GAB Enterprise Architecture Consolidation — ACOI Follow-On

PURPOSE. Produce an architecture GAB can ADOPT, settled enough that the Design and Visual initiative (project ed16b3ff) builds on it rather than inheriting accumulated layers. CEO standard: the design work "will not be meaningful if the architecture is confusing and overlayered." The initiative must do actual work on the findings — establishing ownership and binding — not produce another map. Baobab discovered; Canopy mapped and was explicitly barred from restructuring and product renaming; this initiative changes things. TWO STRANDS WITH DIFFERENT GATING RELATIONSHIPS. STRAND A, consolidation, is what the design initiative depends on: module contract coverage where none exists (website first), binding of every public surface to a named product with URL recorded and a stated subdomain-to-product rule, one division taxonomy consumed from a single source, and disposition of duplicate surfaces. Naming and product-relationship answers flow from Strand A into the design initiative, which consumes rather than decides them. STRAND B, the three open capability decisions — GIS-specialist reasoning capability, company-specialised model or training strategy, and plugin packaging — requires explicit CEO dispositions but DOES NOT GATE the design work. TIMING CONSTRAINT. The plugin decision waits for the 1 October 2026 cross-agent benefits review, because ACOI recorded that no plugin benefit was demonstrated and the measurement intended to test it covered only the in-app Chief of Staff rather than the CEO's actual operating surfaces. BOUNDARIES. Do not reopen Baobab, Canopy, ACOI or Farm Explorer Live — all remain closed and are read as inputs. Do not re-derive their evidence. Strand A must be bounded and sized to complete alongside the design initiative's WP1 to WP3 rather than blocking it indefinitely. Capacity must be reconciled against the concurrent design project before any promotion to Projects. No implementation, plugin build, model tuning, GIS-specialist build, code change or restructuring is authorised by this gate. OUTSTANDING AT G0. Tailoring and scoring were not completed at intake — governance level is set to full but tailoring, strategic fit, value, risk and capacity scores remain unset. G1 should establish them. Linking of the five untriaged ACOI feedback records to this initiative awaits explicit CEO confirmation; only 6aa04059 is linked so far.

SCOPE IS STRAND A ONLY. Consolidation: module contract coverage where none exists — website first (comms owns company website management by contract while docs/modules/website.md does not exist and no deployment tool is mapped), eiras second (registered module, no contract document); binding of every public surface to a named product in the register with its URL recorded and a stated subdomain-to-product rule; one division taxonomy consumed from a single source rather than four variants declared in code; explicit disposition of the duplicate surfaces (flood and wetland-flood serve the same FloodSense application; ramsar and wetland-flood are legacy names predating the completed rename). STRAND B IS DEFERRED, NOT EXCLUDED. CEO decision 2026-09-08: get the architecture right first before considering plugin packaging, fine tuning and GIS specialisation. Feedback records 76280206, bd5d54b0, 98b88d17, 09256d55 and 6d111446 remain live and return in sequence after the consolidation delivers. Their 2026-09-30 due dates will lapse; that is a recorded CEO sequencing choice, not neglect. A formal deferral with a review date has not been set and the date is the CEO's to give. The plugin question additionally carries its own evidence dependency on the 1 October 2026 cross-agent benefits review. THE STANDARD THE WORK IS HELD TO. Establish ownership and binding; do not merely correct current values. Point fixes have been applied repeatedly — Farm Explorer Live task 1ec5ff33 corrected the routing split in August and the divergence returned in another form — because no contract assigns the binding. If the work reduces to a list of individual defects better handled as operational fixes in their owning modules, stop or redirect: consolidation must change ownership or it is not this initiative. BOUNDARIES. Do not reopen Baobab, Canopy, ACOI or Farm Explorer Live; all remain closed and are read as inputs, never re-derived. Do not redefine the company or its divisions, which are settled. No implementation, plugin build, model tuning, GIS-specialist build, code change or restructuring is authorised by this gate. Strand A must be bounded and sized to complete alongside the design initiative's WP1 to WP3 rather than blocking it. Capacity must be reconciled against the concurrent design project ed16b3ff before any promotion to Projects — the design initiative currently holds first call on CEO review time by the CEO's own allocation of 2026-09-06. NEXT: prepare the charter basis for Strand A toward G2, decomposed into bounded ownership and binding decisions with named owners.

