Skip to content

UI coverage comparison

Temporary planning document; planning only. As the planning brief requires, this records what each older and newer design covers, what the current implementation actually does, what the owner decisions require, and for every surface whether the plan retains, revises or must newly design it. No single asset is adopted wholesale. Prototype behaviour is design evidence, never proof of shipped behaviour. Release names follow the integrated plan §5.

Owner session of 4–5 October 2026. Chris's decisions in the owner session change many of the dispositions below, mostly for the v10 pack (E) and Review Prototype v4 (RP4). The changes are marked in place, with the earlier disposition kept, and §6 compares the v10 and RP4 screens with the owner-session decisions. The full change list for the next design iteration of SyRF Prototype v10 is the design-prototype handoff (OS-A30), which also holds the copy-ready prompt for the design session. Producing that handoff is the deliverable; no design session is messaged and no new prototype is built. The owner ledger stays the authoritative decision overlay for the v10 review, and the consolidation wins over older text. Feature implementation is on hold (owner, 5 October 2026).

1. Assets compared

A read-only sub-agent searched the workspace, all 191 PR worktrees, .worktrees, agents, scratch, handover, syrf.old, other repositories, ~/.codex, ~/.claude uploads, ~/buzz-agents, ~/Downloads, ~/Documents, /tmp, git history on every ref, and the whole filesystem by filename. Review C then found assets the first pass had missed; they are added below and marked. Hashes are SHA-256.

ID Asset Location Date Question-interface coverage Reviewed here
A QM v2 prototype (3,890-line HTML, 032fe56e…a999a) docs/features/question-management/prototype.html on branches of #2461 (draft, conflicting) and #2575 (open); local copy in main/.local/qm-designer-handover/05-prototype/ (gitignored) 2026-03-30 Design tree editor, six fixed category tabs, properties panel, Assign per stage, Preview, reviewer form, a four-step publish wizard (Review Changes → Confirm Impact → Answer Handling → Review & Publish) with a reserved step for conflicting per-question decisions within one session that was never prototyped, Impact & Mapping, per-question and project version history Screen inventory (sub-agent and review C, line references)
A2 QM v2 designer specification pack main/.local/qm-designer-handover/01..06/ (UI specification, publishing/versioning UX, Impact & Mapping brief, Figma index) 2026-04-25 Specifications only Headings via sub-agent
A3 QM v2 Angular scaffolds (code) #2575 branch: project-admin/question-management-v2/** (design, properties, impact mapping, assign, publish wizard, preview) 2026-04-17 As A, never wired to the API Via the QM v2 implementation history; present in #2575's tree (review C)
B1 Redesign prototype v7 (React; components-qm-v6.jsx plus v7 rationale, profiles, stage settings, reconciler) main/docs/features/redesign-prototype/ (merged with #2398; Draft) 2026-05-04 Lifecycle icons (draft / published / published with edits), autosave and draft restore, Impact & Mapping, version timeline and diff, Assign with stage history, a three-step "Publish to stage" wizard, Preview, Rationale tab, Profiles, Stage settings in three layouts, Reconciler (side-by-side / diff / table) Component and heading inventory (me and sub-agent)
B2/B3 Redesign working copies v1–v9 and zip /home/chris/workspace/syrf/prototype/, /home/chris/workspace/syrf/SyRF Redesign v9/ 2026-04-26 / 05-04 Same QM as B1; v8–v9 change only the dashboard Hash comparison (sub-agent)
C #2621 landing-page prototypes, all seven plus README: screening profiles (criteria, rationale questions, version history, "used by", and a "migration policy for stages where existing screening data exists": keep on the old version, re-screen under the new one, or archive the stage and start fresh), stage settings, study review, reconciler, PRISMA workflows (workflow templates, stages in a workflow, PRISMA box mappings, template lineage, diagram export), study state (a study-state inspector with per-profile screening-outcome version chains, the annotation reconciliation chain and a worked FT-v1 → FT-v2 re-screen example) and dashboard (the authenticated landing page across projects) docs/features/landing-page/*.html on #2621 (draft, conflicting) 2026-05-02..05 Profile-owned rationale questions, profile versions and their migration policy; PRISMA workflow mapping; study state The first four by sub-agent section inventory; PRISMA workflows, study state and dashboard found by review C, with headings checked directly; not re-read in full by the plan author
D Codex classification / entity-annotation prototypes ~/.codex/visualizations/2026/09/23/01a0cbec-0c32-7103-ab5a-bfc05665deb7/syrf-entity-annotations.html and about 25 intermediates; snapshot in /tmp/syrf-design-handover-2026-09-28/ 2026-09-24..28 Entity types, numbered question tree (add beneath parent), per-stage question selection, concepts & rules, populations, cohorts Section inventory (sub-agent)
D2 Classification site (local build with git history) ~/.codex/visualizations/2026/09/23/01a0cbec-0c32-7103-ab5a-bfc05665deb7/classification-site/dist/index.html 2026-09-27 Classification concepts Found, not reviewed (found by review C)
E v10 pack (SyRF Prototype v10.dc.html, Review Page/Form v10 (byte-identical), Reconciliation Exploration) Handoff package source/design_handoff_syrf_v10/prototype/, under /home/chris/.codex/visualizations/2026/09/23/01a0cbec-0c32-7103-ab5a-bfc05665deb7/syrf-v10-review-2026-10-02/ 2026-10-02 Questions & entity types (c2), screening profiles with screening questions (r1), annotation forms with versions and publish prompt (r6), stage designer with lifecycle, steps and reconciliation (r2), selection preview (r3), history (r5), outcome schema catalogue (o1) Documents read in full; prototype scene and label inventory (me and sub-agent); navigation code checked directly. Re-hashed 5 October 2026 (sha256sum): SyRF Prototype v10.dc.html 1f940684e0bbd32e4b04b0e48f50ecf2d371c20ade94f51eb1e9c1e188ecdbe6; Review Page v10.dc.html and Review Form v10.dc.html both 29adb6b5cf383e5b273aaac7f8817a0c446c2dd22474b504d5231306c240bdbc; Reconciliation Exploration.dc.html 8daf4e3a860407691c09cf3de0602f35cd6c6429323c0cb0b9b3c7ffe261e31c
RP4 Review Prototype v4 and mockups (the stage-review specification) /home/chris/workspace/syrf/handover/2026-09-21-stage-review-design/design_handoff_stage_review_page/ 2026-09-21 Reviewer page only; no question authoring Handoff README read in full (identical to v10's reference copy, reference/stage-review-page-handoff.md); prototype via sub-agent. Owner session: RP4 stays the reviewer-page reference, revised by U1 labels, the step selector and full phone annotation (§6)
RP1–2 / RB Review Prototype v1/v2 (Dockview); Repeatable Branch Pattern handover/2026-09-08-af2-revision2/…; handover/design-return-repeatable-branch/… 2026-08/09 Reviewer page and branches Listed only
— Current implementation main (78c6d097d, unchanged in these areas at 2949ca3a7): project-admin/question-management/ (new editor), project-admin/annotation-question-designer/ (legacy), stage/review/reconcile/overview/settings, membership, data export — See the inventory Code verification (sub-agent, reviews)

Correction to earlier research: COMPARISON.md line 107 calls Review Prototype v4 "absent". It exists in the 21 September stage-review handover folder above; only the v10 pack omits it.

