Radar Log

Free in-browser splat to flythrough-video deliverable; add to browser-splat-viewer vendor table (Arrival Space/Mint/OVR/SuperSplat). SPZ=Cesium interop, LCC2=XGRIDS.

1. 1. Drop a Building C / CorbeauSplat test splat into SplatStroll
2. 2. eval walk-through quality + video export resolution/watermark/branding vs Cesium + custom viewer

Promote SAM3 'sky, water' mask to a default pre-SfM step in CorbeauSplat, especially for water-adjacent sites (VPC, Cameron, Niricson dams).

1. 1. Pull updated kotohibi tool
2. 2. run 'sky, water' mask on a VPC water-heavy set + a Cameron wetland set
3. 3. compare sparse-recon vs no-mask baseline

Canonical 360-3DGS recipe to mirror in CorbeauSplat. Cubemap tool now commercial (on sale). LFS-MRNF 'no cleaning' = potential Brush-cleanup replacement.

1. 1. Price/eval the on-sale cubemap tool vs DIY
2. 2. trial LFS MRNF no-clean vs Brush on a known set
3. 3. spec an AVATA360 Building C interior recapture

Post-train repair pass before SOG/SPZ export; fixes mush-off-capture-path / Building C auto-orbit-into-soup.

1. 1. Clone nv-tlabs/artifixer + weights (Paperspace/Linux GPU)
2. 2. run Building C free-roam A/B vs static animTrack

Post-train repair pass → free-roam no-mush splats; a sellable quality lever vs on-rails competitor tours.

1. 1. Clone nv-tlabs/artifixer + pull weights (likely needs Linux/CUDA GPU or Paperspace), run repair on Building C avata splat, A/B free-roam vs static-animTrack.

New render-to-splat lane vs COLMAP-from-photos; solves coverage/occlusion holes + enables pre-construction splats from CAD.

1. 1. Identify Lightfield Studio (vendor/pricing/local)
2. 2. test Blender model -> batch render -> reconstruct

Native-to-VPC-stack splat-on-map path; no Cesium/Three.js engine swap. Commercial vs $0 DIY tradeoff.

1. 1. Trial GeoSplats with a Building C/Venice splat
2. 2. compare georef accuracy + cost vs Cesium ion

No-upload local conversion = client-data-privacy differentiator (capture never leaves the machine).

1. 1. Run Building C exterior PLY through gauzilla.xyz
2. 2. validate 3DTiles in Cesium + georef-without-EXIF vs SplatTransform

Free full-res viewer-of-record; removes splat-size-vs-web-perf tradeoff, no aggressive decimation.

1. 1. Load Building C/Magnolia full-res splat in upgraded SuperSplat
2. 2. verify load+60fps + WebGPU browser gate

Adds parcel/footprint/asset-inventory deliverable layer without separate GIS data buy or manual asset ID. AI infra-detection maps directly to Entergy utility asset inventory.

1. 1. Verify CAVU's Google Earth tier includes premium layers
2. 2. test AI infra-asset layer on an Entergy/VPC area vs manual inventory
3. 3. fold into utility + site-planning proposals if accurate

Voxel collision = click-on-callout/measurement raycasting with NO Nira and no separate collision mesh. Second $0 browser viewer alongside CesiumJS; pick per project on UX.

1. 1. Stand up a PlayCanvas splat viewer with a Building C exterior splat
2. 2. A/B the collision-callout UX vs the Cesium track
3. 3. choose viewer-of-record per project

Market-credibility air cover for the open-standard 3DGS deliverable; KHR_gaussian_splatting is now de facto standard. $0-SDK chain replaces Nira/Matterport.

1. 1. Use 'no broccoli trees / future of Google Earth' framing in proposals vs legacy ortho/mesh. No new tool — reinforces existing capture->stream pipeline.

Market air cover for the open-standard 3DGS deliverable already being assembled (Cesium streaming + SplatTransform post-process + SAM3 masking). Replaces Nira/Matterport with a $0-SDK open chain; capability is the pitch, not a discount.

1. 1. Use Bilawal's framing as proposal language for 3DGS site twins vs legacy ortho/mesh. No new install -- reinforces the Cesium-first Building C Track 2 plan.

Inserts as a pre-SfM masking step in the CorbeauSplat playbook alongside SplatTransform 2.0 (post-process) and Cesium streaming (delivery). Continues open-source 3DGS pipeline maturation.

1. 1. Watch kotohibi GitHub for next release. Bench on (a) Building C 360 interior (documented 0% align), (b) Cameron Parish foliage-heavy set, (c) PTFI pit. Fold into CorbeauSplat if SfM improves.

118KB collision-per-scene = pricing-talking-point. Pencil PlayCanvas as Plan C viewer-of-record alongside Cesium (Plan A) and Three.js (Plan B) for Building C Phase 5 Track 2.

1. 1. Install on Mac Studio: npm install -g @playcanvas/splat-transform. Validate against a Building C exterior splat. Pair with PlayCanvas viewer test.

Japanese 360-3DGS market consolidating around kotohibi's playbook. CAVU's US-market splat positioning needs to lock in before pipeline gets cloned by every drone operator. AprilTag for real-scale is a missing primitive in CorbeauSplat. Pipeline guide on GitHub is now the canonical recipe — don't reinvent it.

