ippadmin/verdicttank
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

v4.1 go-live cut, Moonshot to Mistral swap, v5.x superseded

Browse Source 
Mark v5.0/v5.1 SUPERSEDED (error-detection thesis failed at -0.40 delta); v4.1 is canonical. Cut v4.1 proposal with 12 version strings bumped. Moonshot to Mistral across production seats; production worker de-kimi'd 2026-08-18. Data-retention posture corrected 7/9 to 8/9 no-training (DeepSeek sole exception). Reconciled COGS with measured Mistral spend. Committed deployed v4.0 content and research docs to resolve the repo/live fork.
This commit is contained in:
root
2026-08-18 20:28:36 -04:00
parent c1e690a122
commit 3c07727f5c
18 changed files with 3409 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
parts/p1.html
+239
View File
@@ -0,0 +1,239 @@
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>VerdictTank v5.0 · Error Detection Density for Proposals</title>
<style>
:root{
--bg:#0b0e14;--panel:#131824;--panel2:#1a2030;--ink:#e6edf3;--muted:#8b97a8;
--line:#232b3a;--accent:#4da3ff;--accent2:#7ee787;--warn:#f0a45c;--bad:#ff6b6b;
--code:#0d1117;--codeink:#c9d1d9;--pass:#2ea043;--below:#d29922;
}
*{box-sizing:border-box}
html{scroll-behavior:smooth}
body{margin:0;background:var(--bg);color:var(--ink);
font:15px/1.65 -apple-system,BlinkMacSystemFont,"Segoe UI",Roboto,Helvetica,Arial,sans-serif}
.wrap{max-width:1180px;margin:0 auto;padding:0 24px}
header.hero{background:linear-gradient(135deg,#0b0e14,#131824 60%,#0f1a2e);
border-bottom:1px solid var(--line);padding:48px 0 36px}
.badge{display:inline-block;font-size:12px;letter-spacing:.14em;text-transform:uppercase;
color:var(--accent);border:1px solid var(--accent);border-radius:999px;padding:4px 12px;margin-bottom:16px}
.badge--warn{color:var(--warn);border-color:var(--warn)}
h1{font-size:34px;margin:6px 0 8px;letter-spacing:-.5px}
.sub{color:var(--muted);font-size:16px;max-width:760px}
.meta{margin-top:18px;display:flex;flex-wrap:wrap;gap:10px}
.meta span{font-size:12px;background:var(--panel2);border:1px solid var(--line);
border-radius:6px;padding:5px 10px;color:var(--muted)}
.meta b{color:var(--ink)}
nav.toc{position:sticky;top:0;z-index:20;background:rgba(11,14,20,.92);
backdrop-filter:blur(8px);border-bottom:1px solid var(--line);padding:10px 0}
nav.toc ol{list-style:none;margin:0;padding:0;display:flex;flex-wrap:wrap;gap:6px;font-size:12.5px}
nav.toc a{color:var(--muted);text-decoration:none;padding:4px 9px;border-radius:6px;border:1px solid transparent}
nav.toc a:hover{color:var(--ink);border-color:var(--line);background:var(--panel)}
section{padding:40px 0;border-bottom:1px solid var(--line)}
h2{font-size:24px;margin:0 0 6px;letter-spacing:-.3px}
h2 .n{color:var(--accent);font-variant-numeric:tabular-nums;margin-right:10px}
h3{font-size:17px;margin:26px 0 8px;color:var(--accent2)}
h4{font-size:14px;margin:18px 0 6px;color:var(--warn);text-transform:uppercase;letter-spacing:.06em}
p{margin:10px 0}
.lead{color:var(--muted);font-size:15.5px;margin-bottom:8px}
ul,ol{margin:10px 0;padding-left:22px}
li{margin:5px 0}
code{background:var(--code);color:var(--codeink);padding:2px 6px;border-radius:4px;
font:13px/1.5 "SF Mono",ui-monospace,Menlo,Consolas,monospace}
pre{background:var(--code);color:var(--codeink);border:1px solid var(--line);border-radius:10px;
padding:16px;overflow:auto;font:12.5px/1.55 "SF Mono",ui-monospace,Menlo,Consolas,monospace;margin:14px 0}
pre.ascii{line-height:1.35;font-size:12px;color:#9fb3c8}
table{width:100%;border-collapse:collapse;margin:14px 0;font-size:13.5px}
th,td{border:1px solid var(--line);padding:8px 10px;text-align:left;vertical-align:top}
th{background:var(--panel2);color:var(--ink);font-weight:600}
td{background:var(--panel);color:var(--ink)}
/* Custom blocks */
.evolution-grid{display:grid;grid-template-columns:1fr 1fr;gap:20px;margin:24px 0}
.evolution-card{background:var(--panel);border:1px solid var(--line);border-radius:10px;padding:20px}
.evolution-card--current{border-color:var(--accent);background:var(--panel2)}
.version-badge{font-size:11px;letter-spacing:.1em;text-transform:uppercase;color:var(--warn);margin-bottom:8px}
.version-badge--current{color:var(--accent)}
.breakthrough-label{font-size:12px;color:var(--accent2);font-weight:600;margin-bottom:6px}
.callout{background:var(--panel2);border:1px solid var(--line);border-radius:10px;padding:16px 20px;margin:20px 0}
.callout--proof{border-left:4px solid var(--accent)}
.callout--fail{border-left:4px solid var(--bad)}
.callout--win{border-left:4px solid var(--accent2)}
.defense-grid{display:grid;grid-template-columns:1fr 1fr;gap:18px;margin:20px 0}
.defense-item{background:var(--panel);border:1px solid var(--line);border-radius:10px;padding:16px}
.positioning-statement{font-size:16px;color:var(--accent2);text-align:center;margin:28px 0;padding:16px;background:var(--panel2);border:1px solid var(--accent);border-radius:10px}
.comp-table__us td{background:#0f1a2e;color:var(--accent);font-weight:600}
.muted{color:var(--muted);font-size:12px}
.table-note{font-size:12px;color:var(--muted);margin:4px 0 14px}
.strategy-cols{display:grid;grid-template-columns:1fr 1fr;gap:20px;margin:16px 0}
.strategy-cols h4{text-transform:none;color:var(--accent);font-size:14px}
.pricing-grid{display:grid;grid-template-columns:1fr 1fr 1fr;gap:16px;margin:20px 0}
.price-card{background:var(--panel);border:1px solid var(--line);border-radius:10px;padding:20px;text-align:center}
.price-card--featured{border-color:var(--accent);background:var(--panel2);position:relative}
.ribbon{position:absolute;top:-10px;left:50%;transform:translateX(-50%);
background:var(--accent);color:#0b0e14;font-size:11px;font-weight:700;
padding:3px 12px;border-radius:999px;text-transform:uppercase;letter-spacing:.06em}
.price{font-size:36px;font-weight:700;margin:8px 0}
.price span{font-size:16px;color:var(--muted);font-weight:400}
.price-card ul{list-style:none;padding:0;text-align:left;margin:12px 0}
.price-card li{margin:6px 0;font-size:13px}
.price-card li::before{content:"✓ ";color:var(--accent2);font-weight:700}
.price-purpose{font-size:11px;color:var(--muted);text-transform:uppercase;letter-spacing:.04em;margin-top:12px}
.walkthrough{display:grid;gap:16px;margin:16px 0}
.wt-input{background:var(--panel);border:1px solid var(--line);border-radius:10px;padding:16px}
.wt-proposal-text{font-style:italic;color:var(--muted);border-left:3px solid var(--warn);padding-left:14px;margin:10px 0}
.wt-context{font-size:12px;color:var(--muted);margin-top:8px}
.wt-scores{display:grid;grid-template-columns:1fr 1fr;gap:14px}
.score-block{background:var(--panel);border:1px solid var(--line);border-radius:10px;padding:16px}
.score-block--narrative{border-left:4px solid var(--accent)}
.score-block--compliance{border-left:4px solid var(--warn)}
.score-label{font-size:12px;color:var(--muted);text-transform:uppercase;letter-spacing:.06em}
.score-value{font-size:40px;font-weight:700;margin:4px 0}
.score-value span{font-size:16px;color:var(--muted);font-weight:400}
.score-block--narrative .score-value{color:var(--accent)}
.score-block--compliance .score-value{color:var(--bad)}
.fixit-list li{margin:12px 0;padding:0}
.fixit-head{display:flex;align-items:center;gap:10px;margin-bottom:4px}
.fixit-tag{font-size:11px;font-weight:700;padding:2px 8px;border-radius:4px;text-transform:uppercase;letter-spacing:.05em}
.fixit-tag--critical{background:rgba(255,107,107,.15);color:var(--bad);border:1px solid rgba(255,107,107,.3)}
.fixit-tag--high{background:rgba(240,164,92,.15);color:var(--warn);border:1px solid rgba(240,164,92,.3)}