Not found (not reviewed, and not claimed to be):

  • Review Prototype v5 (referenced by v10 RECONCILIATION.md line 90 and the exploration file).
  • Review Prototype v3.
  • A distinct embedded "Review Form v10" (it is a byte copy of the Review Page).
  • The Figma file "SyRF Design v2" (cloud only).
  • The remote classification site (the local build D2 exists but wasn't reviewed).
  • The "UI mockups for form screens" zip inside the repository (it exists only as Claude uploads).
  • A QM-only prototype on main.
  • A "PR-E" of the QM v2 stack.
  • QM v2 version-diff, mobile and training-round designs (never prototyped).
  • Any question-interface prototype newer than 2 October.
  • Anything design-related in ~/Downloads or ~/Documents.

2. Surface-by-surface comparison

Key: Retain = keep the current implementation or an asset's design; Revise = keep the core but change it to fit the decisions; New = no asset or implementation covers it.

2.1 Question and definition management

Surface Current main Older (A/A3, B, C, D) v10 (E) Decisions Disposition
Question tree and editor New Design/Assign/Preview editor on the legacy API; answered questions locked; same-parent drag only; seven fixed category tabs; legacy editor is the default path A: tree + properties panel, focus view, inline add, broken-conditional marker, parent-answer mapping grid. B: lifecycle icons, autosaved draft snapshots. D: numbering identical to the reviewer form c2: tree per entity type with Settings/Preview, system-locked wording vs configurable wording, semantic roles, exceptional answers FV1, VU1–VU3, DP4 (same editor for profile questions), TC1 Retain the current new editor as the base (move filtered-options authoring in; remove debug text). Making it the default is a PROPOSAL tied to its two modes (legacy API for legacy projects, canonical for admitted projects; C17). Revise: lifecycle indicators from B once publication exists (R2c); numbering identical to the reviewer page (D/v10); semantic-role display for system questions (v10); locks replaced by versions for canonical forms. Admin draft autosave (B/QM v2 D002) is optional and must never be confused with reviewer drafts. Owner session (D2-11, OS-A01): multiple drafts are allowed with one publication at a time; design pages show authorised viewers and editors (avatars, per-question editing markers), live updates that never overwrite local edits, recoverable conflicts and actor and time stamps (New; UX spec §3.8).
Version history and diff None A: per-question and project version history with restore. B: version timeline and diff rows Version list on forms only; no per-question diff FV1–FV4, VU2, EX1 Retain the A/B history and diff (only they cover it). Revise: "restore" creates a new draft from an old version and never rewrites history; show "Why it changed" and "What reviewers need to do differently" per version (R2a, R2c).
Publication and impact None (edits apply immediately) A: four-step wizard with Impact & Mapping and a reserved, unprototyped within-session conflict step; A3/QM v2 M004: impact service, transition service, reviewer version-transition alert. B: three-step "Publish to stage" wizard r6: "Publish v3?" per-question Compatible/Not compatible prompt; incompatible answers held in the pool FV2/FV3 (all prior versions, admin choice), recovered requireReanswer/autoUpdate/doNothing, PS1–PS3, VU3, FV4 Revise into one dialog combining: per-question impact choice (A/QM v2), per-category treatment for completed, saved-incomplete and draft-only sessions across all prior versions and all bound stages (New), usage-evidence freshness state (New, PS), missing-reason warning (VU3), conflicting per-question decisions within one session (New, the step A reserved), and compatibility classification feeding held reconciliation work (v10). Vocabulary mapping: QM v2 "does not affect / may affect" is the impact classification; its "keep" maps to doNothing, "re-answer" to requireReanswer, and "map answers to updated options" is a data transformation that is only an approved form of autoUpdate if Q-34 says so. Reuse the GuardedReviewSettings conflict and "Apply anyway" patterns. Validate as U6 (R2c). Owner session: Q-34 is decided (the authorised publisher explicitly declares compatibility and mapping, immutable from commit; corrections use guided rollback and republication); a standard-target change is a publication with impact even without question changes (OS-A12); publication may write attributable generated session versions (D2-01 decided-amended); the shared active-work impact preview with recheck at commit and a narrow Apply anyway (UX spec §3.9).
Question templates and import None on main; #3934 adds an import dialog in the new Design tab (open, conflicting, flag annotationQuestionImport) None (library sharing in reconciliation D27/D36 is reference-based) None SET1, DP4 (copies only) New design on the #3934/#2781 contracts: template browse, preview, dependency remapping, copy into the project or into a profile (R1a for project questions; R3b for profile templates). Called "question templates", not a library, because Study Management uses "Library" (Q-13). Validate as U19.
Question selection and form composition Assign tab: two trees per stage; stage-settings question-selection page A: Assign per stage with ancestors and pending publish. D: per-stage selection c3 superseded by r6: questions chosen per form ("Included as ancestor", "Required by data extraction"); stages bind forms SF1/SF2, TC1 Revise: compose form versions with the existing two-tree interaction (full answer-set membership, ancestors and repeated-entity dependencies shown), then bind forms to stages (R2a) and steps (R3a). The stage question-selection page becomes legacy-only. Validate as U14.