1. 1. Clone github.com/Kotohibi/3DGS_pipeline_guide to RAID before next CorbeauSplat session
2. 2. Order/print AprilTag plates for next field capture
3. 3. Watch kotohibi replies on the commissioned-service post for pricing data
4. 4. Update splat-cheatsheet memory

CAVU positioning hardens; watch replies for pricing data

1. 1. Read replies for pricing data
2. 2. update splat-cheatsheet pricing memory

Strengthens CAVU splat-services product-readiness vs kotohibi offering

1. 1. HIGHEST PRIORITY: clone + test against Building C interior 360 set

SMB floor = $300-800; CAVU drone+integration must price 5-10× above

1. 1. Update splat-cheatsheet-stammany pricing memory with SMB floor

Materially shifts post-COLMAP plan if it works at 1000 images

1. 1. Clone on Mac Studio
2. 2. benchmark vs COLMAP on Building C interior set

Free surrounding-area-context layer for site-portfolio proposals

1. 1. Pull repo URL
2. 2. run Venice LA + St John + Hastings test pass

DROPS DIRECTLY INTO CORBEAUSPLAT AS THE POST-BRUSH STEP. Brush handles cleanup but no collision, no LOD, no decimation. SplatTransform replaces or augments Brush: GPU floater filter -> progressive merge decimation -> Morton reorder -> export GLB/SPZ/Voxel. CorbeauSplat output gains a free collision mesh, which unlocks every downstream interactive deliverable. CRITICAL FOR BUILDING C PHASE 5 TRACK 2: the callout/measurement UX previously requiring Nira (paid) now works against any glTF viewer using SplatTransform's collision output as click-target geometry. Combined with Cesium plan, Building C exterior gets render + collision-backed callouts + LOD streaming on a $0 SDK stack. Materially shifts Track 2 economics. PRICING: don't pass savings to client — capability is the pitch.

1. 1. Install on Mac Studio: npm install -g @playcanvas/splat-transform
2. 2. Test against Building C exterior splat: decimated GLB + voxel collision + SPZ for Cesium
3. 3. If quality holds, make it canonical CorbeauSplat post-Brush step
4. 4. Re-cost Building C Phase 5 Track 2 (Nira eval line item likely zeros)
5. 5. Pilot collision-backed callouts as the Phase 5 measurement UX

Cluster (sim 0.70 to FreeGaussian 423985f6, sibling Vision3D 56bf5ef7, 360-drone cluster 8c525ba4). PRICING/POSITIONING signal: $200-500/property Matterport workflow now phone-app-table-stakes. CAVU's site-twin pitch must differentiate on quality + integration (measurement, callouts, BIM-via-APS, Cesium delivery), NOT raw 3D-tour-from-iPhone.

1. 1. Identify actual tool from linked video (likely Polycam/Scaniverse/new entrant)
2. 2. iPhone walkthrough + FreeGaussian quick-look benchmark on Building C garage or one CAVU office room
3. 3. Re-frame Matterport-replacement messaging on next site-twin proposal — emphasize value-add layer not raw scan

Cluster (sim 0.74 to FreeGaussian 423985f6; sibling d98dea07 Cesium, 35ac78fa Moonlake, 97b02e07 SceneMaker — 6+ in 14 days). NEW signal: composition + multi-splat scene editing + web-publish all in one tool — closes capture-to-deliverable loop. Sits between FreeGaussian (capture-only) and Cesium (delivery-only). Open question: real+synthetic splat mixing for proposal pre-viz.

1. 1. Trial-account, run Building C exterior reference set through Mint
2. 2. Eval composition + web-publish quality vs Cesium + custom viewer (Plan A)
3. 3. Test text/image-input synthetic-prop gen as splat — composes with real captured splats

Cluster MEMBER (parent iPhone-only 3D tour scanner, om_patel5 source, May 3 — same underlying tool, similarity >0.81; siblings RohanMital2 maker post + TimJayas use-case). NEW signal: 24h multi-language amplification velocity — phone-3D becomes table-stakes faster than expected.

1. 1. Trial Travvir on Building C garage interior
2. 2. Update CAVU site-twin proposal language BEFORE prospects cite this back
3. 3. Emphasize value-add layer (drone+ground+360+measure+BIM+custom viewer)

Cluster PARENT-IDENTITY (parent iPhone-only 3D tour scanner, om_patel5 source — same underlying tool; similarity >0.85). NEW: tool name + AI model name + splat substrate (World Labs) all confirmed. World Labs = new vendor on the watch list.

1. 1. Install Travvir, walk Building C interior, compare vs RealityScan 0%-aligned and FreeGaussian.ai paths
2. 2. Add World Labs to vendor-watch alongside Cesium/Nira/OVR
3. 3. Update CAVU positioning

DIRECT FIX for Building C interior 0%-alignment failure (memory building-c-realityscan-vagon-handoff). Likely path to Phase 5 Track 2 deliverable.

1. 1. Pull Google Slide PDF to RAID
2. 2. document the kotohibi recipe in CorbeauSplat playbook
3. 3. plan next Building C interior recapture using hybrid rig

DIRECT FIT for VPC stack (Astro 6 + MapLibre + Cloudflare). Replaces hand-rolled tippecanoe + custom hosting with R/Python-native + DuckDB-driven tooling that targets the same Cloudflare backend CAVU is using. Pairs with Google Earth (e12bd13a) for distribution and GeoAgent (today) for pre-tile analysis.

1. 1. Install freestiler venv on Mac Studio
2. 2. Pilot a VPC layer or Cameron Parish nest layer
3. 3. Validate Walker's R2+Workers config
4. 4. Pencil PMTiles delivery into VPC + St John proposals

POSSIBLE BUILDING C INTERIOR UNBLOCKER. Building C interior was blocked at 360-still + video-frame failure (2026-03-20). 360-drone capture into this pipeline could solve it. LFS is a Brush-class renderer worth evaluating as parallel path or replacement. kotohibi tools are open on GitHub. For PTFI/VPC, fewer drone passes per scene with full-360 capture.