.fixit-tag--medium{background:rgba(77,163,255,.15);color:var(--accent);border:1px solid rgba(77,163,255,.3)}
.fixit-delta{font-size:12px;color:var(--accent2);font-weight:600}
.timeline{margin:20px 0}
.timeline-phase{background:var(--panel);border:1px solid var(--line);border-radius:10px;padding:14px 18px;margin:8px 0}
.timeline-phase--risk{border-color:var(--warn);background:var(--panel2)}
.tl-weeks{font-size:11px;color:var(--accent);text-transform:uppercase;letter-spacing:.06em;margin-bottom:4px}
.tag{font-size:11px;font-weight:700;padding:2px 8px;border-radius:4px;text-transform:uppercase;letter-spacing:.05em}
.tag-green{background:rgba(46,160,67,.15);color:var(--pass);border:1px solid rgba(46,160,67,.3)}
.tag-amber{background:rgba(210,153,34,.15);color:var(--below);border:1px solid rgba(210,153,34,.3)}
.tag-red{background:rgba(255,107,107,.15);color:var(--bad);border:1px solid rgba(255,107,107,.3)}
.highlight{background:var(--panel2);border:1px solid var(--line);border-radius:10px;padding:14px 18px;margin:16px 0}
.arch-box{background:var(--code);color:var(--codeink);border:1px solid var(--line);border-radius:8px;
padding:14px 16px;margin:12px 0;font:12.5px/1.5 "SF Mono",ui-monospace,Menlo,Consolas,monospace}
.statgrid{display:grid;grid-template-columns:repeat(4,1fr);gap:14px;margin:22px 0}
.stat{background:var(--panel);border:1px solid var(--line);border-radius:10px;padding:16px;text-align:center}
.stat--hero{border-color:var(--accent2);background:var(--panel2)}
.stat .v{font-size:32px;font-weight:700;color:var(--accent);line-height:1.1}
.stat--hero .v{color:var(--accent2)}
.stat--bad .v{color:var(--bad)}
.stat .k{font-size:11px;color:var(--muted);text-transform:uppercase;letter-spacing:.06em;margin-top:6px}
.errgrid{display:grid;grid-template-columns:repeat(3,1fr);gap:14px;margin:18px 0}
.errcard{background:var(--panel);border:1px solid var(--line);border-left:4px solid var(--bad);border-radius:10px;padding:14px 16px}
.errcard .n{font-size:26px;font-weight:700;color:var(--bad)}
.errcard .t{font-size:13px;font-weight:600;margin:4px 0}
.errcard .d{font-size:12.5px;color:var(--muted)}
footer{text-align:center;padding:40px 0 30px;border-top:1px solid var(--line);margin-top:40px;
font-size:12px;color:var(--muted);line-height:1.8}
footer a{color:var(--accent)}
a{color:var(--accent)}
@media (max-width:768px){
.evolution-grid,.defense-grid,.strategy-cols,.pricing-grid,.wt-scores{grid-template-columns:1fr}
.statgrid,.errgrid{grid-template-columns:1fr 1fr}
h1{font-size:28px}
.price{font-size:28px}
.score-value{font-size:30px}
}
</style>
</head>
<body>
<main class="wrap">
<!-- ====== HERO ====== -->
<header class="hero" style="text-align:center;padding:80px 0 60px">
<div class="badge">v5.0 · Pre-Revenue · Validation-Tested</div>
<h1 style="font-size:44px;letter-spacing:-1px">14 blind errors your solo model missed</h1>
<p class="sub" style="font-size:20px;max-width:840px;margin:0 auto">
A solo frontier model gives you a smooth, confident score. An 11-judge panel gives you the
14 things it was wrong about. VerdictTank does not sell you a higher number. It sells you
the errors that number was hiding.
</p>
<div class="meta" style="justify-content:center;margin-top:24px">
<span><b>Category:</b> AI Proposal Review &amp; Error Detection</span>
<span><b>Panel:</b> 11 seats · 9 vendors</span>
<span><b>Architecture:</b> <a href="architecture.html">Full technical document</a></span>
<span><b>Judge pool:</b> <a href="judge-pool-spec.md">Spec v2.3</a></span>
</div>
</header>
<!-- ====== NAVIGATION ====== -->
<nav class="toc">
<div class="wrap">
<ol>
<li><a href="#thesis">1. The Thesis</a></li>
<li><a href="#pipeline">2. Pipeline Story</a></li>
<li><a href="#validation">3. Validation Data</a></li>
<li><a href="#errors">4. The 14 Errors</a></li>
<li><a href="#panel">5. Judge Pool v2.3</a></li>
<li><a href="#worked">6. Worked Example</a></li>
<li><a href="#pricing">7. Pricing</a></li>
<li><a href="#competitive">8. Competitive Landscape</a></li>
<li><a href="#deployment">9. Deployment</a></li>
<li><a href="#legal">10. Legal &amp; Compliance</a></li>
</ol>
</div>
</nav>
<!-- ====== 1. THESIS ====== -->
<section id="thesis">
<h2><span class="n">01</span>The Thesis Changed, Because the Data Said So</h2>
<p class="lead">
VerdictTank v4.0 was sold on a claim we could not defend: that a multimodel panel produces a
better score than a single strong model. On 2026-08-12 we ran that claim against real proposals
and it failed. What we found instead is a stronger product.
</p>
<div class="statgrid">
<div class="stat stat--hero"><div class="v">14</div><div class="k">Material errors caught</div></div>
<div class="stat stat--bad"><div class="v">-0.40</div><div class="k">Aggregate score delta</div></div>
<div class="stat"><div class="v">3</div><div class="k">Real proposals scored</div></div>
<div class="stat"><div class="v">11</div><div class="k">Judge seats, 9 vendors</div></div>
</div>
<div class="callout callout--fail">
<h4 style="margin-top:0">What failed</h4>
<p style="margin-bottom:0">
The score-elevation thesis. Across three real proposals the panel mean came in
<strong>0.40 points below</strong> the solo baseline. The panel did not lift scores. On two
of three proposals it pushed them down. We are publishing that result rather than burying it,
because the reason it happened is the product.
</p>
</div>
<div class="callout callout--win">
<h4 style="margin-top:0">What worked</h4>
<p style="margin-bottom:0">
Error detection density. The same panel run surfaced <strong>14 material errors</strong> that
the solo baseline missed or underweighted: revenue arithmetic that was wrong by a factor of
seven, a funded direct competitor the solo pass never named, a launch-blocking compliance
cost larger than projected first-year revenue. None of those show up as a score. All of them
decide whether the proposal wins.
</p>
</div>
<div class="positioning-statement">
You do not buy VerdictTank to get a higher score.<br>
You buy it to find the $60K compliance hole and the broken contractor budget before you ship.
</div>
<h3>Why the spread is the signal</h3>
<p>
A single model scoring alone produces low variance. It reads the document once, forms one
coherent opinion, and every dimension it emits is downstream of that opinion. The result feels
authoritative precisely because nothing inside it disagrees.
</p>
<p>
An 11-seat panel of nine different vendors cannot produce that coherence, and the incoherence
is diagnostic. When a Financial Integrity judge scores a proposal 8.2 while an
Execution Feasibility judge scores the same document 2.8, that 5.4-point spread is not noise.
It is a precise statement: <em>the money works, the delivery plan does not</em>. A solo model
averages that tension away into a single confident 6.1 and tells you nothing actionable.
</p>
<div class="highlight">
<strong>Measured, not asserted:</strong> on every one of the three validated proposals, panel
spread exceeded solo spread. RFP Tank 3.7 vs 1.4. VentureBuilt 5.4 vs 2.4. CartMySupply
2.6 vs 1.8. Widening variance is the intended behavior, not a defect to tune out.
</div>
</section>
parts/p2.html
+177
View File
@@ -0,0 +1,177 @@
<!-- ====== 2. PIPELINE STORY ====== -->
<section id="pipeline">
<h2><span class="n">02</span>How We Found This: Four Generations and a Self-Review</h2>
<p class="lead">
VerdictTank is a proposal review engine, not a proposal writer. It ingests a finished document
and returns scored dimensions plus a ranked list of concrete Fix-It items. The pipeline was not
designed in the abstract. It was hardened across four architectural generations, and then it
was pointed at its own proposals.