Preview Preview tab uses AF2 preview mode (v1 fallback) A: published vs pending, simulation, change markers, "Shown because…" — FV1 Retain the AF2-based preview. Revise: published vs pending version comparison, change markers and conditional explanations from A (R2c).
Categories and guidance Seven hard-coded categories; project guidance with per-stage override A/B: same categories c2: entity types with capabilities replace categories; system templates TC1, classification decisions Retain categories through R2 and R3. Revise in C1: entity types with legacy categories as system templates (verify names, E14), with a designed transition from category tabs to entity types. Guidance stays stage presentation text (A-15).
Profile-owned eligibility questions None (criteria are project text) B: Rationale tab (per-profile question set). C: rationale questions with profile versions, "used by", but a cross-profile rationale library (ADR-013) r1: screening questions inside the profile editor, presentation order, decision from answers, agreement and resolution rows DP4, DP3, DP5, RX1 Retain v10 r1's structure (profile editor embedding its own question tree through the shared editor component) plus C's version history and "used by". Revise the "rationale" framing into eligibility questions and reasons. Reject the shared cross-profile rationale library (conflicts with DP4) (R3b, U9, U17).
Screening profiles page Project-level screening settings (method, criteria text, keywords) B: profiles and rationale page. C: list and detail, with a per-stage migration policy for new profile versions r1 DP4, DP5, PV2 Retain v10 r1 + C. Revise C's migration policy (keep on old version / re-screen / archive stage) from per stage to per profile under DP4, with "archive stage" reconciled against LC1 and PV2; the choice itself is Q-26 (R3b). Owner session: Q-26 is decided (profile publication detects affected decisions and version mismatches, shows the impact and confirms the treatment with a final recheck). The profile version also carries Unsure (D4-01), the explicit tie policy with its bound or adjudication step and adjudicator assignment (OS-A13, OS-A16), optional discussion (D4-02), adjudication rationale (Q-32), reconciliation blinding (OS-A02), the exclusion-reason guidance (OS-A17) and external or AI source policies (OS-A20) (Revise/New).
Entity types, templates, concepts and rules None D: entity types, concepts, populations; D2 not reviewed c1/c2 TC1, classification research Retain v10 c1/c2 for C1. Revise: template defaults use verified legacy identities ("Disease Model Induction" vs Chris's "disease-model intervention", E14) (C1).
Outcome schemas and measures AF2 outcome matrix, cell editor, spreadsheet, graph assignment D: outcome workflow intermediates o1 catalogue with an Add-schema dialog; o2 entry (redirected to the reviewer page) OC1, ODIR1, RD16 Retain the current entry pattern (RD16) and the v10 o1 catalogue for project-level schemas. New: project custom-schema authoring beyond the catalogue dialog; series vs observation field roles; and, in the reviewer form, reviewer-created outcome measures with one versioned direction across cohorts in the paper (v10 stores direction nowhere, OD11). Measures are created by reviewers from the paper, never predefined by admins (A-20) (O1, U24).
Owner session: catalogue browser and publication requests None (system templates exist only in plans) — None D2-15 decided-amended (one application-wide CAMARADES catalogue; versioned copies with provenance; publication requests); D4-06 content via T-SI-03 New (R1a, R3b, C1, O1): browse and copy catalogue item versions with copiedFrom provenance, "a newer catalogue version exists" notice, publication-request queue for catalogue administrators (ACD spec §3.4).
Owner session: AI screening model configurations None — None E3 AI expansion (OS-A20): versioned project-level AIScreeningModelConfiguration, permission-aware inspection page New (lane XS1, proposed): a Design page listing model configuration versions, the profile versions and runs that use them, and per-decision provenance, with "not supplied" metadata shown explicitly (RI spec §3.11, RI-R47).
Annotation forms page None — r6: guidance, target, version list, publish prompt, reconciliation settings, questions, extraction toggle SF1/SF2, FV, PS, BL1, RE2, UA1 Retain v10 r6 as the forms page. Revise: the publish prompt becomes the combined dialog above; remove blinding from forms (BL1 is stage-owned) superseded 5 October 2026: keep v10's form-level reconciliation blinding, because the annotation form owns it, blinded by default (OS-A02); remove "confirm each" (RE2), with bulk approve pending Q-11 as an optional per-form setting, off by default, with explicit reconciler confirmation of exact inputs (Q-11 decided); matching weights shown as read-only defaults (A-12); the extraction toggle appears only once O1 exists; add the gold-completeness setting if Q-04 approves (Q-04 decided: requiredness with a form override) (R2a, R4a, U14). Owner session additions: the standard target in the form version (OS-A12) with Study-specific override history; targetOneHandling (automatic Single annotator acceptance or required human reconciliation, OS-A28); the reconciled-hints baseline (OS-A04); the outdated-answer completion mode (D4-17); capacity cap, inactivity timeout and per-reviewer in-progress limit as form operational settings (D2-07, D3-17, D3-18); optional unstarted-assignment expiry, off by default (Q-30).