1. 1. Verify LichtFeld Studio openness/pricing
2. 2. Spec AVATA360 + Metashape + LFS pilot for Building C interior re-capture
3. 3. Audit kotohibi GitHub vs CorbeauSplat

Adds APS Viewer as a 3rd splat-viewer option (alongside Nira and CesiumJS) for clients with existing Revit/BIM. Strong as-built deliverable.

1. 1. Validate Brush splat output compatibility with APS Viewer
2. 2. pitch BIM+splat as as-built upsell on next SMG/PTFI brief

Brand-book deliverables: production cost collapses, market price holds — margin grows.

1. 1. Spin up TMR brand-book trial as first end-to-end test of this workflow.

Strategic for MixMatrix v3 — second supply-side source if API exposes plays/BPM/key. Spotify-alternative framing.

1. 1. Sign up
2. 2. check API docs (BPM/key extraction)
3. 3. A/B vs ACE-Step for Neon Nocturne

BIM+splat fusion = higher-value as-built deliverable than splat-alone. Charge BIM-integration premium.

1. 1. Validate Brush splat → APS Viewer format compatibility (parent memory action).

FREE OPEN-STANDARD ALTERNATIVE TO NIRA. Building C Phase 5 Track 2 was set to evaluate Nira (commercial) for 3DGS+360 hybrid viewer. Cesium now offers same capability with $0 SDK + commodity-priced ion hosting. CorbeauSplat gains a standard interoperable export target (SPZ-compressed glTF). Digital Twin Platform research can revive — LOD splat streaming was the missing piece for city-scale twins. For VPC, PTFI, and Cameron Parish proposals, can now offer 3D site twin add-on at fraction of prior commercial cost while retaining vendor-independence (open glTF means client never locked in). Building C exterior plan locked Potree+model-viewer hybrid 2026-04-17 — that decision should be revisited since Cesium handles mesh AND splat in one stack.

1. 1. Re-evaluate Track 2 with Cesium as primary, Nira as fallback
2. 2. Test KHR_gaussian_splatting export from CorbeauSplat into CesiumJS viewer using sample Building C splat
3. 3. Revisit Building C exterior 3D viewer choice (Potree vs Cesium)
4. 4. Pencil 3D twin add-on into upcoming proposals where applicable

GOOGLE EARTH BECOMES A DISTRIBUTION CHANNEL FOR CLIENT-FACING 3D/GIS DELIVERABLES. Pairs with yesterday's CesiumJS 3DGS announcement — GLB is the same glTF format that now supports KHR_gaussian_splatting. Single authoring source can target Cesium (interactive) AND Google Earth (reach). For St John Intermodal: site boundary SHP + master plan GLB massings + approach elevation profile = client-shareable Earth project link, zero install. For Cameron Parish: wetland nesting GIS data drops native into Earth for LSU team. For VPC: stakeholder-facing layer alongside MapLibre interactive viewer.

1. 1. Pencil Google Earth as deliverable channel into upcoming proposals (St John Intermodal, VPC)
2. 2. Test SHP+GLB workflow with Cameron Parish marsh data
3. 3. Author single GLB source asset that targets Cesium AND Earth
4. 4. Confirm whether Google Earth's GLB import handles KHR_gaussian_splatting extension or just standard mesh

POSSIBLY THE MOST IMPORTANT RADAR ENTRY TO DATE. Anthropic is becoming the cross-software glue layer for creative work — exactly the gap currently bridged by custom MCPs and skills (premiere-pro-mcp 269 tools, ableton-mcp, logic-pro-mcp, blender bridge). Native connectors will replace much of the improvisation. CorbeauSplat's Blender 4-plane bisect crop workflow becomes Claude-conversational. Building C splat-prep iterations compress. NOA pontoon Rev C iterations and Sentinel mount/enclosure design become declarative. For contract bids: workflow cost across multi-tool deliverables is dropping fast — PRICING SHOULD REFLECT CAPABILITY GAIN, NOT TIME-SAVED DISCOUNT (don't undercut yourself).

1. 1. Install Anthropic Blender connector
2. 2. Audit existing custom MCPs vs native connectors as they ship — identify what becomes deprecated and what retains unique value
3. 3. Re-price upcoming contract bids upward to reflect capability gain
4. 4. Track release dates for Adobe CC, Resolume, Affinity connectors specifically (highest CAVU media impact)

CONTRACT-RELEVANT, MULTI-PROJECT PRICING INPUT. Combined with AVATA360 pipeline (53ec3f8a), Cesium 3DGS LOD (d98dea07), and Google Earth GLB delivery (e12bd13a), CAVU now has a credible end-to-end 360-drone → Cesium-streamable site-twin offering at fraction of pre-2026 cost. Don't undercut on price; the capability is the pitch. "Fly within the street" maps directly onto VPC port docks. Mining-site captures (PTFI) need fewer passes.

1. 1. Adopt 360-drone-first capture posture for PTFI / VPC / Cameron Parish proposals
2. 2. Pencil paired non-360 detail pass when sub-cm GSD matters
3. 3. Watch DJI SDK ship

Replaces tier-1/2 motion graphics manual After Effects round-trip; Mike Moncrief retains high-craft. Margin grows.

1. 1. Install skill in Claude Design
2. 2. test on Entergy or wedding lower-third

PARALLEL QUICK-LOOK PATH to CorbeauSplat. The 360-input path may solve documented Building C interior failure (where COLMAP couldn't reconstruct 360 stills, 2026-03-20). Lowers proposal-phase POC cost: send a quick free splat to a prospect within hours of capture. Companion to Diego DT browser viewer post (same OVR team).