</p>
<div class="evolution-grid">
<div class="evolution-card">
<div class="version-badge">v2</div>
<h3 style="margin-top:0">Single-Model Scorer</h3>
<div class="breakthrough-label">Breakthrough: the dual-axis rubric</div>
<p>
v2 established the core insight: a proposal has two independent quality axes. Narrative
quality (clarity, structure, persuasion) and compliance quality (does it actually answer
the scored requirements). One model scored both from one prompt. It proved the concept and
exposed the flaw: the axes bled together. A beautifully written section that missed a
mandatory requirement scored too high, because the same reasoning pass that admired the
prose also graded the compliance.
</p>
</div>
<div class="evolution-card">
<div class="version-badge">v3</div>
<h3 style="margin-top:0">Separated Scoring Passes</h3>
<div class="breakthrough-label">Breakthrough: axis isolation</div>
<p>
v3 split scoring into two independent passes with two purpose-built prompts. The narrative
pass never sees the compliance rubric. The compliance pass never rewards eloquence. This is
the decision that makes the dual score trustworthy: the two numbers can now disagree, and
their disagreement carries information. A 9/10 narrative next to a 4/10 compliance is a
proposal about to lose.
</p>
</div>
<div class="evolution-card">
<div class="version-badge">v4</div>
<h3 style="margin-top:0">Multimodel Adversarial Review</h3>
<div class="breakthrough-label">Breakthrough: cross-model verification</div>
<p>
A single model scoring in isolation is confidently wrong at a predictable rate. v4
introduced a multimodel pipeline: a fast model produces first-pass scores and Fix-It
candidates, then a stronger model reviews that output adversarially, challenging every
deduction and confirming each Fix-It maps to real proposal text. Scores stopped drifting
between runs. This is the generation that made the output defensible.
</p>
</div>
<div class="evolution-card evolution-card--current">
<div class="version-badge version-badge--current">v5 · Current</div>
<h3 style="margin-top:0">Specialist Panel and the Integrity Gate</h3>
<div class="breakthrough-label">Breakthrough: disagreement as output</div>
<p>
v5 replaces the adversarial pair with an 11-seat specialist panel across nine vendors, and
adds a synthesis seat whose only job is to compute panel statistics, flag scores more than
1.5 standard deviations from the mean, and reconcile the verdict against the evidence.
The output is no longer a number. It is a number, a spread, an outlier list, and a ranked
set of material errors with the judge that caught each one.
</p>
</div>
</div>
<h3>The self-review that broke the old thesis</h3>
<p>
Before selling a review engine we ran the engine on our own work. We assembled the panel and
scored three real proposals, in full, with the same prompts and rubric a paying customer would
get. One of the three was VerdictTank's own sibling product. The panel returned a NO GO on it.
</p>
<p>
That run cost roughly $150 in inference and returned 8 of 11 seats. Two seats were lost to a
provider credit wall hit mid-run and one to a model family that could not be dispatched at all.
The incomplete panel is why v2.3 of the judge pool spec now requires a pre-flight health gate
and a pre-baked failover roster, covered in section 5. The results below are what those 8 seats
produced, and we report them at 8 seats rather than extrapolating to 11.
</p>
<div class="callout callout--proof">
<strong>The pipeline is its own reference implementation.</strong> The full architecture is
documented in the companion
<a href="architecture.html">technical architecture document</a> to a standard where an engineer
can implement it from the spec alone. VerdictTank is pre-revenue. We make zero claims about
users, beta cohorts, or external validation. What we claim is narrower and verifiable: the
architecture is built, the pipeline runs, it was executed against three real proposals on
2026-08-12, and it failed its own headline thesis in public.
</div>
</section>
<!-- ====== 3. VALIDATION DATA ====== -->
<section id="validation">
<h2><span class="n">03</span>Validation Run: 3 Real Proposals, 8 Reporting Seats</h2>
<p class="lead">
Every figure in this section comes from the 2026-08-12 validation run. Nothing is modeled,
projected, or illustrative. The solo baseline is Claude Opus 5 scoring the same documents
against the same 10-dimension rubric.
</p>
<h3>Panel mean vs solo baseline</h3>
<table>
<thead>
<tr><th>Proposal</th><th>Panel mean</th><th>Solo baseline</th><th>Delta</th><th>Panel spread</th><th>Solo spread</th><th>Verdict</th></tr>
</thead>
<tbody>
<tr>
<td><strong>RFP Tank v1.0</strong></td><td>4.40</td><td>4.93</td>
<td style="color:var(--bad)">-0.53</td><td>3.7</td><td>1.4</td>
<td><span class="tag tag-red">NO GO</span></td>
</tr>
<tr>
<td><strong>VentureBuilt v2</strong></td><td>6.14</td><td>6.10</td>
<td style="color:var(--accent2)">+0.04</td><td>5.4</td><td>2.4</td>
<td><span class="tag tag-amber">CONDITIONAL GO</span></td>
</tr>
<tr>
<td><strong>CartMySupply</strong></td><td>4.29</td><td>5.00</td>
<td style="color:var(--bad)">-0.71</td><td>2.6</td><td>1.8</td>
<td><span class="tag tag-red">NO GO</span></td>
</tr>
<tr class="comp-table__us">
<td><strong>Aggregate</strong></td><td><strong>4.94</strong></td><td><strong>5.34</strong></td>
<td><strong>-0.40</strong></td><td colspan="2"><strong>Panel spread exceeded solo spread on all 3</strong></td>
<td><strong>THESIS FAIL</strong></td>
</tr>
</tbody>
</table>
<p class="table-note">
Thesis under test: the panel must show a greater than 0.5 point advantage over the solo mean to
justify premium pricing. Result: FAIL on all three proposals and FAIL in aggregate. Panel
composition for this run was 8 reporting judges (4 Band A, 4 Band B) out of 11 specified seats,
a 73% coverage rate.
</p>
<h3>Why the panel scored lower</h3>
<p>
The panel does not elevate scores. It sharpens error detection, and error detection on a flawed
document moves the number down. All three proposals contained severe cross-cutting defects that
additional specialist scrutiny exposed more precisely: fatal execution gaps, competitive
mispositioning, and legal blockers. The solo baseline was directionally correct on all three.
The panel added precision, not points.
</p>
<p>
That is the entire finding, and it inverts the sales pitch. If your proposal is sound, the panel
will roughly agree with a good solo model and cost you more. If your proposal has a hole in it,
the panel finds the hole and the solo model does not. You are not buying a score. You are buying
the probability that a specific, expensive, named mistake gets caught before an evaluator or an
investor finds it for you.
</p>
<h3>Specialist divergence, measured</h3>
<table>
<thead>
<tr><th>Observation</th><th>Evidence from the run</th><th>What it means</th></tr>
</thead>
<tbody>
<tr>
<td>Generalist seats run optimistic</td>
<td>Gemini Pro scored RFP Tank 6.7 as a Band A generalist and 3.0 as the Band B Market specialist. Same model, same document, 3.7 points apart.</td>
<td>Band A generalist scoring without specialist cross-check is systematically over-optimistic. The role, not the model, drives the score.</td>
</tr>
<tr>
<td>Role divergence beats model divergence</td>
<td>DeepSeek V4 Pro scored VentureBuilt 6.4 as Cross-Check C and 2.8 as Execution Feasibility. A 3.6 point split inside one vendor.</td>
<td>Panel diversity is not primarily about buying different vendors. It is about buying different questions.</td>
</tr>
<tr>
<td>One seat can flip a verdict</td>
<td>Remove the 2.8 Execution score from VentureBuilt and the panel averages 6.6, reading as a clean GO. With it, the verdict is CONDITIONAL GO with a named contractor-budget fix.</td>
<td>The lowest score in the panel is frequently the only one doing work. Averaging is what a solo model already does.</td>
</tr>
<tr>
<td>Tight clustering is also a signal</td>
<td>CartMySupply produced zero outliers beyond 1.5 sigma and the tightest spread of the three (sigma 0.89).</td>
<td>Unanimity across nine vendors on a low score is a far stronger NO GO than one model's low score.</td>
</tr>
</tbody>
</table>
</section>
parts/p3.html
+76
View File
@@ -0,0 +1,76 @@
<!-- ====== 4. THE 14 ERRORS ====== -->
<section id="errors">
<h2><span class="n">04</span>The 14 Errors: Every One Named</h2>
<p class="lead">
This is the product. Fourteen material errors the 8-judge panel caught that the solo baseline
missed or underweighted, grouped by failure class. Each is a real finding from the 2026-08-12
run against a real document.