2.2 Stages, steps and project administration

Surface Current main Older v10 Decisions Disposition
Stage designer and settings Settings page: Enable, review settings (with guarded conflict prompts), mock study filters, allocation, question selection, upload, mock permissions; 3-step create dialog; Enable is the only lifecycle control B: stage settings in one-page/tabbed/wizard layouts (profile, activities, resolution, rationale reconciliation, filter set, correction policy). C: stage configuration r2: stage list, lifecycle (complete/reopen, automatic/manual), Setup tab with the review-steps editor (dependencies separate from order, terminal, compulsory, collective satisfaction, target shown from the form, existing-work policy), Overview tab, Reconciliation PV2, DP6, DP7, EW1, VS1, BL1, RX2/LC1, RA1 Retain v10 r2's steps editor and lifecycle controls, in a tabbed layout (B's tabbed variant, as v10 does). New: incoming cross-stage policy control on the destination stage (Collective Include default, own Include advanced), versioned settings publish with impact preview, EW1 stage default with step override, VS1 step policy, BL1 stage setting, assignment expiry default, extra-vote admission (Allow/Stop). Strict within-stage mode only if Q-01 approves. Revise: keep GuardedReviewSettings prompts; replace the mock filters (amended FEAT-008) and the mock permissions page (Members & groups). v10's step-level "who reconciles, per part" is replaced by stage grants (contracts, C9) (R3a, R3c, U12). Owner session, 5 October 2026 (Q-15 replaced, Q-01, Q-02, Q-28, OS-A05, OS-A27): the struck items are superseded. New: a stage study filter editor (clauses on screening-profile outcomes and reconciled answers in AND/OR groups, with a feedback notice when a clause reads an activity bound in the same stage) defines the pool, replacing any cross-stage route control; steps keep in-stage dependencies with own-Include progression by default and a decided stage setting requiring collective Include; exclusion-stop per step; "Finishing work already started" settings (saved work after exclusion, continuation after pool departure; placement SP-AMB-01); a route capacity cap stricter than the form baseline; an optional reviewer-pool browsing setting; system adjudication and training step kinds. BL1 and VS1 move to the form or profile (VS1 as a hint baseline that a step may only hide); assignment expiry is an optional form setting, off by default. v10's "collective satisfaction" maps to the Q-01 setting and its "existing-work policy" to the continuation settings. Lifecycle follows Q-02.
Study selection preview / "Who is offered what" None — r3 (designer view explains absent work) and the Monitor entry C6 admission; OD5 Retain r3 as an admin/designer view backed by the real admission service and its reasons, behind its own capability because it can reveal personal votes; never shown to reviewers (R3a). Owner session (OS-A07, OS-A08, OS-A11): pool-level counts by refusal reason until X-AUTH-RESOLVER (D3-13 brief item); add the admin eligibility explanation timeline below.
Stage overview / monitor Stage overview with materialized-statistics switch-overs, allocation progress, member progress dialogs — Overview tab: progress by step, reviewers, activity C7, C17 Retain the current overview and route SignalStores. Revise: bound forms, targets, steps, gates, readiness and batch frontier from form-unique derivations; no stage-wide presence (not supported). Coordinate with #2469's conflicting chart refactor (R2b, R3a). Owner session: active-reviewer counts for everyone and names only for Monitor holders, never across reconciliation blinding (D3-20 brief item); "capacity not enforced" shown where tracking is off (SP spec §3.11).
Project overview and setup Stages list repeats project-wide criteria under every stage; 8-task checklist in the navigation footer (first stage only, legacy links); CreateProjectWizard (basic details, criteria) C: dashboard is the authenticated cross-project landing page, not a project overview Overview with setup checklist and ownership table SET1, SET2, 21 Sep "Project setup stays" Revise: project overview shows study, form and profile populations separately and stops repeating criteria; the checklist model is replaced with profile/form/step readiness while keeping its navigation-footer placement and its done/next/to-do/locked interaction. New: the guided setup wizard (no asset designs it) (R3d, U11). C's dashboard is reference material for a later landing page, outside this plan's scope. Owner session: the overview carries the pending-admin-changes banner (D3-07); the full global landing page stays deferred.
Members & groups and permissions Administrator/Reviewer only; broken unused CreateProjectGroupComponent; working group×activity PermissionsDialogComponent used only for chart visibility; mock stage permissions QM v2 PGRP; #2224 incomplete frontend; authorization programme WP9 (explanations) and WP11 (generalised dialog) planned Not designed (RD8; RECONCILIATION §8) PM1, PM2, Q-03 New design (U8), delivered with the authorization programme: read-only visibility first (R1b), with "why can or can't I" explanations once its WP9 lands; the generalised dialog and group CRUD after its gates (R1c, its WP11); delegation (R1d). ChangeOwner never appears as grantable; AssignPermissions only within R1d's envelope. Owner session (OS-A24): the membership-disable workflow offers optional contribution exclusion (see the exclusion row below); disabling alone keeps contributions counted (D4-20).
Owner session: eligibility explanation timeline Study history shows no pool history — r3 explains absent work for the current state only OS-A07, OS-A08, OS-A11 (structured history; admin explanation over time) New (SP history release; UI flag stageEligibilityTimeline, PROPOSAL): an admin "Why?" timeline of pool entries, departures, review starts, configuration changes and continuation with the versions in force at the time; "No review is recorded" and tracking gaps shown honestly; a reviewer-facing "Why isn't this available to me?" with own reasons only (UX spec §3.7).
Owner session: active-work impact preview GuardedReviewSettings conflict prompts with "Apply anyway" (eligibility D6) QM v2 impact step r6 publish prompt (questions only) Consolidation §6; Q-24; Q-27; D3-20 Revise the guarded-settings pattern into one shared preview for every admin action affecting reviewers: what changes, who is affected (counts; names only with Monitor), what is kept, which temporary reservations are released, dependent results, notices; recheck at commit; Apply anyway only for temporary reservations (UX spec §3.9; ACD spec §3.10).
Owner session: duplicate review queue, merge and unmerge wizard None FEAT-012 specification only (no prototype) None D2-12 replaced (OS-A29) New (P2a, P2b, P2c): preview with active-work warnings; reconciliation-like conflict resolution of same-reviewer sessions, decisions and accepted results; delegation to the original reviewer as a task; target confirmation; atomic commit; unmerge with per-item carry-forward choices; bibliographic field overrides with provenance (DM spec; validates as U41).
Owner session: contribution exclusion dialog None — None D4-20 decided-amended; OS-A24 New (form scope R2a, stage scope R3a, profile scope R3b): choose form, profile, stage or project; reason required; impact preview including dependent results; mixed route provenance needs an explicit choice; lifting is a separate action (PROPOSAL) (ACD spec §3.2).
Owner session: Deleted projects view and restoration Project deletion route fails closed — None D3-12 decided-amended; OS-A25 New (X-DEL): a restricted list of deleted projects with deletion records; restore with reason and impact preview; restoration rechecks capacity and never revives stale reservations (ACD spec §3.3).
Owner session: training step setup, attempts and assessment None QM v2 "training rounds" brief (never prototyped) None D4-04 decided-amended; OS-A18 New (lane TR1): reference versions, scoring rubrics and pass criteria, manual assessment queue, feedback, retries and optional group admission with the admission record shown (TI spec).
Owner session: inference beta opt-in None D (classification prototypes) c1/c2 concepts and rules Q-18, Q-19 decided; OS-A19 (beta, off by default, explicit project-designer opt-in) New (lane C2): an opt-in screen explaining the beta and its implications, audited enable and disable, inference panels on the reviewer's own snapshot or a pinned accepted result (TI spec §3.7 to §3.9).
Owner session: baseline conversion wizard and findings Project settings › Workflow version (planned panel) — None R4 (OS-A14, OS-A15); Q-05, Q-21, D4-16 New (R6 waves): explicit, reviewable mappings of legacy stages to forms, profiles and steps; findings and blockers in My work; later in-engine redesign through versioned publication (BC spec).
Owner session: external screening import and acceptance CSV screening columns only (upload-search.md) — None E3 AI expansion (OS-A20) New (lane XS1, proposed): preview, validate and accept runs under the profile's source policy; unmatched rows reported; acceptance with impact preview (RI spec §3.13).

