cohort after GA.
The financial model above assumes 15-20% annual AI cost deflation, consistent with the trend over the last several years. That trend is not guaranteed to continue, and the review specifically asked for the reverse case to be shown alongside it. The table below holds Year 3 revenue constant at the base-case forecast and shows how gross margin moves if per-review AI cost rises instead of falling.
| Cost Scenario | Fully-Loaded Cost/Review (Pro) | Pro Gross Margin | Fully-Loaded Cost/Review (Enterprise) | Enterprise Gross Margin |
|---|---|---|---|---|
| Base case (15-20% annual deflation) | $0.98, trending down | 75% | $1.20, trending down | 76% |
| Costs rise 20% | $1.18 | 70% | $1.44 | 71% |
| Costs rise 50% | $1.47 | 63% | $1.80 | 64% |
| Costs rise 100% | $1.96 | 50% | $2.40 | 52% |
Even at a full doubling of AI cost per review, gross margin stays positive at both the Pro and Enterprise tiers, because the current unit economics run well below the price points charged. The bigger risk a cost-rise scenario introduces is not margin collapse, it is that the free tier and Chat-to-Refine's funnel economics (currently modeled as negligible marginal cost) stop being negligible, which would force a free-tier usage cap sooner than the roadmap currently anticipates. That risk is noted here and is not currently modeled into the Year 1-3 forecast above.
The v3.6 financial model presented one base case. That single number implied a confidence this proposal has not earned, since none of the underlying conversion or churn assumptions have been checked against real VerdictTank usage. v3.7 replaces it with two bounds — a validated floor and a validated ceiling — and relabels the old base case for what it actually is.
| Year 1 | Year 2 | Year 3 | |
|---|---|---|---|
| Paying users | 210 | 620 | 1,450 |
| Blended price/mo | $29-49 | $29-49 | $29-49 |
| Free→Pro conversion | 2% | 2% | 2% |
| Monthly churn | 10% | 7% | 5% |
| Net | $34,000 | $118,000 | $305,000 |
| Year 1 | Year 2 | Year 3 | |
|---|---|---|---|
| Paying users | 210 | 780 | 2,450 |
| Blended price/mo | $49-79 | $49-79 | $49-79 |
| Free→Pro conversion | 3% | 3% | 3% |
| Monthly churn | 7% | 5% | 4% |
| Net | $61,000 | $262,000 | $820,000 |
White-Label revenue is zeroed in both the floor and the ceiling, consistent with Section 11 and Section 09 — it is not for sale, so it is not in the forecast. Every dollar in both tables above comes from Free, Pro, and Enterprise only.
The floor uses worse-case assumptions on purpose: lower blended price, lower conversion, higher churn — specifically because the beta survey (Section 02a) has not closed yet and none of these numbers are confirmed. It is the number this proposal is comfortable being held to today. The ceiling uses median assumptions from the beta cohort's early signal, but is only usable as a planning number once the beta survey actually confirms conversion and churn in that range — it is not a number to build a budget against until that confirmation lands.
The v3.6 base case ($72.8K / $353K / $1.282M net across Years 1-3) is retained below for continuity, relabeled honestly: aspirational (includes deferred White-Label revenue that is no longer in the pricing table). It should not be read alongside the floor and ceiling as if it were a third, more-likely scenario — it is the old model, kept visible so the revision is auditable, not the current forecast.
| Scenario | Year 1 Net | Year 2 Net | Year 3 Net | Status |
|---|---|---|---|---|
| Validated Floor | $34,000 | $118,000 | $305,000 | Current commitment |
| Validated Ceiling | $61,000 | $262,000 | $820,000 | Pending beta confirmation |
| v3.6 base case (superseded) | $72,800 | $353,000 | $1,282,000 | Aspirational, includes deferred White-Label |
The eight risks identified in v3 remain in force unchanged and are carried forward in full (corpus confidentiality, accuracy-scoring convergence, prediction attribution difficulty, sanitization gate failure, API abuse, panel cost creep, competitive-comparison leakage, white-label brand risk, training-data recursion). v3.6 adds five new risks introduced by the nine new capabilities.
