frameworkdirectory-build-review-framework-v2 · v11099:anthropic-cowork-vetdir12026-08-12served from databaseAll documents

Directory Build-Plan Review Framework v2 (authority-derived gates)

Directory build-plan review framework v2

Supersedes v1. v1's defect was that four of eight gates were written so the plan passed by construction. v2 fixes that, but an honest version of the fix requires admitting a distinction v2.0 of this document blurred:

gate is failing the law. (G1 fee rules, G4, G5, G6, G7, G8's notice/consent/geofence elements, G6, G9's identity-in-ad-unit rule, G10's editing rule, G13's accuracy elements.)

expensive; the pass condition is the conservative design chosen to avoid litigating it. Reasonable people can disagree, and a reviewer may argue a Type R gate is stricter than the law requires. (G1's listing restriction, G2, G3, G8's no-sale/no-tracker rules, G9's profile-page rule, G11, G13's suppression element, G14.)

G2 and G9 remain circular and the label does not cure it. Their pass conditions restate the plan's own architecture; the authorities establish that the risk is expensive, not that this design defeats it. Vrdolyak in fact blessed sponsored placement on a non-payer's profile page, so G9's prohibition is stricter than the case law requires. Both are kept because the mandate is risk minimization and both are cheap — but a reviewer applying this framework under a growth mandate should feel free to reject them.

Marking the type is the point. A Type R gate is a choice to avoid a fight, and under a different mandate it would be scored differently. Presenting one as if the statute compelled it is how a framework flatters its author.

Objective this instrument now scores against: minimum litigation risk, not maximum viability. The GO bar is therefore 80, raised from 75. A plan that merely survives is not good enough when the stated goal is to stay out of court.

Structure. Part A: fourteen binary gates, each citing authority and typed A/R/I below. Any FAIL blocks the build. Part B: ten weighted dimensions, 100 points, bar 80.

Verdicts: GO (all gates pass, ≥80) · CONDITIONAL GO (all gates pass, 70–79) · NO-GO (any gate fails, or <70).

Evidence rule. Every result cites a specific section of the plan. A mechanism counts; an intention does not. "We will be careful about X" scores as if X were absent.


