1) Executive risk map Scope confirmed: 124-icon B2B analytics SaaS library; navigation, dashboards, alerts, exports, collaboration, and account administration. Target rendering is dense responsive web UI at 16 px and 20 px, derived from a 24 px master grid. Declared tokens: 2 px stroke, round caps/joins, 2 px corner radius, 2 px optical padding; fill reserved for selected states. Because the actual icon-name inventory, SVG measurements, screenshots, and duplicate-candidate list are not present in the supplied evidence here, individual-icon defects cannot be asserted. Risks below therefore distinguish confirmed system constraints from verification-required findings. Rank Finding Severity User harm Reach Reversibility Confidence 1 24 px masters may not remain legible when reduced to 16 px Critical if confirmed High Library-wide Medium Medium 2 Semantically adjacent actions may overlap across export, sharing, alerts, and administration High High Multi-family Medium Low until inventory reviewed 3 Selected fill treatment may reduce state distinction if silhouettes differ significantly from outline forms High High Navigation/state icons Medium Medium 4 Duplicate/near-duplicate concepts can create inconsistent component mappings High Medium–High API + library-wide Low–Medium Confirmed as audit concern; instances missing 5 Stroke/corner/optical-padding drift can become visible at 16/20 px Medium Medium Library-wide High Medium 6 Accessibility semantics cannot be established from icon graphics alone High High Interactive icons High High 7 Localization risk exists for culturally dependent metaphors or directional symbols Medium Medium Unknown subset Medium Low 8 Existing component APIs constrain renames/removals High migration risk Indirect Library-wide Low High Controlling assumption: the declared tokens define the intended canonical system unless supplied SVG evidence demonstrates an intentional exception. Where an SVG conflicts with a token, neither should be silently normalized: retain both observed and intended values until the exception is verified. The highest-value two-sprint strategy is therefore repair geometry/state defects without changing public component APIs, then migrate ambiguous or duplicate semantics through aliases/deprecation rather than immediate removal. 2) Evidence-assumption-gap ledger Item Evidence status What can be concluded Gap / contradiction Smallest verification step Library size Confirmed Audit universe contains 124 icons Icon list absent Supply/export icon names Functional coverage Confirmed Six major product areas are represented Icon-to-family mapping absent Map each supplied name to one family Master grid Confirmed Intended master = 24 px Actual SVG viewBoxes/bounds absent Compare each SVG viewBox/bounds to 24 px Stroke Confirmed intended token Canonical stroke = 2 px Observed SVG stroke values absent Extract stroke-width per asset Caps/joins Confirmed intended token Round caps and joins required Observed attributes absent Extract linecap/linejoin values Corner radius Confirmed intended token Canonical radius = 2 px Not every path exposes explicit radius Inspect geometric equivalents on representative icons Optical padding Confirmed intended token Nominal padding = 2 px Actual occupied bounds absent Measure rendered bounds against grid Filled states Confirmed Fill reserved for selected states Actual state pairs absent Compare outline/selected pairs 16/20 px usage Confirmed Small-size behavior is release-critical Raster/render captures absent Render all icons at both sizes Duplicate candidates Confirmed to exist Deduplication is relevant Candidate identities absent Supply candidate pairs/groups Accessibility Missing No accessibility result can be claimed Labels/ARIA/component implementation absent Inspect icon-button/component semantics Platform conventions Missing No conflict can yet be asserted Specific metaphors/screens absent Review screenshots around navigation/actions Localization Missing No affected icon can be named Markets/locales absent Flag directional/textual/cultural metaphors after inventory review Existing APIs Confirmed constraint Renaming/removal creates migration risk Usage frequency unavailable Search component references before changing names Blocked conclusions: exact non-conforming icon counts, exact duplicate groups, pixel-level violations, accessibility pass rates, and individual icon replacement recommendations. Reporting any of those without the supplied assets would invent evidence. 