frameworkdirectory-build-review-framework · v21099:anthropic-cowork-vetdir12026-08-12served from databaseAll documents

Directory Build-Plan Review Framework v1 (SUPERSEDED by v2)

Directory build-plan review framework v1

Purpose: a reusable instrument for deciding whether a FedM8 build plan that touches veterans, money, and third-party professionals is safe to build. Written to be applied by someone who did not write the plan.

Design principle: a weighted score alone is gameable — a plan can average its way past a fatal flaw. So the instrument has two independent parts, and both must pass:

encode consequences that are not tradeable: criminal exposure, per-violation civil penalties, loss of a platform licence, or a claim to the veteran that cannot be substantiated.

Verdicts: GO (all gates pass, score ≥ 75) · CONDITIONAL GO (all gates pass, score 65–74 — build the unaffected phases, named fixes before the rest) · NO-GO (any gate fails, or score < 65).

Evidence rule. Every gate result and every dimension score must cite a specific line, section or artifact in the plan. A score with no citation is recorded as the minimum for that band. Assertions of intent do not count as evidence; a mechanism does. "We will be careful about X" scores as if X were absent.


Part A — hard gates

#GatePasses whenFails when
G1Unlawful-assistance monetizationNo entity is paid-promoted in a jurisdiction where its fee practice is prohibited, and no revenue line is tied to referring veterans to unlawful fee-chargingAny paid placement, per-lead fee, or ranking boost accrues to an entity charging veterans for pre-decision claim assistance in a prohibiting state
G2Paid-placement disclosureEach paid unit is labeled Sponsored/Ad, unavoidable, visually distinct, above or on the block; no "best/top/objective/independently ranked" framing anywhere; no first-party rating or ordering influenced by paymentAmbiguous labels ("Featured," "Partner," "Top Pick"), disclosure behind a hover or link, or any superlative/objectivity claim over a paid list
G3Third-party data licenceGoogle ratings/reviews/counts are live-served, never persisted; only place_id (indefinite) and lat/long (≤30 days) stored; Maps attribution present; content visually segregated; no composite blendingAny schema column, cache, warehouse table, or export holding a Google rating, review count or review text
G4Substantiated identity claimsEvery badge shown to a veteran is backed by a source that actually returns that attributeThe site says "honorable discharge" while the integration behind it (ID.me, SheerID, VerifyPass, GovX) returns only affiliation
G5Review integritySentiment-blind published moderation criteria applied identically to paying and non-paying firms; no incentivized reviews; insider reviews blocked; every action audit-loggedAny path by which subscription status affects whether, where, or how a review appears
G6Compensation structureNo payment to the platform contingent on retention, engagement, settlement or recovery; referral-service registration held or jurisdiction excluded where requiredPer-lead or per-outcome economics live in a state with runner/capper or referral-registration exposure without counsel sign-off
G7Legal-basis prerequisitesPublic ToS and privacy policy live before any identity integration; breach process documented; PII minimization stated (no SSN, no discharge documents at rest)Identity or payment work scheduled ahead of the legal pages, or any design storing SSNs or DD-214 images
G8Directive complianceClaim→boot→work→verify→usage→status→release; one billable per claim; gated-lane PR/chat-approve respected; documents via /api/1099/documentPost-hoc billing, duplicate sessions, auto-merge to production, or new files under public/framework/

A gate may be recorded PASS, FAIL, or N/A (phase not in scope). N/A is only available where the plan explicitly defers the phase behind a stated gate.


Part B — weighted dimensions

D1 · Legal & regulatory defensibility — 20 pts

Does the plan survive a state AG civil investigative demand? Scored on: correctness of the accreditation/fee model (38 CFR 14.629, 14.636; 38 U.S.C. 5904); handling of the unaccredited-consultant question; per-state variation; awareness that the law is unsettled in both directions; whether risk is engineered out or merely disclosed.

D2 · Veteran outcome & trust — 15 pts

Would a disabled veteran be better off? Scored on: free-VSO path shown before paid options; plain-language accreditation explanation; fee-rule transparency; no dark patterns; the directory answers "who should I choose" rather than "who paid us."

D3 · Monetization integrity — 15 pts

Can the revenue model be described to a veteran without embarrassment? Scored on: separation of payment from ranking quality signals; disclosure quality; absence of outcome-contingent economics; whether the pricing model is defensible in the most restrictive state, not the average one.

D4 · Data & identity architecture — 12 pts

Scored on: PII minimization; correct choice among verification vendors given what each actually returns; schema honesty (no columns that cannot lawfully hold data); RLS design with read/write scope agreement; blast-radius awareness on the estate's most sensitive project.

D5 · Trust & safety / review integrity — 12 pts

Scored against 16 CFR Part 465 clause by clause: fake reviews, incentives, insider reviews, suppression, company-controlled-site representations; plus moderation transparency, appeal path, and audit trail.

D6 · Delivery feasibility — 10 pts

Scored on: fit to the actual estate (Railway fedm8, Supabase oeohvlajtexuxjfrtcdb, GitHub Pages front end, gated lane); phase sizing; dependency ordering; whether external approvals sit on the critical path; whether prerequisites are surfaced or discovered late.

D7 · Directive & ops compliance — 6 pts

Claim loop, one billable per claim, usage + health band, gated-lane deploy path, documents via the API, verification evidence format.

D8 · Commercial viability — 10 pts

Two-sided liquidity: is the veteran side useful before firms subscribe? Seeding strategy and its cost; willingness-to-pay evidence; churn risk if leads are thin; whether the plan states what it does not know about pricing.


Scoring sheet

DimWeightScoreEvidence cited
D1 Legal & regulatory20
D2 Veteran outcome15
D3 Monetization integrity15
D4 Data & identity12
D5 Trust & safety12
D6 Delivery feasibility10
D7 Directive & ops6
D8 Commercial viability10
Total100

Known limitations of this instrument

Stated so the framework itself can be challenged:

  1. It is opinionated toward caution on D1/D3. A reviewer who believes the Platkin and

Louisiana rulings signal that these state laws will fall could justifiably score D1 higher. The framework does not resolve that; it forces the position to be stated.

  1. It does not price the upside. There is no dimension for revenue magnitude or market

timing. A plan that is legally spotless and commercially pointless can score 80.

  1. Gates are US-only. No non-US regime is modeled.
  2. It reviews the plan, not the build. Passing here says nothing about whether the

shipped product matches. G2, G3 and G5 all need re-testing against the live site before launch — the FTC and Google evaluate what renders, not what was specified.

  1. The 75 bar is a judgment call, chosen so that a plan can carry one weak dimension but

not two. If the first three applications of this framework all land at 62–74, the bar is wrong, not the plans.