1. 1. Run Building C interior 360 captures through FreeGaussian.ai
2. 2. Quality-compare against CorbeauSplat outputs
3. 3. Add OVR to vendor-watch list (Cesium, Nira, OVR)

BROWSER-NATIVE CITY-SCALE 3D VIEWER IS THE VPC VISION. MapLibre is currently Plan A; OVR's viewer + FreeGaussian capture is Plan B. Pattern transfers to mining-ops dashboard (PTFI), 151ac site marketing portal (St John Intermodal). OVR is shipping the FULL stack — worth tracking and possibly partnering on white-label/referral basis.

1. 1. Wait for Diego's viewer to be public OR contact OVR for early access
2. 2. Add OVR to vendor-watch list alongside Cesium and Nira

REAL-TIME STREAMING 3D RECON SHIFTS THE WORKFLOW. Currently CorbeauSplat is hours-long offline batch (COLMAP + Brush). Streaming opens drone-flight-time alignment validation — reduces re-shoot risk on remote sites (PTFI Indonesia, Cameron Parish marsh, VPC port). Strategic: SLAM-meets-foundation-model trend will commoditize "send video, get geometry" within 12-18 months. Position CAVU services to consume rather than depend on bespoke COLMAP as a moat.

1. 1. Bookmark @YinghaoXu1 for code/weights release
2. 2. Re-evaluate CorbeauSplat pipeline architecture if real-time SLAM-from-foundation-model ships

Downstream object-isolation primitive (vs kotohibi upstream masking); enables clickable/measurable scene elements without Nira.

1. 1. Get repo (risingodegua.com)
2. 2. test 2D-mask to 3D lift on a CorbeauSplat PLY with known poses

Validates drone to city-scale splat at CAVU scale; aerial integration is the differentiator vs iPhone SMB scanners.

1. 1. Study Niantic+Spexi pipeline as wide-area benchmark
2. 2. add to vendor-watch (World Labs, Cesium, PlayCanvas)

Token-based Gaussian-budget control = the fidelity lever the pipeline lacks; gen-source so pre-viz tier not as-built.

1. 1. Read method
2. 2. watch code drop (logged in 4/4 sibling entry)

Apache-2.0 = legally cleanable to fold in; study token decoder / decoupled primitive-budget control.

1. 1. Clone github.com/nv-tlabs/TokenGS + pull HF weights on Mac Studio
2. 2. bench token-budget vs CorbeauSplat fidelity lever

Splat-native prop generation composes directly with captured 3DGS (no mesh→splat step); easy Mac bring-up given tiny dep footprint.

1. 1. Clone on Mac Studio, test ComfyUI workflow on forklift + mining-equipment refs, eval splat fidelity vs Tripo/Hunyuan/ml-lito.

Opacity-mixing gates generative fill to unobserved regions → repaired splats safe for free-roam marketing, but keep measurement callouts on captured regions only.

1. 1. No separate action beyond release entry
2. 2. informs the trust boundary for repaired-vs-captured splat regions

Hotspot→photo/info UX on the free $0 PlayCanvas viewer = Building C Phase 5 callout template (no Nira); sellable equipment-inspection service.

1. 1. Rebuild experience structure (free-roam + hotspot modal) on CAVU PlayCanvas viewer w/ existing equipment splat
2. 2. save reusable template
3. 3. fold into Building C Phase 5 Track 2

Higher-fidelity synthetic props for splat scenes + single-photo mesh objects for twins; decimate before web.

1. 1. Bench Rodin Gen-2.5 via Hyper3D vs Tripo/Hunyuan/ml-lito on forklift + mining-equipment refs

Template for pre-construction/pre-lease splat sales walkthrough as marketing-portal add-on.

1. 1. Save adnfamily tour URL as reference deliverable for real-estate proposals

Reuse proven Magnolia/Building-C orbit logic as the virtual-camera rig geometry.

1. 1. When testing render-to-splat, tune virtual camera coverage (multi-pitch dense orbits) over raw res

Adopt 'explore before construction' framing for development-marketing proposals.

1. 1. Reuse pitch line for real-estate/development proposals
2. 2. pair with render-to-splat path

Bridges CAVU survey + video worlds: splat pre-viz in Unreal for shoots; AprilTag ties to NexaMark.

1. 1. Watch kotohibi measurement method
2. 2. pre-viz-in-Unreal as low-cost add for video jobs on splat sites

Drop-in last-mile compression for every web GLB deliverable; cuts load time + CF bandwidth.

1. 1. Run Building C decimated GLB + OSM-city GLB through it
2. 2. verify fidelity at -99%
3. 3. make standard pre-deploy step

Add GMRT as a free bathymetry/elevation source; reuse in-browser PNG→ffmpeg export for $0 video deliverables; 'depth context layer' upsell for water-frontage sites.

1. 1. Pencil bathymetry/depth layer into Cameron Parish + VPC + St John proposals
2. 2. try GMRT + Three.js on a marsh/river bbox

Local multimodal pass for footage shot-logging + audio analysis + agentic memory enrichment, zero cloud spend / no client-data upload.

1. 1. Pull gemma-4-12B-it on Mac Studio (Ollama/MLX), test image+video shot-logging, audio transcription, x-radar tool-use loop.

Becomes the post-Brush splat editing/recentering step in Blender (the stack Ian already uses); export glTF KHR_gaussian_splatting to Cesium/PlayCanvas viewer.