3) Taxonomy and semantic-role audit Until individual names are available, classification can be established at the family schema level but not truthfully completed for all 124 assets. Semantic family Functional role Expected states Primary ambiguity tests Baseline risk Navigation Destination / hierarchy default, hover/focus, selected, disabled where applicable destination vs action; outline vs selected recognition High Dashboards visualization / filtering / configuration default, active/selected, disabled where applicable chart types, filter vs settings, expand vs drill-down Medium–High Alerts status / warning / notification unread/read, active, warning/error where represented bell vs warning vs error vs announcement High Exports outbound action / file operation default, hover/focus, disabled, progress if applicable download vs export vs share vs open-external High Collaboration people / comment / share / assignment default, active, unread where applicable user vs team; comment vs message; share vs invite High Account administration identity / security / configuration default, selected, disabled/destructive where applicable user settings vs system settings; remove vs revoke vs delete High Every individual icon should receive four fields: Role: destination, action, status, object, modifier, disclosure, or utility. Family: one of the six confirmed product areas, with cross-family reuse explicitly recorded. State: base, selected, disabled, destructive, warning, unread/read, or other only when evidenced. Risk: Critical / High / Medium / Low based on consequences of misrecognition. Semantic review rules: One metaphor → one primary meaning. If materially identical glyphs expose unrelated actions, classify High ambiguity. One meaning → one preferred canonical glyph. Alternate glyphs for the same action become duplicate candidates unless context requires differentiation. Export/download/share/open-external must remain visually and semantically separable. Notification, warning, error, and informational status should not rely solely on fill or color differences. User, team, invite, permissions, security, and account settings require particular scrutiny because administrative mistakes have wider consequence. A destination icon reused as an action must be documented or replaced/aliased; visual similarity alone is insufficient justification. Missing evidence: without the 124 icon names and candidate duplicate groups, per-icon taxonomy cannot be completed. The minimum verification artifact is a CSV/export of component name → intended meaning → known usage/location. 4) Visual consistency scorecard Scores below are audit readiness/risk assessments, not claims that the current assets pass or fail. Dimension Target Current assessment Risk Required verification Optical grid 24 px master, 2 px optical padding Unverified High Measure occupied bounds Bounding-box use Consistent optical footprint Unverified High Compare extrema/centering Stroke weight 2 px Intended value confirmed; observed value missing High SVG attribute/path audit Cap/join Round / round Intended confirmed Medium SVG attribute audit Corner logic 2 px Intended confirmed Medium Compare repeated primitives Negative space Balanced at 16/20 px Unverified High Render sheet at target sizes Metaphor clarity Distinct action/object meanings Unverified High Screen-context review State pairs Selected fill only; identity retained Partially specified High Compare every state pair 16 px legibility Required Unverified Critical Pixel render + visual QA 20 px legibility Required Unverified High Pixel render + visual QA Platform convention No documented conflicts supplied Unknown Medium Screenshot/context review Duplicate consistency Known candidates exist Unresolved High Overlay + semantic comparison Token conflict rule When measurements become available, record conflicts explicitly: Property Declared token Observed SVG Controlling assumption Stroke 2 px Missing 2 px remains canonical unless exception is documented Padding 2 px Missing optical rather than purely mathematical alignment controls Radius 2 px Missing repeated equivalent corners should follow canonical logic Grid 24 px Missing master geometry evaluated on 24 px canvas Do not automatically force every extremum to identical coordinates. Circular, diagonal, and asymmetric metaphors may require optical compensation. The acceptance criterion is consistent perceived size and centering, supported by measurement plus rendered comparison. 5) Accessibility and interaction-state audit No accessibility pass/fail result is supportable from the evidence provided. Icon appearance, component semantics, keyboard behavior, focus indication, accessible names, and status announcements are separate concerns. Check Evidence Risk if absent Required result Accessible name for icon-only controls Missing Critical/High Meaning available programmatically Decorative icon suppression Missing Medium Decorative SVG does not duplicate labels Keyboard operability Missing High Interactive controls reachable/operable Focus state Missing High Visible state independent of icon fill Selected state Token partially confirmed High Selected meaning survives without relying only on fill/color Disabled state Missing Medium Visually and semantically distinguishable Warning/error distinction Missing High Meaning not dependent solely on color Tooltip use Missing Medium Tooltip supplements rather than replaces accessible naming Destructive action semantics Missing High Delete/remove/revoke metaphors clearly differentiated State announcement Missing High where dynamic Dynamic changes conveyed by containing component as required State-pair requirement: filled selected icons must preserve enough silhouette/metaphor continuity that users can recognize the same destination before and after selection. If the filled version materially changes internal geometry, treat it as a state-design defect rather than a cosmetic variation. Localization review: test arrows, hand gestures, mailbox/document metaphors, currency-like shapes, abbreviations, and any embedded text only if those icons exist in the inventory. No localization defect is currently confirmed. 