</p>
<div class="errgrid">
<div class="errcard"><div class="n">3</div><div class="t">Revenue arithmetic errors</div><div class="d">Headline numbers that contradict the proposal's own inputs.</div></div>
<div class="errcard"><div class="n">4</div><div class="t">Competitive mispositionings</div><div class="d">Named, funded, shipping incumbents the document treated as absent.</div></div>
<div class="errcard"><div class="n">3</div><div class="t">Execution infeasibilities</div><div class="d">Build plans that cannot be delivered at the stated budget or timeline.</div></div>
<div class="errcard"><div class="n">2</div><div class="t">Legal compliance blockers</div><div class="d">Registration and privacy obligations that gate launch entirely.</div></div>
<div class="errcard"><div class="n">2</div><div class="t">Team capacity impossibilities</div><div class="d">Founder hour budgets that exceed the hours available.</div></div>
<div class="errcard" style="border-left-color:var(--warn)"><div class="n">14</div><div class="t">Total, across 3 documents</div><div class="d">Mean 4.7 material errors per proposal reviewed.</div></div>
</div>
<h3>RFP Tank v1.0 · panel 4.40 vs solo 4.93</h3>
<table>
<thead><tr><th style="width:180px">Class</th><th>Error the panel caught</th><th style="width:150px">Caught by</th></tr></thead>
<tbody>
<tr><td><span class="tag tag-red">Revenue</span></td><td>Three mutually inconsistent Year-1 revenue figures inside one document: $1.2M, $1.361M, and $372K. Plus a 22% MRR ramp inconsistency the narrative never reconciles.</td><td>Financial Integrity</td></tr>
<tr><td><span class="tag tag-red">Competitive</span></td><td>CLEATUS is a real, funded competitor at $4M seed with public product-led pricing of $39 to $250/mo, occupying the identical quadrant. The proposal does not name it. The Band A generalist seat actually cited CLEATUS pricing as a positive signal.</td><td>Market Reality</td></tr>
<tr><td><span class="tag tag-red">Competitive</span></td><td>GovEagle pricing referenced at a 15x inconsistency against the proposal's own comparison table.</td><td>Market Reality</td></tr>
<tr><td><span class="tag tag-red">Execution</span></td><td>Five of seven features marked TO BUILD at HIGH effort. The real-time Compliance Copilot alone needs 2 to 3 developers for 8 to 12 weeks. The plan allocates 4 weeks, solo.</td><td>Execution Feasibility</td></tr>
<tr><td><span class="tag tag-red">Team</span></td><td>A solo founder shipping a 7-feature AI SaaS in 10 weeks, with hiring contingent on revenue that requires the product to already exist. A closed loop with no entry point.</td><td>Team / Founder</td></tr>
<tr><td><span class="tag tag-red">Legal</span></td><td>No privacy policy and no terms of service, against FAR and CUI exposure, with ITAR implications on German-hosted infrastructure.</td><td>Legal / Regulatory</td></tr>
</tbody>
</table>
<p class="table-note">
Panel verdict: NO GO. Estimated rework 40+ hours. Recommendation is to cut scope to two features,
extend to 20 weeks, hire a second developer before month one, rebuild the financial model, and
address CLEATUS directly.
</p>
<h3>VentureBuilt v2 · panel 6.14 vs solo 6.10</h3>
<table>
<thead><tr><th style="width:180px">Class</th><th>Error the panel caught</th><th style="width:150px">Caught by</th></tr></thead>
<tbody>
<tr><td><span class="tag tag-red">Execution</span></td><td>Contractor budget broken by a factor of 4 to 7. The stated $1,500/mo implies $11 to $22 per hour against a market rate of $75 to $100. At real rates that budget buys 105 to 140 hours and leaves roughly 800 hours on the founder.</td><td>Execution Feasibility</td></tr>
<tr><td><span class="tag tag-red">Revenue</span></td><td>Year 2 stated on a run-rate basis rather than recognized revenue. Restated correctly, the healthy scenario loses roughly $11K to $18K.</td><td>Financial Integrity</td></tr>
<tr><td><span class="tag tag-red">Competitive</span></td><td>The uniqueness claim is contradicted by shipping products. LivePlan Plan Review and IdeaProof already occupy the space.</td><td>Market Reality</td></tr>
<tr><td><span class="tag tag-red">Team</span></td><td>37 engagements plus 950 hours plus an MSP day job. The three commitments cannot coexist in one calendar.</td><td>Team / Founder</td></tr>
</tbody>
</table>
<p class="table-note">
Panel verdict: CONDITIONAL GO with six named conditions. Estimated rework 15 to 20 hours. This is
the case that most clearly shows the value: the panel mean (6.14) and the solo mean (6.10) are
statistically tied, so on score alone the panel added nothing. What it added was a bimodal split,
Financial 8.2 against Execution 2.8, and the four errors above.
</p>
<h3>CartMySupply · panel 4.29 vs solo 5.00</h3>
<table>
<thead><tr><th style="width:180px">Class</th><th>Error the panel caught</th><th style="width:150px">Caught by</th></tr></thead>
<tbody>
<tr><td><span class="tag tag-red">Revenue</span></td><td>The $2.7M headline is wrong by 7x to 10x against the proposal's own inputs, which compute to $269K. Stripe fees understated by roughly $11K per year. CAC absent entirely.</td><td>Financial Integrity</td></tr>
<tr><td><span class="tag tag-red">Competitive</span></td><td>TeacherLists already solves the identical problem, free, across 2 million lists. Target ships native School List Assist. No technical moat is claimed or demonstrable.</td><td>Market Reality</td></tr>
<tr><td><span class="tag tag-red">Execution</span></td><td>Amazon PA-API 5 removed Cart API support. Target has no self-serve multi-item cart API. Walmart requires separate catalog matching. The core mechanic of the product does not have a supported integration path at any of the three named retailers.</td><td>Execution Feasibility</td></tr>
<tr><td><span class="tag tag-red">Legal</span></td><td>Charitable solicitation registration required in 40+ states at $30K to $75K, plus COPPA exposure and FTC penalty risk. Compliance cost of $60K to $150K exceeds projected Year-1 revenue of $3K to $14K by an order of magnitude.</td><td>Legal / Regulatory</td></tr>
</tbody>
</table>
<p class="table-note">
Panel verdict: NO GO, unanimous, zero outliers, tightest spread of the three. Estimated rework
60+ hours. The build estimate of 116 hours was independently judged 4x to 10x too low.
</p>
<div class="callout callout--win">
<strong>Read the legal row again.</strong> A $60K to $150K registration obligation against
$3K to $14K of projected revenue is not a scoring nuance. It is the difference between a
business and a fine. A solo model reading the same document returned a 5.0 and did not raise it.
That single finding is worth more than every point of score elevation the old thesis promised.
</div>
</section>
parts/p4.html
+203
View File
@@ -0,0 +1,203 @@
<!-- ====== 5. JUDGE POOL v2.3 ====== -->
<section id="panel">
<h2><span class="n">05</span>Judge Pool v2.3: 11 Seats, 9 Vendors, Zero Double-Ups</h2>
<p class="lead">
The panel that produced the validation data ran at 8 of 11 seats because two seats hit a
provider credit wall mid-run and one model family could not be dispatched at all. v2.3 is the
spec written in response to that failure. Full detail lives in the
<a href="judge-pool-spec.md">judge pool specification v2.3</a>.
