Decision Log¶

Updated: 2026-06-06

Purpose¶

This log captures material decisions made about the enterprise — what was decided, when, by whom, and the reasoning. The goal is institutional memory: future advisors, future hires, and future versions of MC-ATG can see how the enterprise's thinking evolved rather than reconstructing from inference.

DECISION-PEOPLE-ROSTER-CORRECTIONS — 2026-05-27 — Andrea and Karin = building cleaning; CTP photo capture = Hector Peck¶

Made by: Brandon Tomkins (5/27 clarification, codified to people pages 2026-06-06)

Prior framework documents misattributed roles for two team members. Corrections (canonical):

Andrea Rodriguez — building cleaning. NOT assistant shop manager, NOT bilingual customer-facing, NOT accounting bridge to Jose's crew (all prior attributions incorrect).
Karin — building cleaning. NOT CTP photo capture operator (prior attribution incorrect).
CTP parts photo capture operator = Hector Peck (filled gap; people/hector-peck.md created 2026-06-06).
Bilingual operational bridge to Jose's crew = Jose himself (the only EN/ES speaker on the floor; updates older "Jose Spanish-only" attribution).
QB / accounting = Claude (in progress; substrate-native accounting agents shipping through 2026 Q2).
1099 crew roster verification = open gap. Stated plainly on the People index; full roster verification on brandon-action-items.md § Team roster verification.

Implication: future questions about who operates which function get answered against this entry. The people-page profile text says role only — no permanent "NOT X" framing on profiles (corrections live here in the log, not on profiles).

This is the chronological decision record. The Deferred Decision Markers (DDM) system tracks decisions that aren't yet made; this log tracks decisions that are.

2026-05-30¶

Cactus Diesel merged into Superior Service (folded under Superior Equipment Inc)¶

Decision: Cactus Diesel ceases to exist as a standalone brand. The diesel-repair / mobile-mechanic / fleet-service offering is merged into Superior Service — a new service-arm of Superior Equipment Inc. Online home: superiorequipmentinc.com/service/. Cactusdiesel.com domain: sunset entirely — site taken offline, registration allowed to lapse at next renewal. Context: Brandon decision 2026-05-30 in the WDM session after reviewing the Q2 2026 audit on cactusdiesel.com (composite 51/100, biggest finding: architectural ceiling on a single-page brochure, no GBP, NAP collision with Superior Equipment already at 156 E Mohave St). Merging the diesel-repair offering under SEI's existing trust signals (LocalBusiness+AutoDealer JSON-LD, 7-state areaServed, established phone/address) inherits years of accumulated SEO + brand equity rather than rebuilding from zero. In-flight CD work disposition: all unshipped CD-side audit recommendations (6 service-detail pages, 6 service-area pages, /fleet/ page, NET-30 application, Service schema) get re-targeted at superiorequipmentinc.com/service/. The CD audit doc itself gets repurposed as a Superior Service launch playbook. The two shipped CD fixes (Formspree → mailto fallback, GTM placeholder flagged) become moot at sunset. Supersedes: the entire Cactus Diesel standalone product strategy. The NAP-collision finding on Yelp (156 E Mohave resolving to Superior Equipment) is auto-resolved by the merge — there is no longer a collision because there are no longer two businesses. Affected artifacts: D:\www\www-cactusdiesel\ (sunset notice deployed), web-site-optimization/cactus-diesel-2026-q2.md (repurposed as launch playbook), web-site-optimization/index.md audit-cadence table, mkdocs nav, dashboard infra.html + web-traffic.html entity references (deferred to next dashboard refresh), and superiorequipmentinc.com IA + sitemap (additive — /service/ becomes a new top-level surface). Brandon-gated for the new /service/ section build: real rate card (trip charge, hourly labor, after-hours premium, diagnostic credit), real response-time guarantee (radius + minutes), real certification list (ASE, OEM training), real fleet-pricing tiers. The audit's Phase 5 template baked in invented specifics that won't ship without verification — same no-fabricated-data rule that's held the rest of the portfolio together. Implication: Service portfolio (1 dealer site + 1 service-area + 5 marketing sites) drops to 4 marketing sites + SEI gains a service surface. Strategy audit-cadence table updated to reflect the merge. Future Q2/Q3 audits of /service/ happen under the SEI Q-audit cycle.