2.3 Reviewer and reconciler surfaces

Surface Current main Older v10 Decisions Disposition
Reviewer page AF2 (default off) with the redesigned shell and Dockview (flagged); legacy grid default; equal-weight decision card; no autosave; no Complete anyway RP4 (Review Prototype v4): three cards, corrections 1–11, efficiency features (the agreed specification). B: study review with an activities switcher p1: one card per step with a task strip; r4: one form area driven by the selected step SL1–SL3, DP3, EW1, VS1/VS2, RC6, RC8 Retain AF2, Dockview and v4 parity (owned by the AF2 programme). Revise: step strip plus one form area (Q-12, U7); the screening card gains profile eligibility questions and derived-decision reasoning, which needs AF2 admission for screening steps or a dedicated renderer (F5, U17); draft/saved copy must distinguish an autosaved draft from an explicit Save (v4's "All changes saved" and v10's "Save draft" would now mislead); history panel (U15); Needs updating; outdated flags and Fix; population chip (U3). New panels and step kinds need the Dockview layout-contract amendment (C17) (R2a–R3b, C1). Owner session: Q-12 is decided (step selector with one form area; the selector shows adjudication steps only to eligible adjudicators and training steps where configured); status shows Draft auto-saved, Version checkpoint saved and Complete as distinct states, with illustrative wording (D3-03, OS-A21); full screening and annotation on phones, tablets and desktops at launch, with the largest project as an acceptance case (D3-05, D2-16 replaced); labelled reconciled-answer hints with explicit click-to-fill (OS-A04); Unsure where the profile enables it (D4-01).
Destructive reviewer actions Hard session delete (reviewer endpoint); v4's "Remove all annotations…" menu item — — SL2, GS1 Revise: no hard delete for canonical sessions; replace with an explicit versioned clear or a draft discard, and an append-only withdrawal (R2a).
Reviewer history and correction None C: study-state inspector shows per-profile outcome version chains r5: submissions, correction drafts, pinned revisions, pending updates, surplus assessments SL3, SF5/SF6, DP2 Retain r5's structure; use C's study-state chains as reference for an admin/support inspector. Revise to SL3 semantics. New: detail for the Fix flow (U4) and DP2 correction (U18) (R2a, R2d, R3b).
Reviewer progress lists My studies, incomplete studies, the no-work page (#2412), StageReviewerProgressStore, all keyed by stage — — SF1/SF2 Revise: a session saved through stage A shows consistently in stage B without double counting (R2b). Owner session (D3-07): project My work, a cross-project tab and a global badge that work with every notification muted or off (New; UX spec §3.5).
Owner session: reviewer-pool browse list and Saved work My studies navigator (stage-keyed) — None OS-A27 (optional, dynamic, blinded browsing within a stage); OS-A05 (finishing started work after departure) New (PROPOSAL: stage setting, off by default): the reviewer's own pool with their own status per step and availability reasons, never other reviewers' decisions, answers, identities or timing; a separate Saved work list for permitted continuation (SP spec §3.6, §3.9).
Reconciliation pool None (route only, no navigation) B: reconciler lists (rationale / extraction / re-screen). C: reconciler r2 Reconcile page: KPI cards, Start reconciling (random, assigned first), open a specific study / Assign, "Assigned by hand" with Release, Held and Corrections banners, step table, bulk approve, Agreement tab RA1–RA5, RE4, AG1, Q-11, FV (held when incompatible) Retain v10's pool structure. Revise: one Reconcile entry per study × form task, even when several stages reach it (RE4), plus a profile part (RX1, R4p); unstarted-only expiry and release/reacquire (RA3/RA4); a Request an additional review action (RA5); Agreement gets its own navigation place and capability rather than living only inside Reconcile (AG1); bulk approve only if Q-11 approves; random start and assigned-first ordering are PROPOSAL. New: a queries queue (R4b). Validate as U20 (R4a). Owner session: expiry of unstarted assignments is optional and off by default (Q-30); an additional-review request raises the effective Study and form target and may name people or groups (consolidation §1); bulk acceptance is an optional per-form setting (Q-11 decided); target-one forms that require human reconciliation appear as one-candidate tasks (OS-A28).
Reconciliation workspace Read-only candidate cards (any number, two per row); no reconciler form; AF2 reconcile host read-only by rule B: two-reviewer side-by-side / diff / table. C: reconciler 4a form: reviewer pills, agreement icon, dashed-outline prefill with per-field CONFIRM, Unresolved only, N to resolve, 4-column side-by-side; 3a matching: aligned pairs, own trees, drag/pick, scoring panel, "Suggest again" discards pairings; 4.5 outcome matrix and two-sided dialog; 4.1 screening decision first SF4/RE3, RE2, RE5, NT1, BL1, MG1, RE1, UA1, VS2 Retain the 4a form-based approach (RD17 recommendation) on AF2 controls, which needs a new editable reconcile host (F4). Revise: pills and side-by-side scale to N candidates with overflow or a candidate selector, never fixed columns, and a selector never hides a disagreeing candidate (U1, New); remove per-field CONFIRM and keep the dashed outline as an autofill marker with provenance (RE2); add exposure tracking with the unseen-control warning and Complete anyway (New); matching groups span all candidates, drag-pairing has a keyboard alternative, and "Suggest again" keeps prior pairings in history (COMPARISON F7, PROPOSAL); the screening part shows must-agree answers and reason reconciliation (RX1/DP5, R4p); outcome reconciliation compares N series (R4c, after AF2 Phase 4 PR 9). Owner session (OS-A02, OS-A03, OS-A28): blinded by default under the form's or profile's setting; random candidate order; context-local labels with no continuity across Studies; no submission times or other chronology clues; a one-candidate layout for required human reconciliation at target one.
Owner session: screening adjudication task None — 4.1 screening decision first OS-A13 (tie adjudication, member or group assignment); D4-02 discussion; Q-32 rationale; OS-A16 bounded Unsure Revise v10's screening-part pattern into an adjudication task in the system adjudication step: tied decision versions under the profile's blinding, optional discussion history, required rationale where the profile asks for it, the individual resolver recorded (RS spec §3.9, §5.11).
Gold history and queries None C: annotation reconciliation chain in the study-state inspector 4.4 gold-history timeline with pinned inputs and "Inputs changed · re-check" GS1, QY1–QY9 Retain 4.4's timeline (GS1); owner session: each entry shows its AcceptedResultVersion authority (Single annotator, Human reconciled, Adjudicated, Merge resolved) and keeps flagged results current until explicit reconsideration (RS-R58). v10's "EDIT RECONCILIATION reopens it" is replaced by the query or new-snapshot route. New: query UI (raise from an accepted answer, one work item per answer version, per-concern outcomes, pending flag, private resolution notices, "my concerns and outcomes") (R4b, U21).
Agreement view None QM v2 RECON-13..15 Agreement tab inside Reconcile (percent agreement, κ, compared counts, "Revise in next version") AG1–AG3, VS2, Q-04, Q-16 Retain v10's content. Revise: separate capability with its own navigation place; independent vs informed split; missing-state reporting; compatible-version flags; κ only after method approval (R5c, U29). Owner session: methods come from specialist input T-SI-02 (Q-16, D4-12); external human and machine sources are separate classes (RI-R46); figures come from the separate rebuildable agreement store (D3-11 brief item).
Data export and history Export page: formats, blinding level open to any ExportData holder ("UNMASK DATA"), "completed sessions only" (forwarded but not honoured by writers) QM v2 EXP-* Data export navigation only EX1, EX2, BL1, VS1 Retain the export page. Revise: unmasking follows the export disclosure contract (who may unmask, candidate vs gold separation, BL1, audit; U26). New: Current / Previous versions (R2a) / As of date (R5a) selector, manifest download, coverage labels (U22). The "completed sessions only" fix ships behind a flag. Owner session: lane X1's comparison-level export, codebook and RIS (D4-09 decided); estimated-from-graph provenance (D4-10); machine-source columns (OS-A20); excluded contributions omitted from current exports unless an exporter opts in (OS-A24, PROPOSAL).
PRISMA views None C: PRISMA workflows prototype (workflow templates, stage and profile mapping to boxes, diagram export) — PR1, FEAT-011 Revise C's mapping into report views over C12: profile-to-phase mapping per project (R3b), frozen report snapshots and coverage (R5b). Reconcile its box mappings with FEAT-011 amendments A–J; several versioned workflows per project would need a separate decision (U23). Owner session: views label imported references, source documents and Study items; show report identity, reason, retrieval and pool-tracking coverage and the machine-assisted share; prioritise actual review through a stage, with pool history as separate audit data (OS-A09); exact box mapping waits for T-SI-05.
Notifications, inbox, study attention Not on main; live banners only — — 3 October scope addition Retain the notification stack's behaviour (owned by its programme); its inbox and preferences screens meet UI-1 to UI-11 before testers see them, and its conversation, issue and PDF screens before production (NS-10, pending D3-01), through a Material 3 inbox redesign PR before R2c with the flood controls in UX strategy §7.9. Review-workflow kinds are added per C15, with feature-owned queues as the source of truth; inbox rendering (action labels, deep links respecting task identity and aliases) is validated jointly (U25, U37); #3944 per Q-10 (notifications integration). Owner session: D3-01 is a brief item; New project notification-content settings (D3-22) and per-project email mute (D3-24); resolved notices leave unread counts and stay in history (D3-23 brief item); delivery remains unauthorised (notifications integration §7).