</p>
<h3>The roster</h3>
<table>
<thead><tr><th>Band</th><th>Seat</th><th>Model</th><th>Vendor</th><th>Scores</th></tr></thead>
<tbody>
<tr><td>0</td><td>Research Agent</td><td>Grok 4.5</td><td>xAI</td><td>No</td></tr>
<tr><td>A</td><td>Primary Reviewer</td><td>Claude Opus 5</td><td>Anthropic</td><td>Yes</td></tr>
<tr><td>A</td><td>Cross-Check A</td><td>DeepSeek V4 Flash</td><td>DeepSeek</td><td>Yes</td></tr>
<tr><td>A</td><td>Cross-Check B</td><td>Gemini Pro Latest</td><td>Google</td><td>Yes</td></tr>
<tr><td>A</td><td>Cross-Check C</td><td>DeepSeek V4 Pro</td><td>DeepSeek</td><td>Yes</td></tr>
<tr><td>A</td><td>Legal / Regulatory</td><td>Claude Sonnet 5</td><td>Anthropic</td><td>Yes</td></tr>
<tr><td>B</td><td>Financial Integrity</td><td>MiniMax-M3</td><td>MiniMax</td><td>Yes</td></tr>
<tr><td>B</td><td>Team / Founder</td><td>Claude Fable 5</td><td>Anthropic</td><td>Yes</td></tr>
<tr><td>B</td><td>Market Reality</td><td>Qwen3.7 Plus</td><td>Alibaba</td><td>Yes</td></tr>
<tr><td>B</td><td>Execution Feasibility</td><td>GPT-5.2 Pro</td><td>OpenAI</td><td>Yes</td></tr>
<tr><td>C</td><td>Synthesis &amp; Integrity Gate</td><td>Kimi K2.6</td><td>Moonshot</td><td>No</td></tr>
</tbody>
</table>
<p class="table-note">
Nine distinct vendors across eleven seats. Nine distinct scoring models. Zero model double-ups:
no single model occupies two scoring seats, which is the constraint that keeps correlated
failure out of the panel mean. Maximum vendor concentration is Anthropic at 3 of 11 (27.3%),
comfortably inside the 40% ceiling. DeepSeek holds 2 of 11 (18.2%). Every remaining vendor holds
exactly one seat.
</p>
<h3>What changed in v2.3</h3>
<div class="defense-grid">
<div class="defense-item">
<h4 style="margin-top:0">Pre-flight health gate</h4>
<p>
Before any scoring begins, the orchestrator pings every rostered model with a 5-second
probe and writes the result to a per-model health file. Any model returning HTTP 400,
HTTP 429, or a no-healthy-deployments error is swapped for its pre-assigned failover before
a single scoring call is spent. The 2026-08-12 run burned roughly 12 dispatches discovering
dead models at runtime. That failure mode is now closed.
</p>
</div>
<div class="defense-item">
<h4 style="margin-top:0">Pre-baked failover roster</h4>
<p>
Every seat carries a named failover from a different vendor, resolved at gate time rather
than improvised mid-run. Failover selection preserves both the vendor-diversity ceiling and
the no-double-up rule, so a degraded panel is still a valid panel rather than an
accidentally correlated one.
</p>
</div>
<div class="defense-item">
<h4 style="margin-top:0">Credit-wall resilience</h4>
<p>
The provider credit exhaustion that cost two seats mid-run is now detected at the gate and
treated as an availability failure, not an error. Anthropic capacity has been restored and
the Primary Reviewer seat runs Claude Opus 5 as specified.
</p>
</div>
<div class="defense-item">
<h4 style="margin-top:0">Permanent exclusions</h4>
<p>
One frontier model family proved structurally incapable of running as a panel seat under
our orchestration and is permanently excluded from the roster, not merely deprioritized.
Excluded models cannot be selected as a failover target either.
</p>
</div>
</div>
<h3>Latency</h3>
<p>
v2.3 targets a critical path of roughly 113 seconds, against 218 seconds measured on the
v2.1 architecture. The improvement comes from band parallelism: Band A and Band B seats execute
concurrently rather than sequentially, and the Synthesis seat is the only stage that must wait
for all scoring seats to return.
</p>
<h3>The ten scored dimensions</h3>
<p>
Every scoring seat rates the proposal 1 to 10 on the same ten dimensions, so panel spread is
computed dimension by dimension and not only in aggregate:
</p>
<div class="arch-box">
Problem Clarity · Market Opportunity · Product Differentiation
Revenue Model Viability · Go-to-Market Strategy · Competitive Moat
Financial Projections · Team / Execution · Risk Mitigation · Legal / Compliance
</div>
<h3>The Synthesis and Integrity Gate</h3>
<p>
The Band C seat never scores. It reads all scoring output and performs a fixed checklist:
verify score arithmetic, compute panel means and per-dimension spread, compute the delta against
the solo baseline, flag every score more than 1.5 standard deviations from the panel mean with a
written rationale, and confirm the verdict follows from panel evidence rather than from the
Primary Reviewer alone. That gate is what turns eleven opinions into one auditable report.
</p>
</section>
<!-- ====== 6. WORKED EXAMPLE ====== -->
<section id="worked">
<h2><span class="n">06</span>Worked Example: VentureBuilt v2, Where the Score Said Nothing</h2>
<p class="lead">
This is the clearest case in the validation set, because it is the one where score elevation
delivered exactly zero and error detection delivered everything. Real scores from the
2026-08-12 run.
</p>
<div class="wt-scores">
<div class="score-block score-block--narrative">
<div class="score-label">Panel mean · 8 judges</div>
<div class="score-value">6.14<span>/10</span></div>
<p class="muted" style="margin-bottom:0">Median 6.45 · standard deviation 1.758 · spread 5.4 (min 2.8, max 8.2)</p>
</div>
<div class="score-block score-block--compliance" style="border-left-color:var(--muted)">
<div class="score-label">Solo baseline · Claude Opus 5</div>
<div class="score-value" style="color:var(--muted)">6.10<span>/10</span></div>
<p class="muted" style="margin-bottom:0">Spread 2.4 · delta +0.04 · statistically tied with the panel</p>
</div>
</div>
<h3>Individual seat scores</h3>
<table>
<thead><tr><th>Seat</th><th>Model</th><th>Score</th><th>Sigma from mean</th><th>Flag</th></tr></thead>
<tbody>
<tr><td>Band B · Financial Integrity</td><td>MiniMax</td><td><strong style="color:var(--accent2)">8.2</strong></td><td>+1.17</td><td>Within 1.5 sigma</td></tr>
<tr><td>Band B · Market Reality</td><td>Gemini</td><td>7.7</td><td>+0.89</td><td>Within 1.5 sigma</td></tr>
<tr><td>Band A · Primary Reviewer</td><td>Opus 5</td><td>6.9</td><td>+0.43</td><td>Within 1.5 sigma</td></tr>
<tr><td>Band A · Legal / Regulatory</td><td>Qwen</td><td>6.5</td><td>+0.21</td><td>Within 1.5 sigma</td></tr>
<tr><td>Band A · Cross-Check C</td><td>DeepSeek V4 Pro</td><td>6.4</td><td>+0.15</td><td>Within 1.5 sigma</td></tr>
<tr><td>Band A · Cross-Check B</td><td>Gemini</td><td>6.2</td><td>+0.04</td><td>Within 1.5 sigma</td></tr>
<tr><td>Band B · Team / Founder</td><td>Kimi</td><td>4.4</td><td>-0.99</td><td>Within 1.5 sigma</td></tr>
<tr><td>Band B · Execution Feasibility</td><td>DeepSeek V4 Pro</td><td><strong style="color:var(--bad)">2.8</strong></td><td>-1.91</td><td><span class="tag tag-red">OUTLIER</span></td>
</tr>
</tbody>
</table>
<div class="callout callout--fail">
<strong>The average is a lie of composition.</strong> A 6.14 reads as a solid, fundable
proposal with room to improve. The distribution says something completely different: the money
is excellent (8.2) and the delivery plan is close to unworkable (2.8). Those are not two
opinions about one thing. They are two accurate findings about two different things, and
averaging them produces a number that describes neither.
</div>
<h3>What the outlier actually found</h3>
<p>
The 2.8 was not a grumpy model. The Integrity Gate challenged it at 1.91 sigma and it survived
the challenge on evidence: a contractor budget broken 4x to 7x, an architecture that regressed
from v1 with no schema and no API contract, and a founder workload of 37 engagements plus 950
hours alongside an MSP day job. The Team seat (4.4) and the Band A generalists (6.2 to 6.9)
all acknowledged the same workload problem. They weighted it less severely. The Execution
specialist is the only seat that forced it into the verdict.
</p>
<div class="highlight">
Remove that one seat and the panel averages <strong>6.6</strong>, which reads as a clean GO and
ships a proposal with an 800-hour founder gap in it. The specialist seat cost a few cents of
inference and changed the verdict.
</div>
<h3>Fix-It items, ranked by materiality</h3>
<ol class="fixit-list">
<li>
<div class="fixit-head"><span class="fixit-tag fixit-tag--critical">Critical</span><span class="fixit-delta">Execution</span></div>
<strong>Fix the contractor budget or cut the scope.</strong> $1,500/mo buys 105 to 140 hours
at market rates, not the volume the plan assumes. Raise to roughly $7,500/mo or reduce scope
to fit the hours actually purchased.
