Bromont 2026-06 — Master Production Plan

Show: MARS Bromont CCI Horse Trials Dates: Thursday 2026-06-04 → Sunday 2026-06-07 Levels offered: CCI4*-L, CCI4*-S, CCI3*-L (incl. U25), CCI2*-L (incl. U25), MARS Bromont Rising Scope target: CCI4*-L + CCI4*-S minimum; expansion possible if foundation supports it Distribution: Bromont Eventing YouTube channel — free live + VOD. ClipMyHorse.tv has the contractual deal; YouTube is the public-facing channel. Brand context: Not an H&C show. Bromont indigo/teal brand stack — assets at streaming-setup/assets/brand/bromont/. Production model: REMOTE — first for RNS. Mike + equipment at Hickory Hills, IL; owners at the show in Quebec. Days from today (2026-05-12): 23 days to Day 1.

This is a living document that captures decisions, lessons, and the evolving plan. Three phases: (1) retro, (2) Claude Code workspace audit, (3) Bromont design.

Phase 1 — Tryon retrospective

Status: ✅ locked (2026-05-12).

What happened at Tryon

• 2026-04-23 plan locked: Approach 1 (Evolution), harden Ocala stack, ship in 2 weeks.
• Actual April 25 – May 7: pivoted to Approach 2 Platform (FastAPI + Postgres + Redis + Next.js, deployed to console.schifftacular.com), Wirecast → OBS (4/25), Captivate → split-engine Captivate+AE (4/29), Lou Ann strategic pivot (4/27), Real → Reel rebrand (5/1), /xc director, /sj scorer, Tier-1 Captivate templates, marketplace, observability, Ontime — 13-day sprint, no rehearsal.
• Show days Fri 5/8 – Sun 5/10: pre-built stack was ignored. Mike + Claude live-built scrapers, director UIs, seed scripts, and Captivate bridge patches on show mornings.

Root cause (corrected)

Architecture choices were not necessarily wrong, but nothing was built correctly. Every piece of the April–May effort needs to be reassessed, redesigned, and rebuilt for Bromont. Tryon's scope was "every idea Mike could think of" with no hardening, no rehearsal, and no foundation that could handle scope evolution. The live-with-Claude pattern in the booth was a survival fallback caused by inadequate pre-production — not a workflow to optimize for.

Causal chain:

1. Wrong job at the wrong abstraction. April designs imagined a "platform operator" job that didn't match show-day reality.
2. No hardening. Features moved from "demo works" to "next feature" without being driven to operator-ready reliability under realistic load.
3. No practice. Mike never operated the system cold, end to end, before show day.
4. Scope creep ate the calendar. Every late addition stole hours from drilling features already built.
5. Result: nothing was trusted, nothing was muscle memory, everything got bypassed.

Lessons for Bromont (load-bearing constraints)

1. Claude's role is bounded but not absent from production. Pre-production = design + build + test (Claude's primary mode). Live-coding in the booth = anti-pattern. Production role for Claude exists and will be defined in Phase 3 — it is not "live operator."
2. "Done" includes muscle memory. A feature isn't done when the code runs — it's done when Mike has operated it cold under realistic pressure, with practice reps logged. Rep count and rehearsal mechanic are open (see lesson 6).
3. Practice is key — concept TBD. Need a way to "recreate show days" so Mike can practice TD work. Don't pre-design the mechanic; let it take shape during Phase 3. Pin the requirement, not the implementation.
4. Scope can evolve, but only on a correct foundation. Tryon's failure wasn't ambition — it was building on sand. Phase 3 must invest in a foundation correct enough that scope additions don't destabilize the system. Then scope can grow honestly.
5. Test in the broadcast chain — but on Mac Mini for daily iteration. Mac Mini needs a parallel OBS setup mirroring the iMac, so the daily build/test loop is local. iMac (remote-only now) reserved for "production test level" runs — full rehearsals, show day. This is a concrete build task for Phase 3.
6. Rehearsals are not calendar-scheduled. Practice affects only Mike; he does it when he can. Plan doesn't bake rehearsal into a schedule — but the capability to rehearse (item 3) has to exist.
7. Bromont ≠ Tryon. Different distribution (YouTube via CMH deal, not H&C), different crew, different brand stack, remote production from Hickory Hills. Don't auto-port Tryon decisions; re-justify each.