6) Prioritized remediation specification Priority Affected family Required change Dependency Owner Verification Path P0 All families Render and inspect complete set at 16/20 px; repair collapsed gaps, merged strokes, clipping, severe imbalance SVG inventory Icon designer Target-size contact sheet + SVG measurement Quick repair P0 Interactive icons Add/fix accessible naming and decorative handling at component level where missing Component implementation FE/design-system engineer Keyboard + accessibility-tree inspection Quick repair P0 Navigation/state pairs Normalize selected-state geometry so outline/fill variants preserve identity State-pair inventory Icon designer + DS engineer Side-by-side default/selected review Controlled migration P1 All Enforce 24 px master, 2 px stroke, round caps/joins, 2 px radius/padding where applicable Measurements Icon designer Automated SVG lint + visual review Quick repair P1 Export/collaboration Separate export, download, share, external-link, invite metaphors where collisions are evidenced Usage screenshots Product designer Contextual comprehension review Controlled migration P1 Alerts/admin Separate notification/warning/error and delete/remove/revoke concepts Semantic inventory Product + icon designer Screen-context review Controlled migration P1 Duplicate candidates Select canonical assets; retain old component names as aliases where possible Candidate list + code usage DS engineer API regression + visual snapshot Controlled migration P2 All Normalize optical scale, centering, negative space and corner construction P0 geometry results Icon designer Overlay/contact-sheet QA Quick repair P2 Cross-platform/contextual icons Resolve convention conflicts only when screenshots demonstrate one Screenshot evidence Product designer Context review Controlled migration P3 Fundamentally ambiguous glyphs Replace metaphor when repair cannot produce distinction Validation evidence Product/design-system lead Comparative recognition/usability check Breaking redesign Two-sprint allocation: Sprint 1 should address P0 plus measurable P1 token violations and create aliases for duplicates. Sprint 2 should resolve evidenced semantic collisions, state-pair inconsistencies, remaining optical defects, and migration documentation. Breaking redesigns should be deferred unless the existing metaphor creates material user risk that cannot be corrected while preserving the API. Where a visual replacement is necessary, preserve the component identifier whenever its semantic contract remains unchanged. 7) QA acceptance checklist with stopping rule Every one of the 124 icons has a recorded name, functional role, semantic family, interaction state, risk level, and evidence status. Every SVG is checked against the 24 px master grid, with measured bounds recorded. Every applicable asset is checked against 2 px stroke, round caps, round joins, 2 px corner logic, and 2 px optical padding. Every token conflict records both declared value and observed value, plus the controlling assumption or documented exception. All icons are rendered and inspected at 16 px and 20 px for clipping, stroke merging, closed negative spaces, loss of metaphor, and optical imbalance. All default/selected pairs preserve recognizable semantic identity; filled treatment appears only where supported by the selected-state rule. Known duplicate candidates are classified as exact duplicate, near-duplicate with same meaning, legitimate variant, or semantically conflicting reuse. Export/download/share/external-link concepts are distinguishable wherever they coexist. Alert/notification/warning/error concepts are distinguishable wherever they coexist. Account, permissions, security, remove, revoke, and delete concepts receive High-risk review where applicable. Icon-only interactive controls are verified at component level for accessible naming, keyboard operation, focus visibility, and correct decorative SVG handling. No accessibility, localization, platform, usability, or measurement result is marked “pass” without supporting evidence. Existing component APIs remain unchanged for quick repairs; controlled migrations use aliases/deprecation before removal. Each High/Critical finding has an owner, remediation path, dependency, and repeatable acceptance test. Snapshot/regression coverage verifies that geometry repairs do not unintentionally change unrelated icons or component mappings. Remaining Medium/Low exceptions are documented rather than silently normalized. Stopping rule: stop the audit only when every material finding has (a) severity, (b) evidence status, (c) affected icon/family, (d) remediation path, and (e) acceptance test, and every Critical/High issue is either verified fixed or explicitly blocked by identified missing evidence. Missing evidence is not a pass: unresolved High/Critical items remain open until the smallest verification step in Section 2 is completed.