2026-05-27¶

LLM infrastructure — managed API, no local LLM hardware for the fleet¶

Decision: Agent fleet stays on managed API. No local LLM hardware purchase. Tier by role: Opus = MC1; Sonnet = builders (OM, SE, WDM, AM); Haiku = volume/extraction (QB Assistant, future Auction agent, photo/parsing). Context: Brandon thinking-session 2026-05-27 — small-business scale doesn't justify $250K frontier hardware or even $10-15K workstation (6+ yr break-even running weaker model than Opus). Frontier open models improve monthly; rent improving capability, own only stable high-volume. Managed API = reliability + redundancy + no bootstrap problem (agent that fixes things can't live inside the box that's down). Supersedes: the "local-substrate DDM" thinking (Mac Studio vs RTX, Q1 2027 revisit, hybrid local+frontier) and any earlier cost-cap early-checkpoint thresholds — replaced by measure-then-decide. Marketing GPU box (decoupled): RTX 5090 32GB approved in principle, justified on Stable Diffusion / Flux / local video alone. LLM serving is bonus overflow, not the reason. 48GB tier deferred until measured token spend justifies. Decision rule on self-hosting: stand up fleet on tiered API, measure one month, decide. < ~$200/mo LLM side → stay on API; several-hundred+/mo daily-heavy → marketing box upgrades to 48GB, self-hosts builder tier as overflow. No AI-marketing-consultant retainer: if Dennis's friend engages, tight time-boxed strategy project only — flat fee, defined deliverables, end date. Execution stays in-house on existing AI stack. Implication: Substrate-agent cost discipline = prompt caching (already shipped on AM-agent) + tier-routing + measured monthly spend. The $20/agent + $100/aggregate caps remain as runaway guardrails, but the real decision input is measured spend vs $200/mo line.

Cactus IMS integration routed to WDM with full latitude¶

Decision: Cactus IMS → cactustruckparts.com integration is WDM-owned. Brandon explicit 2026-05-26 ~18:30 PT: "give the cactus IMS integration to the web design manager and tell her to use her best judgement."

Context: Brandon initially wanted to outright kill Cactus IMS; chat-MC1 wanted to merge. Final call = merge. cactustruckparts.com already runs Supabase + /admin + sitemap + category taxonomy (Scenario C confirmed by mc1-code audit). WDM enumerates the IMS feature gap, sequences the port, decides design judgment. Surfaces only four-test failures to Brandon. Implication: Brandon out of the loop on routine port decisions. Q3 2026 target likely compresses given how far along cactustruckparts.com already is.

MkDocs v0.2 strategic spine staged as canonical source¶

Decision: chat-MC1's 31-doc v0.2 package extracted from Downloads and staged into D:\claude_www\projects\enterprise\strategy\ with full subdirectory structure (operating-arms, future-arms, partner-side, operations, people, decisions, principles, interview-methodology, ai-organization). Old flat-doc /strategy/ subdirectory archived. Context: Reconciled mkdocs.yml at projects/enterprise/mkdocs.yml points docs_dir at the staged source, sites to D:/www/www-strategy, subdomain strategy.copperstatetruckparts.com. Package's curated nav preserved + mc1-code's 6 unique substrate-build docs slotted under "AI Organization". Backup of pre-staging state at .tmp/strategy-backup-pre-v0.2-20260527/. Implication: The spine is now the canonical source-of-truth for strategic documentation. AI roles edit markdown in strategy/; mkdocs build regenerates HTML. IT deploys to the subdomain with Cloudflare Access auth (directive at to_it_manager/20260527-1300 + supplements).