</li>
<li>
<div class="fixit-head"><span class="fixit-tag fixit-tag--critical">Critical</span><span class="fixit-delta">Financial</span></div>
<strong>Restate Year 2 on a recognized-revenue basis.</strong> On run-rate the year looks
healthy. On recognized revenue it loses roughly $11K to $18K. Present both.
</li>
<li>
<div class="fixit-head"><span class="fixit-tag fixit-tag--high">High</span><span class="fixit-delta">Competitive</span></div>
<strong>Withdraw or qualify the uniqueness claim.</strong> LivePlan Plan Review and IdeaProof
already ship in this space. Reposition on a defensible axis.
</li>
<li>
<div class="fixit-head"><span class="fixit-tag fixit-tag--high">High</span><span class="fixit-delta">Team</span></div>
<strong>Name the contractor and the sourcing plan before Phase 2.</strong> A budget line with
no named person is not a capacity plan.
</li>
<li>
<div class="fixit-head"><span class="fixit-tag fixit-tag--medium">Medium</span><span class="fixit-delta">Go-to-market</span></div>
<strong>Map the Year 1 to Year 2 GTM bridge.</strong> Eight net-new signups per month appear
in the model with no acquisition mechanism behind them.
</li>
<li>
<div class="fixit-head"><span class="fixit-tag fixit-tag--medium">Medium</span><span class="fixit-delta">Legal</span></div>
<strong>Complete data protection and trademark clearance before Phase 0 to 1.</strong>
</li>
</ol>
<p class="table-note">
Panel verdict: CONDITIONAL GO. Estimated rework 15 to 20 hours. The solo baseline returned a
6.10 and none of the six conditions above.
</p>
</section>
parts/p5.html
+179
View File
@@ -0,0 +1,179 @@
<!-- ====== 7. PRICING ====== -->
<section id="pricing">
<h2><span class="n">07</span>Pricing: Priced Per Error Found, Not Per Point Gained</h2>
<p class="lead">
Three tiers. The pricing logic follows the revised thesis directly: a panel run is worth what a
caught error is worth, and a caught error is worth far more than a point of score.
</p>
<div class="pricing-grid">
<div class="price-card">
<h3 style="margin-top:0">Free</h3>
<div class="price">$0</div>
<ul>
<li>1 full review</li>
<li>Top 3 Fix-It items</li>
<li>Panel score and spread</li>
<li>Reduced panel size</li>
</ul>
<div class="price-purpose">Purpose: prove it on one document</div>
</div>
<div class="price-card price-card--featured">
<div class="ribbon">Most popular</div>
<h3 style="margin-top:0">Pro</h3>
<div class="price">$79<span>/mo</span></div>
<ul>
<li>5 reviews per month</li>
<li>Full 11-seat panel</li>
<li>Complete Fix-It list, ranked</li>
<li>Panel spread and outlier flags</li>
<li>Solo-baseline delta comparison</li>
<li>Re-score loop with before and after</li>
</ul>
<div class="price-purpose">Purpose: the founder or solo bid writer</div>
</div>
<div class="price-card">
<h3 style="margin-top:0">Enterprise</h3>
<div class="price">$299<span>/mo</span></div>
<ul>
<li>Unlimited reviews</li>
<li>White-label branding</li>
<li>Multi-seat team workspaces</li>
<li>Configurable judge pool</li>
<li>Corpus isolation and data controls</li>
<li>Priority pipeline and support</li>
</ul>
<div class="price-purpose">Purpose: proposal teams running color reviews</div>
</div>
</div>
<h3>What a review costs us, and why the panel is affordable</h3>
<p>
The validation run cost approximately $150 in inference for three full proposals across eight
reporting seats, including retries against dead models before the health gate existed. That
burn is the honest anchor for panel economics: a clean 11-seat run on one proposal, with the
health gate preventing wasted dispatches, sits well inside single-digit dollars.
</p>
<p>
The reason a full 11-seat panel fits a $79 tier at five reviews per month is vendor mix. Only
a minority of seats run premium frontier models. The specialist Band B seats run strong
mid-tier models from five different vendors, which is where the error-detection value came from
in validation. Panel diversity is cheaper than panel depth, and diversity is what caught the 14.
</p>
<h3>Why the value question is not the score question</h3>
<table>
<thead><tr><th>Error class</th><th>Real example from validation</th><th>Cost of missing it</th></tr></thead>
<tbody>
<tr><td>Legal blocker</td><td>Charitable solicitation registration in 40+ states</td><td>$30K to $75K of registration, against $3K to $14K of projected revenue</td></tr>
<tr><td>Compliance total</td><td>Full first-year compliance load on the same proposal</td><td>$60K to $150K, exceeding Year-1 revenue by roughly 10x</td></tr>
<tr><td>Execution gap</td><td>Contractor budget short by 4x to 7x</td><td>Roughly 800 unbudgeted founder hours</td></tr>
<tr><td>Revenue arithmetic</td><td>$2.7M headline against $269K computed from the document's own inputs</td><td>Credibility with any investor who checks the math, which is all of them</td></tr>
<tr><td>Competitive blind spot</td><td>A $4M-seed funded direct rival never named in the document</td><td>The first question in the room, unanswered</td></tr>
</tbody>
</table>
<p class="table-note">
A single caught item in the top two rows pays for a decade of the Pro tier. That is the entire
pricing argument, and it does not depend on the panel producing a higher score, which it does
not.
</p>
<h3>Positioned against the authoring category</h3>
<table>
<thead><tr><th>Comparison</th><th>Their price</th><th>VerdictTank</th><th>Multiple</th></tr></thead>
<tbody>
<tr><td>Pro vs Bidara Starter</td><td>$499/mo</td><td>$79/mo</td><td><strong>6.3x cheaper</strong></td></tr>
<tr><td>Pro vs AutoRFP.ai Scale</td><td>$899/mo</td><td>$79/mo</td><td><strong>11.4x cheaper</strong></td></tr>
<tr><td>Enterprise vs Bidara Starter</td><td>$499/mo</td><td>$299/mo</td><td><strong>1.7x cheaper</strong></td></tr>
<tr><td>Enterprise vs AutoRFP.ai Scale</td><td>$899/mo</td><td>$299/mo</td><td><strong>3.0x cheaper</strong></td></tr>
</tbody>
</table>
<p>
We are not a proposal team in a box. We are one high-value pass in the workflow. A buyer already
spending $499 to $899 per month on an authoring tool should be able to add the error-detection
layer without a second budget conversation. Pricing Pro at $79 makes VerdictTank an add-on
decision rather than a platform decision.
</p>
</section>
<!-- ====== 8. COMPETITIVE LANDSCAPE ====== -->
<section id="competitive">
<h2><span class="n">08</span>Competitive Landscape: Nobody Sells the Errors</h2>
<p class="lead">
Every AI-native player in this space is an authoring tool. They generate drafts. The nearest
substitute for what we do is not a competitor product at all. It is a single frontier model
and a prompt, and validation showed exactly what that substitute misses.
</p>
<table>
<thead>
<tr><th>Product</th><th>Category</th><th>Published price</th><th>Relationship to VerdictTank</th></tr>
</thead>
<tbody>
<tr><td><strong>AutogenAI</strong></td><td>Enterprise authoring</td><td>Custom, sales-led, no self-serve</td><td>Complementary. We find the errors in what it writes.</td></tr>
<tr><td><strong>Civio</strong></td><td>Gov RFP authoring</td><td>Custom, sales-led</td><td>Complementary. Downstream reviewer.</td></tr>
<tr><td><strong>Bidara</strong></td><td>Mid-market authoring</td><td>$499/mo Starter</td><td>Complementary. Transparent pricing, natural comparison anchor.</td></tr>
<tr><td><strong>AutoRFP.ai</strong></td><td>Response automation</td><td>$899/mo Scale</td><td>Complementary. Reviews its drafts.</td></tr>
<tr><td><strong>DeepRFP</strong></td><td>Lean-team authoring</td><td>$89/user/mo</td><td>Complementary. Lowest per-seat price in the category, natural partner.</td></tr>
<tr><td><strong>A solo frontier model</strong></td><td>DIY substitute</td><td>API cost only</td><td><strong>The real competitor.</strong> Measured: misses or underweights the material errors a panel catches.</td></tr>
<tr class="comp-table__us"><td><strong>VerdictTank</strong></td><td><strong>Panel error detection</strong></td><td><strong>Free / $79 Pro / $299 Enterprise</strong></td><td><strong>The only 11-seat, 9-vendor review panel with a published integrity gate</strong></td></tr>
</tbody>
</table>
<p class="table-note">
Competitor prices are vendors' own published rates as of July 2026. Tools without public pricing
are shown as sales-led. Every named product was verified to exist and to occupy the authoring
category.