1. 1. Star/watch Blender issue #159470 for release
2. 2. when it ships, benchmark splat import/recenter/export vs Brush+SplatTransform

x-radar cron (fetch->cluster->research->write) could migrate to declarative YAML + run-inspection; unifies API+agent+file ops for batch jobs.

1. 1. Review Claude Platform CLI docs on Mac Studio
2. 2. evaluate YAML-config + run-inspection for the radar cron + AI-memory sync. Watch GA vs preview

Mesh-cleanup stage for scan-to-mesh deliverables (when a client needs an editable textured mesh, not a splat); candidate cleanup step for Building C decimated OBJ before Nira/viewer.

1. 1. Trial on a Building C decimated mesh + a photogrammetry capture
2. 2. compare retopo quality/time vs manual Blender/ZBrush

Credible $0 QGIS alternative when deliverable targets MapLibre; author-once ship-to-web same renderer.

1. 1. Update Mac Studio GeoLibre to v0.7.0
2. 2. load VPC layers
3. 3. test WFS/WMS + Whitebox + project round-trip

Adoption-confidence signal — Cesium streaming substrate for site twins is actively maintained; pin Cesium track to 1.142/2.27.0.

1. 1. When the Building C Cesium track spins up, pin versions + skim release notes for 3DGS/3D-Tiles changes. Pair with installed CesiumJS Skills.

Internal pre-flight airspace-awareness layer for CAVU drone shoots (controlled airspace, restrictions, LAANC) without manual FAA portal checks; reusable FAA-to-SQLite pattern.

1. 1. Clone jlmcgraw/aviationMap, run the FAA-data updater on Mac Studio
2. 2. if usable, wire an airspace-check step into the shoot-scheduling workflow

$0 QGIS replacement for the case where the deliverable already targets MapLibre: author once, ship to web with the same renderer, no QGIS->web restyle. Pairs with freestiler (PMTiles) + GeoAgent.

1. 1. Install desktop build on Mac Studio
2. 2. load a VPC layer set
3. 3. test lidar + GeoAgent plugins
4. 4. eval project-persistence round-trip into the Astro/MapLibre stack

Stronger Opus on CAD raises first-pass fidelity for NL->CAD (Fusion connector, cad-skill). Caveat from coverage: visual-reasoning gains don't always equal 3D-solid correctness -- bench before trusting.

1. 1. 30-min head-to-head when next CAD task surfaces: same sketch/prompt through Opus 4.8 + Fusion connector vs prior baseline
2. 2. compare dimension accuracy. If it holds, default Fusion + cad-skill workflows to 4.8

This radar pipeline is itself a fan-out workflow -- candidate to re-express as a dynamic workflow for big bookmark batches. Fits codebase audits, doc-gen batches (PTFI), multi-file refactors. Token-intensive; reserve for genuinely parallel high-value work.

1. 1. On Mac Studio set Opus 4.8 + /ultracode
2. 2. pilot a dynamic workflow on a safe parallelizable task to learn orchestration-script behavior + token profile before relying on it for production cron jobs

Image-to-3D prop gen becomes a $0 local capability alongside Tripo/Hunyuan3D/Seed3D 2.0 (cloud). r&d-tier work shifts to ml-lito; cloud vendors stay for final-deliverable polish. Strengthens sovereign-stack positioning for privacy-sensitive clients.

1. 1. Clone apple/ml-lito on Mac Studio. Bench on forklift / mining-equipment / Vesta product reference photos vs Tripo + Hunyuan3D + Seed3D 2.0. If quality is competitive at $0, default r&d to ml-lito.

Lovable absorbs the 'Maps + Sheets + Gmail wiring' tier of web deliverables. CAVU positioning shifts toward design quality, Astro/Cloudflare ops, MapLibre+Cesium 3D, and pipelines beyond Sheets — not raw integration time.

1. 1. Trial Lovable on a small CAVU need (request-a-quote mini-app with Sheets inbox + Maps embed). Validate first-pass quality. Update tier-1 proposal templates if it ships shippable.

Add satellite-scale 'box-one, find-all' feature inventory as CAVU upsell line item for any client with public imagery + asset-inventory need

1. 1. Watch Kyle's profile + walker-data.com for open-source repo. Test on Cameron Parish wetland tile when released.

MIT license unblocks commercial use in CAVU deliverables. Adds 'single-image, must-be-faithful' tool to the 3D toolbox alongside multi-view splat path.

1. 1. Clone repo on Mac Studio, run HF demo on a Building C interior occlusion shot, compare to SceneMaker de-occlusion result.

Install accelerates any Cesium integration. Differentiator: 'we ship Cesium AI-coded deliverables correct on first pass.' Prereq for the Cesium-as-viewer-of-record track.

1. 1. Install CesiumJS Skills into ~/.claude/skills/ (via hubs.li/Q04h2FNk0 → GitHub). Validate: prompt Claude to write a CesiumJS viewer for KHR_gaussian_splatting GLB.

Internal-dev process improvement. Pilot before adoption.

1. 1. Trial on Charles install Phase 2-5 OR a Trade Lab phase

Enables photo-to-callout binding when productized

1. 1. Research-watch
2. 2. bookmark for repo/paper link

Direct overlap with CLI Cockpit's Driver pane (multi-session command center). Either folds into Anthropic's surface or forks to focus on what Anthropic doesn't ship (Autopilot, X-Radar widget, project dashboards).

1. 1. Opt into research preview on Mac Studio
2. 2. screenshot agent view surface
3. 3. write a one-page integration-vs-replacement assessment before next CLI Cockpit phase

Marketing language now matches the buyer surface CAVU's client work touches (SMG, Hastings, VPC, St John, PTFI). Borrow the pitch phrasing verbatim for proposals. Adoption confidence is high — feature is stable, Google is leaning into AEC enterprise.