What we keep vs. drop from Tryon

Bromont rebuilds. Specific reassessment of which Tryon pieces (if any) survive happens during Phase 3. Default for Phase 3: assume nothing carries over; force every Tryon component to earn its way back in.

Phase 2 — Claude Code workspace audit

Status: ✅ functionally complete (2026-05-13). Tier A applied.

Approach: Six deep-dive research passes (hooks, plugins, skills, subagents, MCP servers, memory hygiene) producing concrete RNS-specific proposals. All proposals consolidated into a single selection form at .claude/workspace-upgrade.md for Mike to mark up.

Selection artifact: .claude/workspace-upgrade.md (workspace root). ~85 items across 8 categories with checkbox + comment per item.

Audit findings — short version:

• Zero hooks configured. Major missed leverage on automatic correctness checks (which directly address the "nothing built correctly" Tryon root cause).
• 106 user-level skills installed, ~80 plugins listed but only studio-coach enabled. Mike has a hardware-store catalog instead of a toolbox.
• One custom subagent (twentycrm-schema-architect). No RNS-specific subagents.
• No project-level .mcp.json. ~50 remote MCPs authorized via claude.ai, many unused.
• No RNS-specific skills. No Real Neat Solutions/.claude/skills/ or agents/ directory.
• Root CLAUDE.md describes a "multi-product workspace" but RNS is effectively streaming-setup now.
• Memory: 2 explicitly-superseded entries, several Tryon-era memories that need "under reassessment" marking.

Output: once Mike marks up workspace-upgrade.md, I apply only the checked items, then phase 2 closes.

Phase 3 — Bromont design

Status: 🔄 in progress (2026-05-13 → 2026-05-16). Substantial progress this session — handed off to a new session for closure.

Disciplined brainstorming pass. Starts from: "What absolutely must work for Mike, alone, in Hickory Hills, to direct Bromont's live program over YouTube?" Scoped outward only as the foundation correctly supports growth. Will define Claude's production role, rehearsal mechanic, remote-production architecture, and what (if anything) from Tryon survives.

Phase 3 snapshot — 2026-05-16 (compact handoff)

Locked this session:

• Eventing rules reference — FEI 2026 27th Edition + EC + USEF + Cheval Québec consolidated at EVENTING-RULES.md. Key change: missed flag 15 → 9 penalties (FEI 2026); frangible activation stays 11; SJ 2nd refusal = elim. Bromont OC publishes Long divisions (CCI4*-L, 3*-L, 2*-L + U25), not the Short-only mix one agent claimed.
• Bromont specifics — Course designer Derek di Grazia (Kentucky 5*, Burghley). SJ designer Brody Robertson. TD Martyn Johnson. GJ President 4*-L Peter Gray. MARS Bromont Rising = USEA grant + education layer, NOT a separate scoring division.
• Quebec language law scope established — French-predominant on outdoor signage + sponsor overlays + promo landing page; English-language broadcast itself is NOT required to be French.
• Section 1.1.1 Dressage + 1.1.2 Show Jumping + 1.1.3 Cross-country viewer experience all fleshed in BUILD-PLAN.md.
• XC layout palette — 5 locked (HERO, HERO+ROSTER, HERO+MAP, DUAL, START BOX PiP) + 3 added (FINISH LINE PiP, REPLAY+LIVE, LEADERBOARD+INSET). Killed: QUAD (max 3 riders not 4), WATER HERO (Hank context ≠ broadcast feature).
• Primary-slot graphics-follows-rider mechanic + per-rider stopwatch model locked. Earlier TPS-fed rider-to-camera coupling was REVERSED later in the same session — Bromont is operator-driven with no TPS integration. See [[project_bromont_no_tps]].
• Console role: orchestration + situational awareness, NOT a video mixer. OBS handles PGM/PVW/transitions; console fires graphics + commands OBS via WebSocket + tracks rider state. See [[project_console_role]] memory.
• Grandma-test usability bar — plain language, big buttons, no jargon, no keyboard-as-primary. See [[feedback_console_grandma_usable]] memory.
• Direction A — "Now Playing" picked as the operator-experience pattern. Working prototype at xc-director-now-playing/ and live at https://bromont.schifftacular.com/console/mobile/.
• Console hub in the brand book at https://bromont.schifftacular.com/console/ — two tools so far (Mobile Console + Show Jumping Timer). Designed to grow.