</p>
<h3>Why the DIY substitute is the row that matters</h3>
<p>
Any buyer sophisticated enough to want proposal review can paste their document into a frontier
model and ask for a critique. That is the honest competitive threat, and it is the one we tested
against rather than around. The result is in section 3: the solo model returns a defensible,
directionally correct score, and it returned none of the six VentureBuilt conditions, none of
the CartMySupply compliance exposure, and none of the RFP Tank competitive reality.
</p>
<div class="defense-grid">
<div class="defense-item">
<h4 style="margin-top:0">1. Incumbents cannot sell honest criticism</h4>
<p>
Authoring tools sell the promise that they write your proposal. A brutal error list on the
output that same tool just produced is a direct admission the generated draft is losing. It
is structurally against their interest. We have no draft to defend. The verdict is the
product.
</p>
</div>
<div class="defense-item">
<h4 style="margin-top:0">2. Review is where the money is decided</h4>
<p>
Every serious bid already goes through a review gate, the color-team pass organizations run
manually by pulling senior staff off billable work. That labor is expensive, slow,
inconsistent between reviewers, and unavailable to the solo consultant. The demand is proven
by the existence of the manual process.
</p>
</div>
<div class="defense-item">
<h4 style="margin-top:0">3. Panel orchestration is a real moat</h4>
<p>
Nine vendors, health gating, pre-baked failover, no model double-ups, and an integrity gate
that challenges its own outliers is not a prompt. It is an operations problem, and the
2026-08-12 run is the evidence of what it costs to learn.
</p>
</div>
<div class="defense-item">
<h4 style="margin-top:0">4. An empty category sets its own price</h4>
<p>
A crowded category means the budget line exists and you fight for share. An empty review
category means we define the line and set the reference price, while remaining complementary
to every authoring tool in the table above.
</p>
</div>
</div>
<div class="positioning-statement">
The authoring tools write proposals. A solo model grades them smoothly.<br>
VerdictTank tells you the fourteen things both of them got wrong.
</div>
</section>
parts/p6.html
+228
View File
@@ -0,0 +1,228 @@
<!-- ====== 9. DEPLOYMENT ====== -->
<section id="deployment">
<h2><span class="n">09</span>Deployment Options</h2>
<p class="lead">
Two supported deployment shapes. Both are managed by IT Pro Partner below the application layer.
</p>
<div class="strategy-cols">
<div class="defense-item">
<h4 style="margin-top:0;text-transform:none;color:var(--accent);font-size:15px">Option A · ITPP-INFRA Shared</h4>
<p>
Runs on existing netcup RS 4000 infrastructure alongside IT Pro Partner operations. Same
Wasabi S3 backup pipeline, same Caddy reverse proxy, same monitoring stack (Prometheus and
Grafana). Zero new infrastructure cost. Suitable for launch through Series A.
</p>
<ul>
<li>netcup RS 4000 (app3), Docker Compose</li>
<li>Wasabi S3 daily backups plus 15-minute sync</li>
<li>Managed by the IT Pro Partner infrastructure team</li>
</ul>
</div>
<div class="defense-item">
<h4 style="margin-top:0;text-transform:none;color:var(--accent);font-size:15px">Option B · Dedicated</h4>
<p>
Dedicated netcup or Hetzner instances with a dedicated S3 bucket. Full isolation from ITPP
operational infrastructure. Recommended for post-Series A or enterprise white-label
deployments requiring independent compliance scope.
</p>
<ul>
<li>Dedicated netcup RS or Hetzner CPX instances</li>
<li>Dedicated Wasabi S3 bucket, separate backup schedule</li>
<li>Managed by IT Pro Partner below the application layer</li>
</ul>
</div>
</div>
<p class="table-note">
<strong>Shared responsibility:</strong> IT Pro Partner manages everything below the application
layer (OS, container runtime, networking, backups, monitoring) under both options. The
VerdictTank application and its model pipeline are the product team's responsibility.
</p>
<div class="highlight">
<strong>Panel-specific operational requirement.</strong> Under either option the orchestrator
must hold credentials for nine separate model vendors and must run the pre-flight health gate
before every panel dispatch. Vendor credential rotation and per-vendor spend ceilings are
application-layer concerns and sit with the product team, not with infrastructure.
</div>
</section>
<!-- ====== 10. LEGAL, PRIVACY, COMPLIANCE ====== -->
<section id="legal">
<h2><span class="n">10</span>Legal, Privacy &amp; Compliance</h2>
<p class="lead">
Trademark clearance, the Minimum Viable Legal framework, the controller and processor role map,
the sub-processor training guard, corpus confidentiality, and incident response. Carried forward
from v4.0 and updated for the nine-vendor panel.
</p>
<h3>10.1 USPTO trademark clearance: "VerdictTank"</h3>
<p>
<strong>Status: preliminary clearance only. This is not a substitute for a formal search.</strong>
This assessment was performed with open-web search tools only. USPTO TESS is a
JavaScript-rendered application and a static fetch returns only the search shell with no query
results. Before any trademark application is filed, a live interactive TESS search or a paid
clearance search through a trademark attorney is required.
</p>
<h4>Open-web common-law search results (performed)</h4>
<table>
<thead><tr><th>Search</th><th>Result</th><th>Assessment</th></tr></thead>
<tbody>
<tr><td>"VerdictTank" exact, web-wide</td><td>Only hit is verdicttank.com itself</td><td>No third-party commercial use found</td></tr>
<tr><td>"Verdict Tank" space variant</td><td>Two incidental unrelated hits, neither a business nor a registered mark</td><td>No competing commercial use. Matches are noise.</td></tr>
<tr><td>Trademarkia and Justia proxy queries</td><td>No results returned</td><td>Consistent with no existing registration, but not equivalent to direct TESS</td></tr>
<tr><td>Domain: verdicttank.com</td><td>Live, owned, serving the product</td><td><strong>Primary domain.</strong> Confirms operational use in commerce.</td></tr>
<tr><td>Domain: rfptank.com</td><td>Legacy holding, same naming convention</td><td><strong>Defensive only.</strong> Retained against a family-of-marks argument. Not a product surface.</td></tr>
</tbody>
</table>
<h4>Recommendation</h4>
<ul>
<li>Before Series A close or any public marketing scale-up, commission a formal USPTO clearance search for Classes 9, 42, 35 and 45.</li>
<li>File an intent-to-use application for VERDICTTANK as a standard character word mark, Class 42 primary and Class 9 secondary.</li>
<li>Do not file on the basis of this document alone. It is a preliminary desk review.</li>
</ul>
<h3>10.2 Minimum Viable Legal (MVL) framework</h3>
<p>
MVL is the internal gate name used in the architecture documents as the precondition for
onboarding white-label and enterprise customers. It is not one document. It is five interlocking
instruments that must all exist and be internally consistent before the white-label provisioning
gate turns green.
</p>
<table>
<thead><tr><th>Component</th><th>Purpose</th><th>Applies to</th><th>Status</th></tr></thead>
<tbody>
<tr><td>Terms of Service</td><td>Governs the contractual relationship with every direct user</td><td>All tiers</td><td>Drafting required</td></tr>
<tr><td>Privacy Policy</td><td>GDPR and CCPA compliant notice of collection and use</td><td>All tiers</td><td>Drafting required</td></tr>
<tr><td>Data Processing Addendum</td><td>Article 28 GDPR processor terms</td><td>Enterprise, white-label</td><td>Hard gate on white-label</td></tr>
<tr><td>AI Disclaimer (DISC-001)</td><td>Non-removable versioned notice: output is AI opinion, not professional advice</td><td>Every scored surface</td><td>Engineering spec complete, legal copy needs counsel sign-off</td></tr>
<tr><td>Limitation of Liability</td><td>Caps aggregate liability at the lesser of $100 or fees paid in the preceding 12 months</td><td>All tiers, embedded in ToS</td><td>Drafting required</td></tr>
<tr><td>Governing law and venue</td><td>Recommend Delaware law with Georgia venue, pending confirmation of incorporation state</td><td>All tiers</td><td>Pending counsel</td></tr>
<tr><td>GDPR readiness</td><td>Lawful basis mapped per role. Articles 28, 33 and 34. SCCs or IDTA for EU transfers.</td><td>Any EU user</td><td>Framework mapped, SCC execution pending white-label launch</td></tr>
<tr><td>CCPA and CPRA readiness</td><td>Service-provider contract terms and a consumer rights workflow</td><td>Any California resident</td><td>DSAR workflow build pending</td></tr>
</tbody>
</table>
<p class="table-note">
The Free, Pro and Enterprise tiers require Terms of Service, Privacy Policy and the AI Disclaimer
at minimum before any paid launch.