Manager-as-substrate-agent ordering ratified¶

Decision: Order for future manager-agents going substrate-native: OM (next, pending migration-gate re-run by SE-Infra) → IT → WDM → SE-App → SMM → AMM → AI Comm Monitoring. AMP/AM done. Context: AM-agent established the template (callable tools, judgment-bearing, mid-response Claude API calls via tool-use, prompt caching, SHADOW-mode validation before LIVE flip). Pattern is the substrate-native equivalent of how chat-MC1 operates: context + judgment + tool-use to ground judgment in data. Implication: SE-Infra owns each Phase 0 runtime build per the AM template. Each manager-agent passes through SHADOW → LIVE-with-Brandon-ratification gate. Cost discipline (prompt caching, tier routing) inherits from AM-agent's template.

DET activated¶

Decision: Desert Equipment Transport (DET) activated — insurance bound, paperwork filed. Status: Operational Context: DET had been operationally ready for over six months, gated only on insurance. With insurance bound and authority filed 2026-05-27, DET is ready to begin commercial operations. Implication: DET Operating Model becomes active. The load-board read-only build (Phase 1 of dispatch capability) is appropriate near-term work.

Honest-state markers operating principle established¶

Decision: All framework documentation uses red/yellow/blue admonitions to signal confidence (information needed / verify before relying / documented assumption). Context: Andrea and Karin role corrections surfaced that chat-MC1 had been documenting inference as fact. Brandon ratified the admonition system to make documentation visually honest. Implication: Documents become trustworthy for external readers (Substrate Eye AI, future CPA, banking). The corrections loop is visible rather than hidden.

Pivot to MkDocs for strategic documentation¶

Decision: Replace custom WDM-built strategic documentation system with MkDocs + Material theme. Context: Custom WDM build had been in flight without visible progress. MkDocs provides industry-standard documentation infrastructure with mature features. WDM's bandwidth returns to truck-operations dashboard work. Implication: Strategic documentation lives at strategy.copperstatetruckparts.com (when IT deploys). Source markdown files in D:\claude_www\projects\enterprise\strategy\. AI roles read/write markdown natively.

Andrea and Karin correction¶

Decision: Andrea Rodriguez and Karin are building cleaning staff, not office operations. Context: Brandon clarified 2026-05-27 that prior framework documentation incorrectly documented Andrea as "assistant shop manager" with operational responsibilities and Karin as CTP photo capture operator. Both were incorrect. Implication: Email architecture and any other workflows assuming Andrea or Karin as informal triage layer must be fully substrate-driven instead.

Brandon Tomkins, not Brandon Wynn¶

Decision: Principal's correct name is Brandon Tomkins. Dennis Wynn is partner, not family. Context: Multiple framework documents had incorrectly used "Brandon Wynn" — chat-MC1's misattribution propagated across weeks of work. Implication: Name correction sweep across all framework documents. Brandon Tomkins as principal, Dennis Wynn as 20-year partner.

2026-05-26¶

Substrate v1.0 LIVE¶

Decision: Claude OS substrate v1.0 ratified as LIVE. Context: SE-Infrastructure shipped substrate code-complete (271 tests, 11 packages) 24 days ahead of target. Four cutover gates closed in ~67 minutes under standing operating mode (short timer + tactical override). One temperature-deprecation bug caught at smoke layer, fixed in-flight. Implication: Framework can now operate substrate-native agents. Manager-as-substrate-agent pattern enabled.

AM-agent shipped as first judgment-bearing substrate agent¶

Decision: AM (Accounting Manager) deployed as substrate-native agent with callable tools (read_qb_warehouse, read_categorization_rules, read_accounting_snapshot). Flipped from SHADOW to LIVE production by Brandon explicit ratification 2026-05-26 ~17:28 PT. Status: Operational; v0.3 with prompt caching (~4× cost reduction at scale). Today's spend ~25¢ against $20/day individual cap and $100/day aggregate guardrail. Context: AM-agent became the template for manager-as-substrate-agent pattern. Calls Claude API mid-response via tool-use to ground judgment in data. Sonnet tier per the post-decision tiering (Opus = MC1, Sonnet = builders, Haiku = volume/extraction). First post-flip loop bug (MM ?since= returned same posts) caught + fixed with post-ID dedupe state; subsequent exchanges single-fire clean. Implication: Other managers migrate to substrate-native in chat-MC1-ratified order: OM → IT → WDM → SE-App → SMM → AMM → AI Comm Monitoring (AMP/AM done). OM-agent blocked on rehydration-eval re-run by SE-Infra.