Prototype artifacts (all in streaming-setup/shows/bromont-2026-06/):

Phase 3 closure status (updated 2026-05-16):

Phase 3 is closed except for two items:

1. § 1.3 Venue crew experience — PARKED pending Lou Ann's POC vendor test results ([[project_bromont_comms_state]]). Not blocking other sections; revisit after Lou Ann's POC meeting.
2. Schedule shape reconciliation — BLOCKED waiting on 2026 omnibus drop from the OC. Resolve when published.

Everything else closed this session:

• ✅ § 1.4 Sponsor experience — blended model (tier-1 time-slot / tier-2 content-attached / tier-3 long-tail menu). Logo-only. 3 marquee fence bumpers. Proof-of-value reporting + phase-presenting structure deferred to post-Bromont (OC alignment).
• ✅ § 3 Bromont slice — Tier 0 floor / Tier 1 target / Tier 2 stretch / Slipped breakdown locked.
• ✅ § 4 Build order — 3 phases (Foundation May 17-22 / Heavy build May 23-31 / Integration + rehearsal Jun 1-3) + show week (Jun 4-7). Build force = solo + Claude Code + Claude Design + AE-MCP.
• ✅ Q4 (graphics standards) — N/A. Bromont is YouTube-only, Bromont brand only. No CMH constraint.
• ✅ Q6 (Claude's production role) — Mac Mini NDI rig per [[project_claude_production_role]]. Cuts every ride, tracks riders across cameras, output to console banner read-only. Failure mode is no-broadcast-impact.
• ✅ Q7 (rehearsal mechanic) — Phase 3 rehearsal phase June 1-3 uses recorded Ocala/Tryon source captures as synthetic feeds; full-chain test once Bromont team is on-grounds June 3.
• ✅ Highlight-capture pipeline — Mac Mini segment ring + scrub UI + Shuttle Pro mark input + OBS replay scene + replay bumper. Full-stack in Tier 1.
• ✅ Story-of-the-Short — KILLED. Won't be built for Bromont.
• ✅ XC block open / close — Saturday's Short finale + Long middle treatment folded into the build-plan tiers and AE template work in Phase 2.

Other locks this session (see decision log below):

• ✅ No TPS integration for Bromont ([[project_bromont_no_tps]]). Supersedes the earlier TPS-fed coupling lock.
• ✅ Console built in Claude Design, backend wired by Claude Code ([[project_bromont_build_decisions]]).
• ✅ AE + SPX-GC are co-primary graphics; Captivate is live redundancy running open on the iMac (NOT a cold fallback) ([[project_bromont_architecture_corrections]]).
• ✅ AE lives on the Mac Mini, pre/post-show only — not a live overlay.
• ✅ SPX-GC on the VPS, consumed by iMac as a browser source.
• ✅ iMac belongs to RNS Videomedia (not "Mike's"). Mark = partner + XC location-scout (not DP).

Locked context (2026-05-13)

• Mike's deliverable: directing the full live program remotely from Hickory Hills. (Resolves Q1.)
• Signal path: same LiveU stack used at Ocala and Tryon, staged in Hickory Hills Mon/Tue of show week. Lou Ann owns the LiveU relationship. Mike does not call the LiveU rep — equipment trusted to function as intended. (Resolves Q2.)
• Crew at venue: Mark + Lou Ann on site. Hank = primary camera operator — dressage, show jumping, and the XC water jump. Up to 5 locally-hired camera operators for the rest of cross country. Fallback: if Hank no-shows, Mark covers dressage and SJ, XC coverage adjusts accordingly. (Resolves Q3.)
• Gear in Hickory Hills: LiveU rental + iMac (OBS rig) + supporting kit, all set up and running by Mon/Tue before show week. Mike arrives the following day with ≥2 full days solo with the equipment before going live. (Resolves Q5 and Q8 — iMac is the production rig, not a fallback.)
• Rehearsal model: Mike arrives at the Hickory Hills setup with everything built, tested, and rehearsed. The 2+ days alone with real gear are validation, not first-touch. All design/build/practice happens before the move-in. (Frames Q7 as a pre-Hickory-Hills problem, not an on-site one.)