3. Retained, revised and uncovered: summary lists

Retained (kept largely as designed or implemented):

  • The current new question editor as the base.
  • Version history and diff from QM v2 (A) and redesign v7 (B).
  • AF2-based preview.
  • v10 profile editor (r1), forms page (r6), steps editor and lifecycle (r2), selection preview (r3), history structure (r5), pool structure, 4a form-based reconciliation, 4.4 gold timeline, agreement content, entity types and concepts (c1/c2) and the outcome catalogue (o1).

  • 2621's profile version history and "used by" (C).

  • The shipped outcome entry pattern (RD16).
  • AF2, Dockview and Review Prototype v4 parity.
  • The current stage overview.
  • GuardedReviewSettings prompts and the setup checklist's interaction pattern and placement.
  • The notification stack UI.
  • Owner session: v10's form-level reconciliation blinding on the forms page (r6), retained because the annotation form owns that blinding (OS-A02).

Revised (core kept, changed to fit the decisions):

  • Publication dialog: all prior versions, three session categories, freshness gate, warnings, within-session conflicts, vocabulary mapped to the recovered transitions.
  • Question selection moves from per stage to per form.
  • Reviewer page: steps, drafts vs saves, eligibility reasoning, Needs updating, Fix, no hard delete. Owner session: step selector with one form area (Q-12); Draft auto-saved, Version checkpoint saved and Complete (OS-A21); full phone annotation (D3-05); hints with click-to-fill (OS-A04).
  • Reconciliation workspace: N candidates, no per-field confirmation, exposure warning, editable host. Owner session: blinded by default with random order and context-local labels; a one-candidate layout (OS-A02, OS-A03, OS-A28).
  • Matching: N-way groups, history-preserving re-suggestion, keyboard pairing.
  • Forms page: no blinding or confirm-each settings no confirm-each setting; owner session: form-owned blinding, standard target, target-one handling, hints baseline, outdated-answer mode, capacity, timeout and in-progress limit.
  • Profiles: eligibility rather than rationale framing; no shared library; migration policy per profile. Owner session: Unsure, tie policy and adjudication, discussion, rationale, blinding and source policies in the profile version.
  • Stage settings: tabbed, versioned, DP7 control stage study filter editor (Q-15 replaced); stage grants instead of "who reconciles"; owner session: continuation settings, route cap, browsing setting.
  • Owner session: guarded-settings prompts generalised into the shared active-work impact preview.
  • Project overview: separate populations.
  • Lifecycle indicators added to the question tree.
  • Export page: disclosure contract for unmasking.
  • PRISMA workflow mapping (C) reconciled with FEAT-011 and C12.

Gaps that need design before their release (each with its validation):

  1. Question templates and import UI (R1a, U19; R3b for profile templates).
  2. Publication impact for the three session categories across all prior versions and bound stages, with statistics freshness (R2c, U6).
  3. Profile-version publication impact on decisions: partly designed by #2621's per-stage migration policy and its study-state re-screen example, which must become per profile (R3b, Q-26). Owner session: Q-26 decided (mismatch detection, impact, confirmed treatment).
  4. The incoming cross-stage route policy control (DP7) and its effective-policy preview (R3a, U12). Superseded 5 October 2026 (Q-15 replaced): the stage study filter editor with its AND/OR clauses, the in-stage step dependency graph and its effective-policy preview (R3a, U12).
  5. Members & groups with scoped grants and delegation (R1b–R1d, U8).
  6. The guided setup wizard (R3d, U11).
  7. The completed-stage change-approval dialog (R3c, U10).
  8. Reconciliation with three or more candidates across answers, matching and outcomes (R4a, U1).
  9. Exposure-based unseen-control warning with Complete anyway (R4a, U4).
  10. Assignment expiry, release and reacquire controls, and Request an additional review (R4a, U20). Owner session: expiry optional and off by default (Q-30); the request raises the effective target and may name people or groups (consolidation §1).
  11. The accepted-answer query UI and queue (R4b, U21).
  12. As-of export with manifests and coverage labels (R5a, U22).
  13. PRISMA report views: partly designed by #2621's PRISMA workflows prototype (R5b, U23).
  14. Outcome-measure creation with versioned direction, and custom schema authoring (O1, U24).
  15. Notification item views for review-workflow events (per release, U25).
  16. Reviewer states: autosave vs Save vs Complete vs current version (R2a, U13); forms page and stage binding (R2a, U14); history panel (R2a, U15). Owner session: the distinction is fixed as Draft auto-saved, Version checkpoint saved and Complete, with illustrative wording (OS-A21).
  17. VS1 accepted-answer display with exposure capture (R4a, U16). Owner session: labelled reconciled-answer hints shown by default from the form baseline, with explicit click-to-fill and recorded exposure and adoption; no automatic prefilling (OS-A04).
  18. Profile eligibility questions and decision reasoning on the screening renderer (R3b, U17); DP2 correction from history (R3b, U18).
  19. Export disclosure (R2a, U26); coexistence navigation (R2a/GA, U27); pilot rollback to read-only (R2a, U28).
  20. The agreement view's own capability-gated place and its independent/informed split (R5c, U29).

Gaps added by the owner session (5 October 2026). None of these has a prototype. Each is specified in the named specification and listed for the design iteration in the design-prototype handoff.

  1. Full screening and annotation of the largest form on a 390 px phone and a 768 px tablet, including the outcome-data grid, Jump to the next required answer and the action bar above the on-screen keyboard (D3-05; UX spec §3.3).
  2. The stage study filter editor and the in-stage step graph with exclusion-stop and continuation settings (Q-15 replaced, Q-01, Q-28, OS-A05; SP spec).
  3. The admin eligibility explanation timeline and the reviewer "Why isn't this available to me?" view (OS-A08; UX spec §3.7).
  4. Presence, live updates and conflicts on question-design pages (OS-A01; UX spec §3.8).
  5. The shared active-work impact preview with recheck at commit and Apply anyway (consolidation §6; UX spec §3.9).
  6. Reviewer-pool browsing and the Saved work list (OS-A27, OS-A05; SP spec §3.6, §3.9).
  7. The duplicate merge and unmerge wizard with conflict delegation and carry-forward choices (OS-A29; DM spec; U41).
  8. The screening adjudication task, discussion and bounded Unsure handling (OS-A13, OS-A16; RS spec).
  9. Training setup, attempts, assessment and admission (OS-A18; TI spec).
  10. The inference beta opt-in and panels (OS-A19; TI spec).
  11. The Deleted projects view, restoration and the contribution exclusion dialog (OS-A24, OS-A25; ACD spec).
  12. The catalogue browser and publication-request queue (D2-15; ACD spec §3.4).
  13. AI screening model configurations, external import acceptance and machine-assisted labels (OS-A20; RI spec).
  14. Project notification-content settings and per-project email mute (D3-22, D3-24; ACD spec).
  15. The baseline conversion wizard and its findings (OS-A14, OS-A15; BC spec).
  16. My work, the cross-project tab and the global badge (D3-07; UX spec §3.5).