LedgrAI Pro killed¶

Decision: LedgrAI Pro project terminated. Context: The 41-account holding-company chart of accounts + intercompany work was scaffolding for a QB-replacement project that proved deprioritized. Brandon confirmed: dies with LedgrAI. Implication: Learning preserved (the intercompany account structure informs the real entity billing work); the project itself is closed.

CapitalBridge parked¶

Decision: CapitalBridge hard-money-lending ERP project parked. Status: Revisit when Tamarack lending operations activate Context: Built as learning project for the eventual Tamarack platform. Tamarack remains dormant; CapitalBridge waits. Implication: When Tamarack activation triggers, revisit CapitalBridge as starting point for the lending platform.

Cactus IMS absorbed into Cactus Truck Parts¶

Decision: Cactus IMS (separate inventory management project) absorbed into cactustruckparts.com. Context: cactustruckparts.com already runs Supabase + /admin + sitemap + category taxonomy. Merging IMS scope into the existing web app is the right path; maintaining separate IMS isn't. Implication: Feature gap list needed (which IMS scope already exists in /admin vs. needs porting) — pending Brandon enumeration.

2026-05-25 and earlier¶

SP entity description corrected¶

Decision: SP = Superior Water Truck Parts (NEW parts imported from China). Distinct from CTP. Context: Previously documented as "role pending clarification." Implication: SP-MX expansion thesis became coherent.

CTP-MX maps to Ritchie Bros opportunity, not SP¶

Decision: Ritchie Bros VP visit insight applies to CTP (used parts conduit), not SP. Context: chat-MC1 had attributed the cross-border opportunity to SP initially; Brandon corrected. Implication: Two separate cross-border thesis tracks: SP-MX (near-term, asset-light, new parts) and CTP-MX (dormant, partnership-dependent, used parts).

Framework-as-product strategic context documented¶

Decision: Dennis's vision of deployable framework documented as future arm. Context: Brandon-Dennis conversation 2026-05-22 raised the thesis. Implication: Phase 1 priority remains ATG at design quality. No engineering attention diverted to productization until ATG framework demonstrably works.

DET liability isolation reasoning documented¶

Decision: DET stays separately Brandon-owned for liability isolation. Context: Brandon corrected chat-MC1's initial recommendation to fold DET into ATG. Commercial trucking liability is categorically different from dealership liability; separation prevents veil-piercing exposure. Implication: Captured in Entity Structure & Reasoning as a deliberate-and-protected decision.

How this log gets maintained¶

When material decisions are made (strategic, structural, operational), an entry lands here. chat-MC1 produces draft entries; Brandon corrects/ratifies. The log is append-only — corrections happen via new entries that supersede, not by rewriting history.

Decisions still in flight (not yet made) live in the DDM (Deferred Decision Markers) system when authored.

Living document. Last updated 2026-05-27 by chat-MC1.

2026-07-08 — Jeep entity code standardized to SJB / Sedona Jeep Builders¶

The site carried two codes for the Jeep entity: "SJB — Sedona Jeep Builders" (entity page, index, holding structure) and "SJR — Sonoran Jeep Rebuilder" (marketing & sales plans page). Per the strategy-site fix directive (MC-ATG, 2026-07-08): one code site-wide until Brandon settles the brand naming. SJB / Sedona Jeep Builders is the standard; the marketing-plans heading was brought in line. Open item unchanged: brand name vs. the live sedonajeep.com domain — Brandon's call, tracked in deferred decisions.