1. 1. When next VPC or St John proposal draft surfaces, lift Google's pitch phrasing verbatim
2. 2. Adopt Google Earth as a standard proposal-phase deliverable channel for AEC/infrastructure clients

Adoption confidence signal — Cesium committed long-term, not a research drop. Safe to lean on Cesium's three-runtime story in client proposal language. No new technical capability; parent action items still hold.

1. 1. No new action — parent d98dea07 action items still stand: re-evaluate Track 2 with Cesium primary, test KHR_gaussian_splatting export from CorbeauSplat, revisit Building C exterior viewer choice

Pencil as future small productized CAVU service

1. 1. When next route-video need surfaces, hand-roll in MapLibre+CF

Adds Apple-native channel to distribution-channel list

1. 1. Visit App Store + maker site
2. 2. capture pricing/embed/splat support

Replaces manual OpenSCAD authoring step for fast-iteration prints

1. 1. Install skill on Mac Studio
2. 2. trial on Bark Blaster bracket or Sentinel mount

Pre-sales demo channel for SMG/Hastings pitches

1. 1. Trial Mint on Building C exterior reference set

Adds measurement layer to generation pipeline

1. 1. Test virality_predictor via Claude→Higgsfield MCP on a CAVU reel

Native-color LiDAR collapses photogrammetry-plus-LiDAR-fusion into a single sensor pass. Not a CAVU buy yet at this price tier, but adds Ouster to the vendor-watch list alongside XGRIDS PortalCam. When a client surfaces with sub-cm color-fidelity requirements (PTFI mining ops, port as-built compliance, marsh canopy mapping), REV8-class LiDAR becomes a sub-contractor referral. Strategic positioning: pitch the deliverable (segmented twin, callout-ready scene), not the rig — capture pipeline goes sensor-agnostic over the next 12-24 months.

1. 1. Watch for REV8 list price + lead time
2. 2. Watch for drone-mountable variants or partner integrations
3. 3. Watch for sub-contractor / rental-channel availability
4. 4. Add Ouster to vendor-watch list with XGRIDS, World Labs

TWO PILOT TRACKS. (a) CAVU INTERNAL: month-end close + statement auditor templates against CAVU's FreshBooks (existing OAuth in env). Reduce monthly bookkeeping if workflow holds. (b) VESTA DECK: pitchbook + comparable-company templates against the existing 11-doc Vesta bundle (2026-04-30) + ongoing VestaLease audit. Partially template-drive future investor-materials revisions instead of hand-rolling. Pairs naturally with existing ms365 MCP setup. CONSTRAINT: don't push sensitive CAVU/Vesta finance data through templates until workflow is validated on non-sensitive samples.

1. 1. Clone github.com/anthropics/financial-services-plugins on Mac Studio
2. 2. Audit which of 10 templates have minimal config burden
3. 3. Install the lightest one as a Claude Code plugin
4. 4. Dry-run against a non-sensitive sample
5. 5. If validated, pilot CAVU month-end close
6. 6. If validated, pilot a Vesta pitchbook iteration against the existing 11-doc bundle

Sibling to HeyGen HyperFrames + OpenArt Smart Shot. Chat-native surface (no install, no skill, no separate UI) lowers floor below MCP/skill tools.

1. 1. Watch rileybrown 28-min tutorial
2. 2. extract prompt patterns
3. 3. bake-off vs HyperFrames + Smart Shot on a known lower-third or slate deliverable
4. 4. document winner in creative-skills-asset-audit-2026-05-05.md

Adds Arrival Space to vendor-watch list alongside Cesium ion, Nira, Mint.gg, OVR, APS Viewer. Cluster member with parent 53ec3f8a (kotohibi AVATA360→LFS pipeline), siblings 8c525ba4 / e986622c / 23bb50cc / 3e4a4dee. Validates parent pipeline is in production. Pricing / embed / branding policy unknown — needs vendor-info fetch.

1. 1. Visit arrival.space directly
2. 2. If brandable + cheap, add as Plan-C distribution channel for Building C exterior splat
3. 3. Cross-reference parent kotohibi cluster

CLI = scriptable into cron/launchd/n8n pipelines (vs MCP which is Claude Desktop surface). Pairs with parent 8889ae72 (Higgsfield MCP), a55b9df6 (meta-ads brand-book framing), HyperFrames (motion). Completes the Claude → Higgsfield → HyperFrames automation chain for paid-ad creative. Open-source skills = composable templates analogous to ~/.claude/skills.

1. 1. Clone github.com/higgsfield-ai/skills, audit immediately usable skills
2. 2. Install Higgsfield CLI on Mac Studio + Mac Pro
3. 3. Pilot batch run on Adam+Emily save-the-date (cheapest test)
4. 4. Update higgsfield-integration-research.md and higgsfield-tools-guide.md with CLI + skills entries
5. 5. Cross-orchestrate with higgsfield-animated-slide-deck-workflow.md

Cluster (parent dbcc568e forge3d, sibling maptheclouds Lego City, e12bd13a Google Earth — 4+ in 14 days on browser-3D-terrain). NEW: country-scale R3F+WebGL streaming-budget validated; supports site-portfolio-scale (multi-county/regional) deliverables, not just single-site context.

1. 1. Pull source URL from AhmedShahnab profile/replies if open-source
2. 2. Pencil 'regional elevation overview' as upsell line on St John Intermodal + Cameron Parish proposals

Cluster (parent STream3R, SceneVerse++, LingBot-Map, A LoD of Gaussians, SceneMaker — 6+ in 14d on post-COLMAP recon research). NEW: addresses the LABELED-DATA gap (SceneVerse++ auto-annotated; NAS3R goes label-free).