Recommended navigation (Q-13). Correction: v10's default navigation mode is "steps", whose Setup section runs 1 · Concepts & rules, 2 · Entity types & questions, 3 · Outcome data schemas, 4 · Annotation forms, then profiles and stages, followed by Monitor and Data. "Library" is v10's alternate navigation mode (checked in the v10 prototype's navigation code). This proposal follows the alternate mode's grouping, with stages kept separate, and doesn't reuse its "Library" label:

  • Overview: project dashboard with setup progress. The Project setup checklist stays in the navigation footer (21 September decision), with readiness-based content.
  • My work (PROPOSAL, pending D3-07): the reviewer's and reconciler's landing inside a project, listing every actionable item by role with the four feature-owned queues as filters and the inbox as history; a flag-independent global badge, a cross-project "My work" tab beside "Pending Projects" and an admin banner for pending change requests (R3c, R4a, R4b; UX strategy §3.1). Owner session: D3-07 is Decided (U3); the badge counts work even when notifications are muted; rows add Saved work, adjudication for your group, merge conflicts, training, external runs and conversion findings (UX spec §3.5).
  • Workflow version: a badge in the overview and rail and a panel in Project settings (admitted or classic, read-only containment, who and when; the admission action) (R0, R2a).
  • Review: per-stage review entry.
  • Reconcile: one entry per study × form task, with the profile part alongside, so a shared form's task isn't listed under several stages (RE4).
  • Design: Questions, Entity types (C1), Concepts & rules (C1), Outcome schemas (O1), Forms, Screening profiles. The reusable collection is "question templates". Owner session: AI screening models (lane XS1, proposed); copies from the CAMARADES catalogue carry provenance (D2-15).
  • Stages: per stage: Overview, Steps & settings, Monitor ("Who is offered what", behind its capability). Owner session: Steps & settings include the stage study filter; Monitor links to the eligibility explanation timeline.
  • Members & groups: Members, Groups, Permissions.
  • Data: Export (current, previous versions, as of), History, Agreement (behind AG1), PRISMA.
  • Project settings: general project settings keep their place. Owner session: notification content settings (D3-22) and the inference beta opt-in (OS-A19).
  • Study Management keeps "Library" (/studies), along with its Searches and Processing pages. Owner session: the duplicate review queue and the merge and unmerge wizard live here (OS-A29).
  • Owner session: Deleted projects is a restricted view reached from the project index for the owner, Delete holders and application administrators (PROPOSAL, AC ambiguity B4); the catalogue browser and request queue are application-level pages for catalogue administrators (D2-15).
  • Coexistence: legacy projects keep today's Studies and Screening sections until adopted; the navigation branches per project mode (C17, U27). Owner session: coexistence is temporary; every project converts to the faithful baseline after pilots (OS-A15), and until then unconverted projects get the refreshed design and every compatible new feature (OS-A23).
  • Join: the M3 navigation owner, who rebuilt the rail and checklist footer in September, is an L16 join for every IA change.

5. Cross-cutting UI requirements

  • Material 3 for every new and updated screen (Chris, 3 October, UI1): consistent, modern and built with Material 3 under FEAT-023's semantic contract, so it renders correctly before and after that programme's atomic cutover. Criteria UI-1 to UI-11 are in acceptance criteria §3; the sequencing question is D3-01. Older assets (QM v2, redesign v7, #2621, v10) are design inputs; where their styling predates Material 3, the retained structure is rebuilt with Material 3 components and roles, in sentence case (D3-04).
  • Design system of record and pattern inventory: FEAT-023's emitted token contract plus the shared Angular components; every new pattern is specified once (states, tokens, keyboard model, narrow behaviour, copy keys), built once and shown on a flagged gallery route, with a handoff on the UX strategy's template (UX strategy §4).
  • Width matrix from break-points.ts plus 320, 390 and 925 px, 200% zoom, 400% reflow and a short-height layout (acceptance criteria §3.1); touch targets at least 44 px below 600 px; phone screening for title and abstract steps pending D3-05. Owner session: D3-05 is decided-amended; phones, tablets and desktops support screening and full annotation at launch, with the largest project as an acceptance case. A PWA is an exploration beyond the MVP with no offline writes (OS-A22).
  • Accessibility for every new surface: the harness and screen-reader matrix in UX strategy §8 (axe in journey states, keyboard alternatives for drag-pairing and the step strip, screen-reader meaning for candidate pills, agreement icons and the step strip, contrast of the autofill outline in both themes, forced colours, reflow); Firefox and WebKit smoke journeys and CDK pointer-based drag, never native HTML5 drag (D3-15).
  • Reviewer efficiency: the keyboard screening path (decide, next, skip; at most two actions per decision) and live completeness are in R3a/R3b scope; input latency under autosave is budgeted (AC-UX-03, AC-UX-04).
  • Copy deck (C17) before any R2a copy: typed message constants, a guard spec and user-guide glossary parity; terms pending D3-03. Owner session: D3-03 and D3-04 are decided (U1, with illustrative draft and checkpoint labels).
  • Coexistence chrome: the minimum shared chrome identical in both project modes, the workflow version badge and panel, the legacy explainer, and the legacy restyle under D3-06 (UX strategy §3.3 and §3.4). Owner session: D3-06 is decided-amended: restyle plus as many compatible new features as possible, with workflow semantics unchanged until a confirmed conversion (OS-A23).
  • Narrow screens and performance budgets for the reconciliation workspace (candidate selector plus single column below lt_md; a selector never hides a disagreeing candidate) and the stage designer; AF2's bounded rendering assumes one form instance, so N-candidate reconciliation needs its own budget.
  • Today's screening-reconciliation path and how R4p retires it are documented before F4.
  • Pilot operations UX: the workflow version panel is the admit or remove screen; the read-only containment banner is what reviewers see after a rollback (U28, U32).
  • Help and change communication: at most three reviewer-visible change bundles before GA; the "What changed" panel from R2a; the anchored tour from R3a; contextual help on every new screen through the user-guide URL pipe; user-guide pages drafted under target markers.
  • Design QA: tier 1 per PR by agents (theme guards, screenshot matrix, axe, keyboard transcript, copy-deck compliance, a second agent against the handoff); tier 2 by Chris per release on staging, plus per-PR acceptance for new shared patterns and five high-risk surfaces (pending D3-02). Owner session: D3-02 is a brief item; the proposed additions to the high-risk list are the merge and unmerge wizard, import acceptance, the eligibility explanation, large-form annotation on phones and the impact preview (PROPOSAL).
  • Research with real users: a baseline study before the R2a build, formative sessions per freeze gate, summative sessions per release, pilot diaries and a terminology card sort (UX strategy §9); the tester panel's shape was approved under D1-06 on 3 October and Chris named the panel that night (decision register §1.14); no external SyRF users are named yet, and external participants and telemetry are pending D3-08. Owner session: D3-08 is decided (U3): the agreed tester panel with no extra external recruitment now, and consent-based timing telemetry without answer content.

