--- contractVersion: 1 title: AI support vendor decision packet description: A fictional security and procurement decision comparing hard gates, weighted evidence, and conditional adoption. language: en theme: light layout: document preset: editorial --- # AI support vendor decision packet **Fictional showcase · 06 August 2026 · decision ready** This demonstration follows a fictional procurement review for Lantern & Field, an invented customer operations company. Every vendor, score, control, price, and evidence record is sample data created for the report engine; none describes a real organization or commercial claim. :::callout{kind="info" title="Decision in one minute"} Select **Cedar Assist** for a 90-day reversible pilot. Meridian Reply has the highest weighted score, but it fails a non-negotiable regional-processing gate. Quill Support passes every gate but trails Cedar on workflow fit and implementation evidence. ::: ::::cards :::card{title="Recommended"} **Cedar Assist · 84/100** Passes all four mandatory constraints with two time-bound conditions. ::: :::card{title="Highest raw score"} **Meridian Reply · 89/100** Disqualified until its support telemetry remains inside the approved region. ::: :::card{title="Qualified alternative"} **Quill Support · 77/100** Meets the baseline but needs more workflow and migration evidence. ::: :::: The review treats a :term[hard gate]{key="hard-gate"} as a pass/fail constraint, never as a weighted preference. An :term[evidence expiry]{key="evidence-expiry"} prevents an old assurance from remaining valid indefinitely. :::glossary{key="hard-gate" term="Hard gate"} A mandatory security, legal, or portability condition. One failed gate disqualifies a candidate regardless of its weighted score. ::: :::glossary{key="evidence-expiry" term="Evidence expiry"} The date after which a control claim must be re-tested or replaced before it can support the decision. ::: ## Gate decision {{include: partials/gates.md}} ![Sample evidence map connecting requirements, vendor evidence, gates, scoring, and a conditional decision](assets/evidence-map.svg) :::popover{title="Ranking exception" trigger="Why not the top score?"} Meridian's 89-point result measures preferences only. Its failed regional-processing gate is non-compensating, so additional usability or price points cannot make the candidate eligible. ::: ## Weighted comparison :::::chart{type="bar" title="Weighted score after evidence review" description="Sample weighted scores rank Meridian Reply first at 89, Cedar Assist second at 84, and Quill Support third at 77; Meridian remains ineligible because it fails a separate mandatory constraint." x-label="Fictional candidate" y-label="Weighted score out of 100"} ::::series{label="Weighted score"} ::point{label="Cedar Assist" value="84"} ::point{label="Meridian Reply" value="89"} ::point{label="Quill Support" value="77"} :::: ::::: | Weighted criterion | Weight | Cedar Assist | Meridian Reply | Quill Support | | --------------------- | ------: | -----------: | -------------: | ------------: | | Agent workflow fit | 30 | 27 | 29 | 21 | | Evidence quality | 25 | 22 | 20 | 19 | | Implementation effort | 20 | 16 | 17 | 15 | | Portability depth | 15 | 12 | 13 | 14 | | Three-year cost | 10 | 7 | 10 | 8 | | **Total** | **100** | **84** | **89** | **77** | ::::tabs{title="Review lenses"} :::tab{label="Scoring method"} Weights were fixed before proposals were opened. Reviewers scored traceable evidence on a five-point rubric, then normalized the result to 100. Hard-gate outcomes were not converted into points. ::: :::tab{label="Evidence quality"} Cedar supplied current audit extracts and a tested export. Meridian supplied strong workflow evidence but an unresolved regional data-flow statement. Quill supplied complete policies but only a tabletop export. ::: :::tab{label="Residual risks"} Cedar must prove deletion propagation and peak-volume response quality during the pilot. The buyer retains weekly export snapshots and a termination rehearsal as portability safeguards. ::: :::: :::disclosure{title="Read the scoring assumptions and limits" open="false"} Scores are fictional ordinal judgments, not measurements of real products. A two-point difference is not treated as statistically meaningful. The recommendation depends on the stated four gates, pilot scope, and evidence dates; changing any of them requires a new decision record. ::: :::modal{title="Reviewer evidence checklist" trigger="Open the evidence checklist"} - Confirm each gate cites a current owner and evidence date. - Recalculate the weighted total without changing pre-declared weights. - Verify the selected vendor has a tested export and deletion route. - Record every pilot condition with an owner, deadline, and exit consequence. ::: ## Decision and conditions :::decision{title="Approve Cedar Assist for a reversible pilot"} Proceed only for the EU support queue and exclude payment data. Security owns deletion verification by 21 August; Customer Operations owns peak-volume acceptance by 28 August. If either condition fails, stop ingestion and execute the tested export within one business day. Meridian and Quill remain alternatives, not fallback approvals. ::: ::::timeline{title="Governed adoption path" description="Four sample checkpoints keep evidence, pilot access, acceptance, and the final rollout decision separate."} :::event{date="08 Aug" title="Evidence locked" kind="accent"} Freeze the reviewed documents, hashes, owners, and expiry dates. ::: :::event{date="12 Aug" title="Restricted pilot starts" kind="warning"} Enable the EU queue with payment data excluded and weekly exports enabled. ::: :::event{date="28 Aug" title="Conditions reviewed" kind="accent"} Evaluate deletion propagation and peak-volume response against the written thresholds. ::: :::event{date="04 Sep" title="Go or exit" kind="success"} Approve wider use only if every gate remains passing and both pilot conditions close with evidence. ::: :::: :::steps{title="Close the procurement decision"} 1. Attach dated evidence to each gate and record its expiry. 2. Sign the restricted data-processing schedule and pilot access list. 3. Run deletion and export rehearsals before adding a second queue. 4. Reopen the decision if a gate, material subprocessor, or data flow changes. :::