1. 1. Archive repo for watch
2. 2. Re-eval alongside STream3R when either ships runnable code on real CAVU data

Cluster (parent OpenGeometry, sibling Pascal Editor, Drafted, Anthropic-Fusion-connector, Tripo+Claude+Blender, Jerrod Lew donut — 5+ in 14d on AI-arch / image-to-CAD). NEW signal: image-of-groundplan as input modality (closer to Drafted than Tripo); ForgeCAD as cross-vendor target.

1. 1. Next CAD task (PTFI/TMR/Sentinel): 30-min head-to-head Anthropic-Fusion vs ForgeCAD/Codex on same hand-sketch
2. 2. No install today

Counter-cluster to cloud video gen (~5 entries in 14d). Adjacent to local-music stack (ACE-Step, Stable Audio Open). NEW signal: local Mac stack hardening — sovereign creative-content infra reducing cloud dependency for r&d.

1. 1. Install via Pinokio on Mac Studio
2. 2. Bench v2.0 vs Phosphene 1.x on known prompt
3. 3. Head-to-head vs Seedance 2.0 cloud on same prompt — if quality ≥70% of cloud at $0, default for r&d

Cluster MEMBER (parent iPhone-only 3D tour scanner; siblings taziku_co + RohanMital2 maker post). NEW: use-case granularity refines which CAVU site types Travvir most threatens (small interior, trailer exterior) vs differentiated (drone-required wide-area).

1. 1. Use stories in CAVU positioning: 'phone apps cover X — CAVU integration covers Y'
2. 2. Pair with maker-post action items (no separate trial needed)

Cluster (parent Tripo+Claude+Blender, Gizmo, SceneMaker, Moonlake, Hunyuan3D MCP, Hyper3D MCP — 6+ in 14d on image/text→3D-object). NEW vendor: ByteDance enters 3D-gen w/ scale + GPUs + research budget startup vendors lack.

1. 1. Watch for Seed3D 2.0 to land in Higgsfield/Replicate/fal.ai/MCP within ~30d
2. 2. When it does, head-to-head vs Tripo+Hunyuan3D+Hyper3D+Gizmo on a forklift / sentinel-bracket prompt

Cluster (same author as freestiler, sibling QGIS OpenGeoAgent + GeoAgent + Topographical Explorer + Google Earth — 6+ in 14d on geospatial-tooling). NEW signal: ANALYTICAL/optimization axis (not tile-gen, not viz). Closes the loop: analyze → tile → distribute via Kyle's stack.

1. 1. Install R+spopt-r on Mac Studio
2. 2. Pilot on Entergy SLE territory or VPC dock-staging hypothetical
3. 3. Pencil 'spatial optimization analysis' as upsell line on VPC + St John + Cameron Parish proposals

VPC stack is Astro+MapLibre+Cloudflare. Adding 3D building context to the port master plan now needs zero custom Three.js — hit OSM Buildings tiles directly. Replaces the maptheclouds 'Lego City' custom-WebGL pattern when OSM coverage is sufficient. Cluster member with parent MapLibre Lego City + Topographical Explorer + freestiler PMTiles + forge3d.

1. 1. Pull maptheclouds source code if open
2. 2. Build OSM Buildings + MapLibre pattern as reusable Astro component for VPC
3. 3. Add to VPC architecture options as 'tier-1 free building context'
4. 4. Test St John Intermodal surrounds for OSM coverage quality

Cluster (parent a3483511 Anthropic creative connectors; sibling Tripo+Claude+Blender). NEW signal: agent-NATIVE not agent-bolt-on. Open-source claim + triple export covers more chains than commercial single-vertical agent-3D (Moonlake, Tripo paid). Watch: open-source claim unverified until repo surfaces.

1. 1. Pull repo URL from gen_holography profile/thread
2. 2. Audit license + install path
3. 3. Benchmark export quality on a single forklift / Sentinel-bracket prompt vs Tripo + Hunyuan3D MCP

Cluster (sibling c82b9325 HeyGen, eec2eb64 Pika Agents, 81b12b96 OpenArt Smart Shot, 18f7e7ef oliviscusAI — 5+ in 14 days on AI-video). NEW signal: not GEN, not local-SEARCH — URL-IN-CLAUDE understanding of any short-form social video. Different stack layer (research/competitive-analysis vs gen vs local-archive search).

1. 1. Trial Algrow TikTok+Instagram path on NSIT venue Reels + NextSecGame SEC TikToks
2. 2. If structural breakdowns (hook/cut/CTA timing) are useful, promote to pre-shoot research workflow for Cavu branded content

First open-source text-to-precision-BREP-CAD pipeline demonstrated end-to-end on the web. Authoring-side complement to Pascal Editor (mesh-approximated, `71e36366`). Pairs with Anthropic Fusion connector (`feb17abf`/`a3483511`) for paid engineering-grade path. BREP precision matters when fab/engineering accuracy is required (TMR, NOA pontoon) — mesh approximations don't.

1. 1. Clone OpenGeometry repo on Mac Studio
2. 2. Run sample text→floorplan→BREP via opengeometry-three
3. 3. Benchmark BREP fidelity vs Pascal Editor mesh output
4. 4. Decide default authoring tool per use case (free-precision: OpenGeometry; free-mesh: Pascal Editor; paid-engineering: Fusion+Claude)

Cluster member of Cesium 3DGS LOD ecosystem (sim 0.82 to parent d98dea07). Unreal-path relevance is highest for desktop-runtime kiosk/ops-display deliverables (VPC investor meetings, PTFI ops display). Browser-path (CesiumJS) remains default for web client deliverables.