| Risk | Likelihood | Impact | Mitigation |
|---|---|---|---|
| Chat-to-Refine coaching drifts into ghostwriting the proposal for the user, undermining the "we critique, we don't write" positioning | Medium | High | Coach is scoped to structuring questions and gap-identification prompts only, never full-paragraph generation. Coach transcripts are logged and periodically spot-checked for drift toward ghostwriting behavior. |
| Dual scoring is perceived as a gimmick or confuses users if the two scores are not clearly differentiated in the UI/report | Medium | Medium | Every report leads with a two-line plain-language framing ("your idea is strong, your pitch needs work" style) before showing the numeric breakdown. User-test the framing before Phase 2 exit. |
| Per-dimension "Explain the Low Score" freeform text becomes the highest-risk sanitization surface -- more opportunity for an internal detail to leak than a fixed-template report ever had | Medium | High | Sanitization gate (v3 Infrastructure Fix #10) explicitly extended in Phase 0 to scan all freeform explanation and remediation text, not just fixed report sections. Manual spot check on the first 200 explanation outputs before Phase 2 GA, not just the first 50 API responses as in v3. |
| Fix-It remediation plans give away enough of the pipeline's evaluation logic that a sophisticated user reverse-engineers what the panel is checking for, gaming future submissions rather than genuinely improving them | Medium | Medium | Remediation plans stay tied to what is missing (data, evidence, structure), not to how the panel weighs it. Track re-review score deltas for a pattern of superficial compliance (template-filling without substance) versus genuine improvement, and feed that signal back into reviewer accuracy scoring. |
| Roast/Boost community layer attracts low-signal noise, brigading, or becomes a vector for harassment tied to a submitter's real proposal | Medium | Medium | Opt-in only, off by default, no default visibility to anyone outside the submitter's own dashboard unless the submitter explicitly shares it. Ships behind a feature flag in Phase 4 and is evaluated after one full quarter of opt-in data before any default-on consideration. |
v3.6 is an integration, not a rewrite -- it takes the v3 memory foundation as given and wraps it in a full product journey (coach, dual score, transparency, fix-it, re-score) that the competitive scan shows nobody else in this category has assembled end to end. The two strategic themes it centers, transparency and the fix-it loop, are the two changes most likely to move retention and repeat-use, which is where v3 alone was weakest. All four v3 conditions carry forward unchanged; three new conditions are added specific to the new surfaces.
This proposal will be submitted to the VerdictTank pipeline for its own review before implementation begins, consistent with house policy that every IT Pro Partner product proposal is reviewed by the tool it is proposing to improve.
Priority Fix #5, Legal Killers #1-5 identified in the critical review: the v3.5 proposal shipped with zero disclaimer language anywhere in the product, no consultant-facing liability allocation, and no defined pipeline for what happens to submitted content across jurisdictions. This section summarizes the Minimum Viable Legal (MVL) framework built in direct response to that finding. The full specification, including exact contractual language templates for counsel, data handling requirements, and acceptance criteria for each workstream, lives in the companion document and is not duplicated here.
Every verdict, explanation, audit finding, and remediation plan the platform generates must carry the following disclaimer, verbatim or as approved by counsel, rendered as a stored data attribute at generation time rather than a page-footer afterthought:
AI-Generated Content Disclaimer
This content, including any score, critique, verdict, audit finding, or remediation plan, is an automated output generated by artificial intelligence models. It is not professional advice of any kind, including but not limited to legal, financial, investment, accounting, tax, procurement, or business advice. The opinions, scores, and recommendations expressed are opinions generated by AI models based on pattern analysis of the submitted material, and do not reflect the professional judgment, endorsement, or certification of VerdictTank, its officers, employees, or any White-Label partner.
Do not rely on this content for investment, procurement, funding, hiring, financing, legal, or comparable high-stakes decisions. AI-generated outputs may be incomplete, inaccurate, outdated, or wrong. You are solely responsible for independently verifying any information before acting on it and for obtaining qualified professional advice where appropriate.
To the maximum extent permitted by applicable law, VerdictTank's total liability arising from or related to this content, and any decision made in reliance on it, is limited to the amount paid by you for the applicable service in the twelve (12) months preceding the claim, or one hundred US dollars (USD 100), whichever is greater. VerdictTank disclaims all warranties, express or implied, regarding the accuracy, completeness, or fitness for a particular purpose of this content.
Short form, for space-constrained surfaces such as chat bubbles, mobile cards, and API metadata fields: "AI-generated opinion. Not professional advice. Do not rely on for investment, procurement, or funding decisions. Liability limited per Terms of Service."
The companion MVL document specifies five workstreams, each tied to a specific Legal Killer surfaced in the critical review: the AI liability disclaimer above (Legal Killer #2), consultant and White-Label MSA liability allocation and indemnity terms (supporting the channel-conflict mitigation in Section 09), data handling and multi-jurisdiction confidentiality controls for submitted proposal content, a defined content-moderation and takedown pipeline for any user-facing or shareable output, and a versioned audit trail so historical outputs can be reconstructed against whatever disclaimer and terms were live at the time they were generated. None of these five workstreams is treated as optional or deferrable; per the MVL document, they are existence conditions for multi-tenant, multi-jurisdiction operation, not features to schedule around.
The full specification, including required MSA checklist items for counsel, disclaimer versioning requirements, and per-workstream acceptance criteria, is maintained at docs.itpropartner.com/verdicttank/legal-framework.md and must be reviewed by outside counsel before that document is published to any customer or partner, and before White-Label Pilot #1 is allowed to onboard.