Still open

• Q4 — graphics standards: what does CMH / Bromont Eventing expect? Brand book, template pack, or blank canvas?
• Q6 — Claude's production role: pre-staged dashboards? On-call monitoring? Operator-triggered automations? Read-only observability? Needs definition before the rehearsal mechanic can be designed.
• Q7 — rehearsal mechanic, refined: with the "arrive ready" constraint, the question becomes "how does Mike rehearse directing a live multi-cam YouTube broadcast without the venue feeds existing yet?" Recorded Ocala/Tryon source captures? Synthetic feeds? Crew-in-the-loop video calls during a sim?

Phase 2 artifacts available during Phase 3:

• 13 new skills + 2 new subagents at workspace .claude/
• 10 new plugins enabled
• Community OBS MCP (pending Mike install)
• All Phase 1 lessons + corrected memory locked in

Updated Phase 3 starting moves:

1. Run /feature-opportunity-scan to map Tryon attempts → correct-implementation recommendations under the new constraints (remote director, no LiveU concerns, Bromont brand stack).
2. Talk to /expert-obs about scene architecture for: 1 primary cam op feeding 3 disciplines + 5–6 XC feeds via LiveU, directed remotely.
3. Define Claude's production role (Q6) — this gates the rehearsal mechanic.
4. Resolve graphics standards (Q4) by reading the Bromont brand assets and any CMH documentation we have.

North star (locked 2026-05-13)

Build the feature-complete console for the perfect broadcast of eventing. Bromont is the first show that uses it, not the boundary of what gets built. Ocala and Tryon are evidence — they showed what's needed and what fails — they are not code dependencies.

Ground rules for Phase 3:

1. No Tryon code carries over by default. Every feature is re-spec'd, re-designed, re-built. Old code can inform design (e.g. "the slot model worked"), but it does not start as a code dependency. See [[feedback_bromont_ground_up_rebuild]].
2. The product is the eventing console, not "the thing we ship for Bromont." Bromont is the forcing function for the first useful slice.
3. "Feature-complete" is the design target. That means defining the full set first, then honestly scoping which slice ships for Bromont and which slips to a later show.
4. Operator-ready + rehearsed is the definition of done for anything that ships, per Phase 1.

Scan output (2026-05-13) — overridden

The infrastructure-mapper scan biased toward "carry forward Tryon code." Mike overruled it: ground-up rebuild is the framing. The scan's observations about what shapes a show day (scene topology, graphics chain, rehearsal mechanic, monitoring) are useful as prompts for the build plan, not as recommendations. Don't cite the scan's verdict columns when planning.

Decision log

Running assumptions

• A1: Mike directs the full live program remotely from Hickory Hills, IL. RNS Videomedia (Mark + Lou Ann) is on the ground at the venue.
• A2: Distribution is the Bromont Eventing YouTube channel; CMH is the contractual back-end but invisible to public viewers.
• A3: Venue crew = Mark + Lou Ann + Hank (primary cam: dressage / SJ / XC water) + up to 5 hired local XC camera operators. If Hank no-shows, Mark covers dressage/SJ and XC adjusts.
• A4: Rehearsals are not calendar-scheduled — they affect only Mike, who does them when he can. The capability to rehearse is required; the schedule is not. With "arrive ready" locked, rehearsals must work before Mike reaches the Hickory Hills gear.
• A5: Tryon's pre-built stack is not assumed forward. Every component re-justifies or stays out.
• A6: 23-day window to Day 1. Real build window is shorter once foundation work is honest. Mike arrives at Hickory Hills the day after the gear is staged (Wed/Thu of show week), giving ≥2 full days solo before going live.
• A7: LiveU + iMac stack is Lou Ann's responsibility to stage and trust. Mike does not contact the LiveU rep — equipment is assumed to function as configured.

Open questions (Phase 3)

• Q4: What graphics standards does CMH / Bromont Eventing expect — brand book, CMH template, or blank canvas?
• Q6: What does Claude's production role look like? (Pre-staged dashboards? On-call monitoring? Specific automations triggered by Mike?) Gates the rehearsal mechanic.
• Q7 (refined): How does Mike rehearse directing a multi-cam YouTube broadcast before the venue feeds exist? Recorded Ocala/Tryon captures replayed into the chain? Synthetic feeds? Crew-in-the-loop video calls during a sim?

(Q1, Q2, Q3, Q5, Q8 resolved 2026-05-13 — see "Locked context" in Phase 3.)