Charter v1.0 approved. Scope is Strand A only, six work packages: WP1 surface enumeration and baseline; WP2 product-to-surface binding with CEO selection of the naming rule and each duplicate disposition; WP3 contract coverage authoring docs/modules/website.md and the eiras contract and reconciling with comms.md; WP4 single-source division taxonomy; WP5 retirement discipline — the contract duty plus a CI check that fails on a duplicate declaration; WP6 handoff to project ed16b3ff, acceptance and closure. CAPACITY ALLOCATION REMAINS UNRESOLVED AND BLOCKS PROMOTION. The charter requires the CEO to state what review allocation this initiative receives before promotion to Projects. G2 approval did not specify it. The tension is live and explicit: on 2026-09-06 the CEO allocated first call on his review time to the design initiative ed16b3ff with all five other active projects continuing unpaused; this initiative carries four further CEO decision points. An explicit workable allocation is required — not a cleared backlog, and not an assumption that the design initiative's allocation extends to cover this work. STRAND B REMAINS DEFERRED, NOT EXCLUDED. Plugin packaging (76280206), GIS-specialist capability (bd5d54b0), model specialisation (98b88d17), enum validator hardening (09256d55) and post-closure context-layer maintenance (6d111446) are sequenced behind this work by CEO decision of 2026-09-08. The plugin question additionally awaits the 1 October 2026 cross-agent benefits review. STANDARD AND BOUNDARIES UNCHANGED. Establish ownership and binding; do not merely correct current values — stop or redirect if the work reduces to a defect list better handled as operational fixes in owning modules. Estimates do not become commitments until WP1 closes the recorded coverage limits: the public surface list is not exhaustively enumerated and gab-website was not audited for surface coverage. All code change passes the Lane B pre-build gate. Public URL or subdomain retirement requires link migration, an approval path and a recorded rollback. Closed projects stay closed. No implementation, code, contract or surface change is authorised by this gate; G3 approves the delivery baseline, naming and disposition selections are made at WP2.

Execution, monitoring and control complete against the G3 baseline of 2026-09-11. All eight work packages are done; project progress 100 percent. ONE REBASELINE OCCURRED AND IS RECORDED, NOT ABSORBED. WP3b-ii's pre-build gate found the original 4-5 day estimate for the FloodSense and RamsarMap rewrite had been withdrawn and never replaced, and resized the work at 6-9 focused days (FloodSense 2-3, RamsarMap 4-6). That resizing is what surfaced the RamsarMap question rather than letting it run silently over. TWO IN-FLIGHT CORRECTIONS TO THIS PROJECT'S OWN G1 FINDINGS, both made at pre-build gates and both narrowing the claimed defect: 1. WP4: the G1 assessment stated four live division taxonomies. The gate proved three of the four apparent variants are legitimately distinct taxonomies under separate DB constraints (finance division_budgets six values; project_registry.category seven different values), and real divergence is four sites across three files, not four taxonomies. The wider framing is withdrawn. 2. WP5: the same pattern recurred across the Lite suffix, the eiras API path pairs, farm-explorer in two repositories and two held company assets. That constraint shaped the mechanism — check-bound-values.mjs encodes legitimate difference via distinct[] rather than failing on textual duplication, which a naive checker would have done on every CI run. MONITORING EVIDENCE. The estate's gates are live and enforcing: MCP parity at 233 handlers / 231 model-visible tools; check-bound-values in CI; check-brand-surfaces added 2026-09-20 and proven against two real drifted surfaces rather than a fixture; check-dependency-audit replacing a binary npm audit gate that had been failing since 2026-09-06 and silently skipping Deploy Supabase Migrations for fourteen days. That fourteen-day blind spot was found and closed during this project's execution window and is the clearest evidence the monitoring duty was actually discharged rather than assumed.