Part A — hard gates

#GateAuthorityPasses when
G1Accreditation & fee law38 CFR 14.629, 14.636; 38 U.S.C. 5901, 5904Only VA-accredited attorneys, claims agents and VSO representatives are listed as able to assist; fee rules (post-decision, ≤20% presumed) are stated accurately; no revenue accrues from an entity charging unlawfully
G2Referral-service lineCal. B&P §6155 + §6156.5 (SB 37, 2025 — private action, $5,000–$100,000/violation + fees); Jackson v. LegalMatch (2019) 42 Cal.App.5th 760; Tex. Occ. Code ch. 952 + Tex. Ethics Op. 573; Fla. Bar R. 4-7.22; Ohio Gov.Bar R. XVI; RCW 2.48.180Open enrollment (no caps, panels, waitlists, exclusivity); search returns all matching firms by name; the consumer selects a named firm before any contact; the contact form delivers only to that firm; no intake questionnaire is distributed to a filtered set
G3Compensation structureABA Model Rules 5.4(a), 7.2(b); NJ ACPE Op. 741; NY Ops. 1132/1294; Fla. Ethics Op. 18-1; IL RPC 7.2(c)(8); Tex. Penal §38.12; La. R.S. 37:219Flat, retention-independent pricing only. No percentage of fees, no per-retention, per-signed-case, per-outcome or success-based component. No per-lead pricing
G4Paid-placement disclosureFTC Act §5 and the Enforcement Policy Statement on Deceptively Formatted Advertisements (the operative authority); In re LendEDU (FTC 2020); 16 CFR §465.1(c) supplies the enforceable definition of "clear and conspicuous," not the labeling duty itselfEach paid unit labeled Sponsored, unavoidable, visually distinct; no "best/top/objective/screened/vetted/independently ranked" framing; no first-party rating or ordering influenced by payment
G5Review integrity16 CFR Part 465 §§.2, .4, .5, .7 (eff. 21 Oct 2024); FTC Act §5(m)(1)(A) — $53,088/violation (2025 level carried into 2026 by OMB M-26-11)No incentivized reviews; insider reviews blocked; sentiment-blind published moderation applied identically to paying and non-paying firms; review text never edited
G6Third-party data licenceGoogle Maps Platform Service Specific Terms §14.3; Maps attribution policyGoogle ratings/reviews/counts live-served, never persisted anywhere — DB, cache, CDN, logs or export; only place_id (indefinite) and lat/long (≤30 days) stored; attribution present; visually segregated; never blended into a composite
G7Substantiated identity claimsVendor documentation: ID.me community payload; SheerID published FAQ; VA VSHE service_history.readEvery badge shown is backed by a source that actually returns that attribute. No "honorable discharge" claim on an affiliation-only integration
G8Consumer health dataRCW 19.373 (MHMD — private right of action via RCW 19.86.090: treble to $25,000 plus attorney's fees); NV SB 370; Cal. Civ. Code §1798.140(ae), §1798.121; FTC §5 (GoodRx, BetterHelp, Cerebral)Standalone consumer-health-data notice linked from the homepage; separate opt-in to collect and to share; deletion cascading to backups with downstream notification; appeal path naming the AG; no sale or sharing of health data; no third-party trackers on authenticated or review pages; no geofencing near VA medical facilities
G9Right of publicity765 ILCS 1075 (IRPA — $1,000 statutory minimum per violation + fees); Cal. Civ. Code §3344; Ohio R.C. 2741; Ind. Code 32-36-1; Vrdolyak v. Avvo (N.D. Ill. 2016); Lukis v. Whitepages (N.D. Ill. 2020)A listed professional's name, photo or identity never appears inside an ad unit, in copy selling subscriptions, or in a paywalled teaser; no sponsored competitor placement on a non-paying professional's own profile page; no paid tier whose benefit is removing competitors' ads
G10Platform speech posture47 U.S.C. §230 and its limits (Roommates.com, Barnes v. Yahoo!, Lukis); Browne v. Avvo; Davis v. Avvo; state anti-SLAPP / UPEPAReview text never edited or co-authored; the platform's own rating disclosed as opinion with published, pay-neutral methodology; moderation policies drafted as discretionary, not promissory; notice-and-response process exists and is documented
G11Legal-basis prerequisitesVA production-access requirements; MHMD §.020; insurance underwriting normsPublished ToS, general privacy policy, standalone health-data notice, documented breach process, retention/deletion schedule, and media/multimedia E&O bound — all before any identity integration or public listing goes live
G12Directive compliance (Type I — internal, not law)hi.jbnx.io v35claim → boot → work → verify live → usage + health band → status → release; one billable per claim; gated-lane PR then chat approval; documents via /api/1099/document with project_slug
G13Third-party accuracy, defamation & trademarkLanham Act §43(a); 765 ILCS 1075; state defamation law; Browne v. Avvo, Davis v. AvvoEvery factual claim made about a listed professional is sourced and dated; roster staleness downgrades the badge; third-party ratings render only where the underlying listing's ownership is verified; a correction/takedown path exists for the named professional; free suppression on request; no third-party org's logo or mark reproduced beside a paid product
G14Written outside-counsel sign-off(Type R — the plan's own highest-leverage control)A written opinion, scoped to the specific contested question and obtained before the dependent phase ships, exists for each jurisdiction whose exposure is (a) uncapped or class-scale and (b) resting on an unconstrued authority — currently California §6155/§6156.5, Washington MHMD, Illinois IRPA, and Florida Bar R. 4-7.22

A gate is PASS, FAIL, or N/A (phase explicitly deferred behind a stated gate).