6. v10 and Review Prototype v4 versus the owner-session decisions (5 October 2026)

This section compares what the original prototypes show with the decisions of 4–5 October. The prototypes are SyRF Prototype v10 (asset E, hashes in §1) and Review Prototype v4 (asset RP4). The owner ledger is the decision overlay for the v10 review; the owner session amends that overlay. Each row names the change the next design iteration needs. The complete, self-contained change list, the acceptance scenarios and the copy-ready design-session prompt are in the design-prototype handoff. Nothing here authorises building a new prototype, messaging the design session or implementing any screen.

Prototype screen What the prototype shows Owner-session decision Change needed in the next iteration
RP4 reviewer page (three cards, corrections 1–11) and v10 p1/r4 p1: one card per step with a task strip; r4: one form area driven by the selected step; RP4 three cards Q-12 decided: step selector with one form area Adopt r4's single form area with a step selector; state by shape and text; a labelled menu below lt_sm; adjudication steps only for eligible adjudicators; training steps where configured
RP4 "Unsaved changes" chip and footer "All changes saved"; v10 "Save draft" Status copy that blurs draft and save D3-03 with U1 amendment (OS-A21) Show Draft auto-saved, Version checkpoint saved and Complete as three visibly distinct states (wording illustrative); keep an earlier checkpoint visible when newer edits are autosaved; confirm before an incomplete Save replaces a completed version
RP4 layouts at desktop widths; phone screening only in the earlier plan Desktop-first reviewer page D3-05 decided-amended; D2-16 replaced Design phone (390 px) and tablet (768 px) layouts for full annotation of the largest form with no feature removed: outcome grid presentation, drag alternatives, action bar above the on-screen keyboard, embedded source on narrow screens
RP4 ALL-CAPS letterspaced buttons, 4 px radii, hex values Pre-M3 styling D3-04 decided (U1); UI1 Material 3 sentence case, M3 shape and colour roles, no literals
v10 r2 step setting "Earlier gold standard" (RECONCILIATION.md line 31) Show to reconcilers or hide, set per step; candidates never see it Q-28 hints amendment; Q-30 (OS-A04) Move the setting to the form as a hints baseline shown by default to reviewers as labelled hints; a step may only hide it; "Use accepted answer" per question and "Use all available accepted answers"; no prefilling; hint authority and snapshot shown
v10 r6 forms page Guidance, target, versions, publish prompt, reconciliation settings including blinding, questions, extraction toggle OS-A12; Q-11; Q-04; R1 final (OS-A28); D4-17; D2-07; D3-17; D3-18; Q-30 Keep form-level blinding; add the standard target as part of the version (target change publishes), target-one handling, hints baseline, outdated-answer mode, completeness override, capacity cap, inactivity timeout and in-progress limit; bulk acceptance optional and off; remove "confirm each"
v10 r6 "Publish v3?" prompt Per-question compatible or not compatible Q-34 decided; D2-01 decided-amended; consolidation §6 Publisher declares compatibility and mapping explicitly; impact by session category and route; generated session versions named in the impact; target-only publications; recheck at commit with the shared impact preview
v10 r2 stage designer (lifecycle, steps, collective satisfaction, existing-work policy, who reconciles per part) Steps editor with dependencies separate from order Q-15 replaced; Q-01; Q-02; Q-28; OS-A05; OS-A27; OS-A13 Add the stage study filter editor; keep in-stage step dependencies; map "collective satisfaction" to the collective-Include stage setting and "existing-work policy" to the continuation settings; add exclusion-stop, route capacity cap and browsing settings; remove any cross-stage route control; who reconciles comes from stage grants; Q-02 lifecycle copy
v10 r3 selection preview and Monitor Current-state explanation of absent work OS-A07, OS-A08, OS-A11; D3-13; D3-20 Add the eligibility explanation timeline with historical versions, "No review is recorded" and tracking gaps; pool-level counts until X-AUTH-RESOLVER; names only for Monitor holders
v10 r1 screening profiles Screening questions, presentation order, decision from answers, agreement and resolution rows D4-01 (S1, OS-A16); tie amendment (OS-A13); D4-02; Q-32; Q-26; OS-A17; OS-A20 Add Unsure, the explicit tie policy (extra review with a bound, or adjudication) and adjudicator assignment to a member or group, discussion, rationale, blinding, the exclusion-reason guidance (no question-count limit) and source policies; profile publication impact
v10 DECISIONS.md line 242 (blinding on the profile and the form version; values global in the prototype) Blinding drawn on both profile and form version, not stored per scope Blinding and ordering amendment (OS-A02) Keep both settings and store them per form version (annotation reconciliation) and per profile version (screening reconciliation), blinded by default; no stage or step blinding setting
v10 4a reconciliation form and 3a matching Reviewer pills, agreement icon, dashed prefill with per-field CONFIRM, four-column side-by-side Blinding and ordering amendment (OS-A02, OS-A03); R1 final (OS-A28); Q-04 Context-local labels, random order, no chronology clues; scale to N candidates with a selector that never hides a disagreeing candidate; a one-candidate layout for required human reconciliation; blank comparisons shown as Not assessed; the reconciler's own prefill marker stays without per-field CONFIRM (RE2, unchanged by the owner session)
v10 4.1 screening decision first Screening part inside reconciliation OS-A13; D4-02; OS-A16 Separate screening adjudication task in a system adjudication step, with discussion history and rationale where required
v10 4.4 gold history Timeline with pinned inputs and "Inputs changed · re-check" R1 final; RS-R58 Show accepted-result authority (Single annotator, Human reconciled, Adjudicated, Merge resolved); flagged results stay current until explicit reconsideration
v10 Reconcile pool KPI cards, Start reconciling, Assign, Held and Corrections banners, bulk approve, Agreement tab Q-30; consolidation §1; Q-11; D3-11; T-SI-02 Optional expiry, off by default; additional-review requests that raise the effective target; bulk acceptance only where the form enables it; Agreement in its own place with source classes kept apart
v10 c2 questions and entity types Tree per entity type with settings and preview D2-11; OS-A01 Presence avatars, editing markers, live updates that keep local edits, recoverable conflicts, author and time in history; "Publishing by … since …" banner
v10 navigation (steps mode and Library mode) Setup runs through concepts, entity types, schemas, forms, profiles, stages; Library alternate mode D3-07; D2-15; OS-A20; OS-A25 Add My work, the cross-project tab and the global badge; the catalogue; Design › AI screening models; Deleted projects (restricted); keep "Library" for Study Management
Not in either prototype — OS-A27, OS-A29, OS-A18, OS-A19, OS-A24, OS-A25, OS-A14/OS-A15, D3-22, D3-24 New screens listed as gaps 21 to 36 in §3: browsing and Saved work, merge and unmerge, training, inference beta, exclusion, deleted projects, conversion wizard, notification content and mute

The PWA exploration (OS-A22) and accepted-answer branching between steps (D4-15) are outside the MVP; the handoff lists them as deferred so the design iteration does not draw them as launch scope.