Two CEO actions remain outside the repository and do not block acceptance, because neither serves a branded company surface any longer: delete the superseded Vercel projects gab-flood-tool and gab-sitesense-lite, both verified clear of branded domains on 2026-09-20. Separately, if ramsar.gabclimatesmart.com is removed from gab-ramsar-map, the app's new origin must be supplied so WETLAND_FLOOD_ORIGINS can be updated, or its own API calls will be rejected.

ONE EXPECTED BENEFIT IS NOT MET AND IS CARRIED FORWARD, NOT CLOSED. benefits_check is recorded false for that reason. This project's stated benefits included "recorded dispositions for the three expiring ACOI capability questions". Verified against the operational_feedback register on 2026-09-20: all three are still status 'new', undispositioned, and due 2026-09-30 — ten days out. - 76280206-c529-4671-8ea1-5d7d3d10903d — plugin packaging for the GAB OS operating context layer - bd5d54b0-99a3-4b84-9d96-976a9497a16c — GIS-specialist reasoning capability - 98b88d17-8731-47dd-b3a7-a9fc5030f75f — company-specialised model or training strategy TWO FURTHER ACOI ITEMS ARE ALREADY PAST DUE, also still 'new': - 09256d55-7da1-41fc-a196-924182ba8c61 — enforce enum membership in the central tool input validator (change_request, was due 2026-08-31) - 6d111446-0951-4915-8352-f8bedd564025 — maintain the ACOI context layer after closure (operational_fix, was due 2026-08-13) These are CEO and Engineering dispositions, not agent work, which is why they were not closed by delivery. They are recorded here rather than allowed to lapse silently — an expiring question that nobody dispositioned is exactly the accumulation pattern this project exists to stop.

# GAB Enterprise Architecture Consolidation — ACOI Follow-On ## Charter basis v1.0 — draft for Daniel's G2 review Prepared 8 September 2026. Portfolio initiative 044bc747-1ee6-424d-9c0d-14990d53f572. Full governance. G0 and G1 are approved; this document does not record G2 approval, authorize implementation or spending, or start any code change. ## 1. Purpose and authority Produce an architecture GAB can ADOPT — settled enough that the Design and Visual initiative (project ed16b3ff) builds on it rather than inheriting accumulated layers. The CEO's standard, stated 2026-09-08: the design work "will not be meaningful if the architecture is confusing and overlayered." This is the consolidation

This project created the website module, which did not exist when it started. Routing and public surfaces had been owned by comms.md by default while no website contract existed at all. That module is now the operating owner of what this project produced, and the project closes rather than continuing as a wrapper. WHAT TRANSFERS TO WEBSITE. Subdomain-to-product rule v1.0 (one branded subdomain, one public tool, short name without the Lite suffix), CEO-confirmed 2026-09-12. The TOOL_SUBDOMAINS map in proxy.ts, now nine entries. The RETIRED_SUBDOMAINS map and the wetland-flood 308 redirect to /tools — which stays ATTACHED to gab-os deliberately, because an unattached subdomain under the wildcard DNS gets no TLS certificate and the redirect would break, not become redundant. WHAT TRANSFERS ELSEWHERE. eiras owns the public tool backends and lib/eiras/public-tools.ts. brand owns lib/brand.ts and the identity values. finance owns lib/divisions.ts as the single division source. RECURRING DUTIES NOW OWNED, NOT PROJECT WORK. Run check:brand-surfaces after a deploy, never before — it inspects production, so running it early faithfully reports the old build. Keep check-bound-values' registry current when a new bound value appears; lib/eiras/public-tools.ts is listed there as identified but NOT yet enforced and is the next one to wire. Prune check-dependency-audit's ACCEPTED list when an entry stops being reported — its own STALE check enforces this, and that is what caught the two next@14 entries on 2026-09-20. THE FAILURE MODE THIS HANDOVER EXISTS TO PREVENT. Every defect this project fixed came from a surface nobody owned. Farm Explorer served two different experiences at once; wetland-flood cross-linked every Lite tool to the wrong product; two tools sat on a retired identity for weeks while both source gates stayed green. None were caused by bad code. They were caused by no named owner and no check that looked at the live thing.