Why G13 and G14 exist. An independent review of v2.0 observed that the most likely actual lawsuit — a listed professional suing over something the site says about them — was measured only inside a 7-point dimension and could not fail anything, and that the plan named written counsel review as its highest-leverage control while the framework never required it. Both are now gates.


Part B — weighted dimensions

DimWeightScores
D1 Legal & regulatory defensibility16Accreditation/fee model correctness; per-state variation handled as data not prose; risk engineered out rather than disclosed
D2 Veteran outcome & trust14Free VSO path privileged in layout; fee rules explained at the decision point; no dark patterns; the site answers "who should I choose," not "who paid us"
D3 Monetization integrity12Defensible in the most restrictive state, not the average one; payment structurally separated from ranking, ratings and reviews
D4 Data & identity architecture10PII minimization; vendor choice matches the claim made; schema encodes licence constraints; RLS read/write scope agreement
D5 Review integrity10Every 16 CFR 465 clause has a named mechanism; appeal path; audit trail; "clearly false" defined
D6 Delivery feasibility8Fit to the estate; phase sizing; external approvals off the critical path; hidden work items surfaced
D7 Directive & ops compliance5Claim loop, billing, usage band, gated lane, verification evidence format
D8 Commercial viability8Cold-start liquidity; seeding cost; pricing uncertainty named honestly
D9 Privacy & health data10MHMD/SB 370/CCPA obligations as built artifacts; tracker posture; retention and deletion; consent logging
D10 Third-party accuracy & defamation7Accuracy of claims made about listed professionals; roster staleness; correction and takedown path; insurance

Band guidance — expressed as a fraction of the dimension's weight, so it produces a number rather than a mood. v2.0's bands ("top decile / upper-middle") mapped to no point values, which made any total between roughly 65 and 85 defensible from identical evidence.

BandFraction of weightTest
4 — Verified0.86–1.00Mechanism specified concretely and a named verification test exists in the plan's verification section
3 — Specified0.66–0.85Mechanism specified concretely; verification vague or absent
2 — Policy0.31–0.65Stated as intent or policy; no mechanism
1 — Absent0.00–0.30Missing, wrong, or contradicted elsewhere in the plan

Bands are contiguous by design — v2.0 left unreachable gaps between them.

Round to the nearest whole point. A dimension containing an internal contradiction cannot score above band 2 regardless of how much else is present.

Mandate weighting. D7 (5) and D8 (8) do not measure litigation risk. Under the stated risk-minimizing mandate, report the litigation-risk subtotal out of 87 alongside the total out of 100, and note if the two verdicts diverge. If a plan clears 80 only because of D7 and D8, it has not met the mandate.


Scoring sheet

DimWeightScoreEvidence
D1 Legal & regulatory16
D2 Veteran outcome14
D3 Monetization integrity12
D4 Data & identity10
D5 Review integrity10
D6 Delivery feasibility8
D7 Directive & ops5
D8 Commercial viability8
D9 Privacy & health data10
D10 Accuracy & defamation7
Total100Bar: 80

Known limitations

Stated so the instrument can be challenged rather than trusted.

  1. It cannot deliver "no litigation risk." No instrument can. A platform that publishes

star ratings of named professionals will receive demand letters; the achievable goal is that every claim is defensible, cheap to defeat, and insured. Gates G9 and G10 measure defensibility, not immunity.

  1. It is US-only and models no non-US regime.
  2. It reviews the plan, not the build. G4, G5, G6, G8 and G9 must all be re-tested

against the rendered live site before launch — the FTC, Google and a plaintiff evaluate what renders, not what was specified.

  1. Two authorities cited are young. Cal. §6156.5 (2026) has no construing case law, and

MHMD has no published merits ruling as of August 2026. Gates G2 and G8 are therefore calibrated to the statutory text, which is the conservative reading.

  1. The 80 bar reflects a risk-minimizing mandate. Under a growth mandate, 75 would be

the right number. Changing the mandate should change the bar, explicitly.

  1. No dimension prices the upside. A legally spotless, commercially pointless plan can

score 90.