</p>
<h3>10.3 Controller and processor role map</h3>
<table>
<thead><tr><th>Data flow</th><th>Role</th><th>Legal basis</th><th>Agreements required</th></tr></thead>
<tbody>
<tr><td>Free tier submission and review</td><td>Controller</td><td>Contract plus legitimate interest</td><td>ToS, Privacy Policy</td></tr>
<tr><td>Enterprise org admin and org users</td><td>Joint controller</td><td>Performance of contract</td><td>ToS, Enterprise DPA (Art. 26 GDPR)</td></tr>
<tr><td>White-label tenant end-users</td><td>Processor</td><td>Tenant's instructions</td><td>DPA, SCCs or IDTA, published sub-processor list</td></tr>
<tr><td>Panel model API calls, all nine vendors</td><td>Controller of the vendor relationship. Each model vendor is a sub-processor.</td><td>Legitimate interest</td><td>Sub-processor training guard plus a DPA with each vendor</td></tr>
<tr><td>Corpus contribution (aggregate scores and structural metadata)</td><td>Controller, secondary-use basis</td><td><strong>Opt-in consent.</strong> Cannot ride on contract or legitimate interest under the purpose limitation principle, Art. 5(1)(b).</td><td>Explicit opt-in UI, anonymization pipeline, retention separate from the review record</td></tr>
</tbody>
</table>
<div class="callout">
<strong>The nine-vendor panel raises the sub-processor row from a formality to the primary
compliance surface.</strong> Every additional vendor seat is an additional sub-processor that
must carry a confirmable training opt-out before it can receive customer content. A vendor that
cannot satisfy the guard below cannot hold a panel seat, regardless of how well it scores.
</div>
<h3>10.4 Sub-processor training guard</h3>
<h4>Model clause for vendor DPAs</h4>
<div class="arch-box">
Vendor shall not use Customer Data (including all inputs, outputs, prompts, completions,
and any content submitted via the Vendor's API) to train, fine-tune, retrain, or otherwise
improve any machine learning model, except with Customer's prior written consent on a
per-instance basis. Vendor shall not retain Customer Data beyond the minimum period
technically necessary to provide the API response.
</div>
<p>
<strong>Enforcement rule:</strong> a vendor without a public, contractually confirmable training
opt-out is excluded from the panel roster entirely and cannot be selected as a failover target.
The only acceptable path for a non-compliant provider is a customer-side, explicit, revocable
opt-in. Never a silent default, and never for corpus-eligible content. Each of the nine rostered
vendors is audited against this clause before it is eligible for a seat, and the audit is
re-run at each roster revision.
</p>
<h3>10.5 Corpus confidentiality</h3>
<p>
The corpus is VerdictTank's most valuable long-term asset and its highest confidentiality
exposure.
</p>
<table>
<thead><tr><th>Data type</th><th>Corpus-eligible</th><th>Rationale</th></tr></thead>
<tbody>
<tr><td>Dimension scores and panel spread statistics</td><td>Yes, opt-in</td><td>Structural, not identifying. Core signal.</td></tr>
<tr><td>Structural metadata (vertical, length bucket, revision count, deltas)</td><td>Yes, opt-in</td><td>Enables content and moat analytics</td></tr>
<tr><td>Raw proposal text</td><td><strong>Never</strong></td><td>Confidential business content plus potential third-party PII</td></tr>
<tr><td>Explanation and audit finding text</td><td><strong>Never</strong> in raw form</td><td>Critique text frequently quotes the submission verbatim</td></tr>
<tr><td>Chat refinement transcripts</td><td><strong>Never</strong> as transcript content</td><td>Highest incidental-PII risk of any input surface</td></tr>
</tbody>
</table>
<h4>Anonymization pipeline</h4>
<ol>
<li><strong>Source-content exclusion.</strong> Raw text fields excluded at the schema and ETL level.</li>
<li><strong>Structural extraction only.</strong> ETL reads scored and aggregated fields, never freeform text.</li>
<li><strong>Identifier stripping.</strong> Review, user and org identifiers replaced with a one-way surrogate key.</li>
<li><strong>Free-text quarantine.</strong> A stricter named-entity pass before any inclusion.</li>
<li><strong>k-anonymity floor.</strong> Public content published only when the cohort exceeds a minimum threshold.</li>
</ol>
<p class="table-note">
Corpus contribution is off by default and requires explicit, separate opt-in. It is not bundled
into ToS acceptance and is revocable at any time from account settings.
</p>
<h3>10.6 Incident response</h3>
<p>Structured around the six functions of NIST CSF 2.0.</p>
<table>
<thead><tr><th>Function</th><th>VerdictTank action</th></tr></thead>
<tbody>
<tr><td>Govern</td><td>Named incident commander. Breach classification criteria documented before any incident.</td></tr>
<tr><td>Identify</td><td>Asset inventory: transactional database, corpus database, credentials for all nine model vendors, white-label tenant segments.</td></tr>
<tr><td>Protect</td><td>Row-level-security multi-tenant isolation, sanitization gate, sub-processor training guard, per-vendor spend ceilings.</td></tr>
<tr><td>Detect</td><td>Alerting on anomalous data access, bulk export, and cross-org query attempts. Health-gate telemetry on every panel dispatch.</td></tr>
<tr><td>Respond</td><td>GDPR: 72-hour notification to the supervisory authority (Art. 33). CCPA: notification without unreasonable delay.</td></tr>
<tr><td>Recover</td><td>Post-incident review documented. White-label tenants notified per their individual DPA terms.</td></tr>
</tbody>
</table>
<h4>Wrong-verdict liability</h4>
<p>
<strong>Scenario:</strong> a customer submits a proposal, receives a favorable panel verdict,
acts on it, and the verdict was wrong in a way that led to a bad decision. This is primarily a
reputational risk. The liability cap bounds legal exposure and does nothing for reputation.
</p>
<ul>
<li><strong>Legal layer:</strong> the AI disclaimer fails closed on every surface, a $100 or 12-months-of-fees liability cap applies, and the terms explicitly instruct users not to rely on AI output for investment decisions.</li>
<li><strong>Confidence calibration:</strong> every verdict ships with panel spread, standard deviation, and outlier flags. A verdict with a 5.4-point spread is a materially different signal from a unanimous one, and the report says so on its face.</li>
<li><strong>The published FAIL:</strong> section 3 of this document is itself part of the defense. We publish the case where our own thesis failed, which is a stronger honesty posture than any disclaimer.</li>
<li><strong>Incident playbook:</strong> do not litigate merits publicly, point to the auditable disclaimer version shown to the user, offer a private re-review, and disclose plus correct any systematic flaw found.</li>
</ul>
<h4>Model provider outage disclosure</h4>
<ul>
<li><strong>Public status page</strong> distinguishing VerdictTank infrastructure incidents from upstream model provider incidents.</li>
<li><strong>Degraded-mode behavior:</strong> a panel that ran short of its full eleven seats is flagged visibly with the seat count and which roles failed over. We never silently substitute a provider without disclosure. The 2026-08-12 run is reported at 8 of 11 seats throughout this document for exactly that reason.</li>
<li><strong>SLA language:</strong> uptime commitments are qualified as dependent on upstream provider availability.</li>
</ul>
</section>
</main>
<footer>
<div class="wrap">
VerdictTank v5.0 · <a href="https://verdicttank.com">verdicttank.com</a> ·
<a href="architecture.html">Technical Architecture</a> ·
<a href="judge-pool-spec.md">Judge Pool Spec v2.3</a> ·
<a href="index-v4.0.html">Prior version (v4.0)</a><br>
Pre-revenue. Thesis revised on measured data from the 2026-08-12 validation run.
All domain references are <code>verdicttank.com</code>; <code>rfptank.com</code> is retained
as a defensive holding only.<br>
Every score, delta, spread and error count on this page comes from that run. Nothing is projected.
</div>
</footer>
</body>
</html>