1. 1. Skim tutorial when scoping any Cesium-for-Unreal POC
2. 2. Bookmark in Obsidian Building C Phase 5 Track 2 plan if Unreal runtime becomes a candidate

Cluster with Pika Agents, HeyGen HyperFrames (Claude Design + Codex), Higgsfield MCP. NEW signal: storyboard-explicit pre-viz layer upstream of Seedance — turns AI video gen into a pre-production tool, not just content. Use as brainstorm/board surface; Higgsfield-via-Claude remains client-orchestration pick.

1. 1. Run Smart Shot test with Adam+Emily save-the-date concept
2. 2. Compare output vs Pika Agents and Higgsfield seedance_2_0 on same prompt
3. 3. Watch for OpenArt MCP/skill release

Cluster with d98dea07 (Cesium 3DGS LOD) — XGRIDS is Khronos co-developer of KHR_gaussian_splatting, so this is the CAPTURE/AUTHOR side of the same standard. NEW signal: segmentation of resulting splat into editable objects, not just reconstruction. Crosses 'pretty cloud' → 'structured asset' line. Directly relevant to Building C Phase 5 callout/measurement UX (today's plan = Nira on monolithic splat).

1. 1. Pull LCC Studio download from xgrids.com/support, check macOS support + import format
2. 2. Test if LCC Studio accepts drone-photogrammetry inputs (would slot into CorbeauSplat post-process)
3. 3. Pencil PortalCam capture as referral/sub-line on high-end twin bids

Single-image walkable splat = lower threshold for proposal-phase mockups.

1. 1. Identify lab via WebSearch (likely World Labs/Tencent)
2. 2. watch for inference access

Cluster with 426a0ea0 (30-min vibe coding) + 53df5587 (24-min CC feature deep-dive). Adds MCP-design layer. Watch all three for ~75-min Anthropic-direct curriculum, then audit custom MCPs (mem0, ableton-mcp, ppmcp, blender) vs native connectors.

1. 1. Watch 18-min workshop + skim 50-MCP guide
2. 2. cross-check against claude-creative-connectors-2026-04.md and custom-MCP inventory

Cluster with d98dea07 (Cesium 3DGS LOD rendering). This is the TRAINING-side complement: Cesium solves render LOD streaming, this paper solves training LOD on consumer GPU. Specifically targets the Building C documented 0%-alignment failure (mixed-scale aerial+street). Watch for code release.

1. 1. Bookmark felixwindisch.github.io/ALoDOfGaussians/ for code release
2. 2. Pull Uni10K HF dataset for reference
3. 3. Once code drops, benchmark on Building C image set vs current CorbeauSplat/Brush pipeline

Cluster with a3483511 (Anthropic creative connectors) + f73626d9 (Jerrod Lew donut). Validates multi-tool agent orchestration, not just single-tool. Compare Tripo vs already-installed Hunyuan3D + Hyper3D Blender MCPs to pick CAVU default.

1. 1. Compare Tripo MCP vs Hunyuan3D vs Hyper3D on a known prop (e.g., generic forklift for VPC)
2. 2. Pick default for CAVU prop-gen
3. 3. Wire into CorbeauSplat + Building C proposal-render workflow

Cluster with 1f231598 (QGIS OpenGeoAgent) — same opengeos org. GeoAgent IS the framework, OpenGeoAgent is the QGIS plugin surface. Pair with freestiler (today) for tile delivery and Google Earth (e12bd13a) for distribution.

1. 1. Install GeoAgent via pip (Anthropic provider extra) on Mac Studio
2. 2. Test against a Cameron Parish marsh tile
3. 3. Pencil into proposal pre-step for any GIS-touching CAVU bid

Cluster with af476340 — operational threat model layer on top of the prior 'reduce prompts' settings article. Anthropic-CISO sourcing: 90% of Anthropic's own code is written by Claude, so they had to solve this for themselves.

1. 1. Pull deny-rule config from video
2. 2. Diff against ~/.claude/settings.json
3. 3. Push to all CAVU machines via network-sync-procedure
4. 4. Bake into Gene/Charles/Allen/Keller onboarding scripts
5. 5. Reference in VestaLease audit deliverables

Fusion is the SINGLE creative-connector item with strongest engineering-community traction (vs Blender, Adobe). Engineering teams are paying attention. NOA pontoon Rev C iterations + Sentinel enclosures become Claude-conversational.

1. 1. Set up Fusion connector before next NOA pontoon iteration
2. 2. Spec a Sentinel enclosure design through Claude+Fusion as stress test

Strengthens the 360-drone equipment-add case for proposals where canopy is in scope. Pencil into Cameron Parish marsh + cypress canopy and PTFI forested approach corridors.

1. 1. Pencil 360-drone line item into Cameron Parish proposal
2. 2. Cross-reference parent 53ec3f8a / 8c525ba4 / 23bb50cc cluster

Pure research signal. CorbeauSplat is committed to 3DGS pipeline and Cesium glTF export (memory d98dea07). If Power Foam ships open code with quality wins, pipeline target may shift in 6-12 months. Don't pivot now; track inference + viewer code + glTF/Khronos standardisation.

1. 1. Bookmark powerfoam.github.io for code release
2. 2. Re-evaluate pipeline target if (a) inference code ships, (b) viewer SDK exists, (c) glTF extension proposed

Adds "scripted GIS analysis from prompt" line to deliverable menu for site-planning clients. Replaces hand-built QGIS toolboxes for proposal-phase analyses. Pairs with Google Earth (e12bd13a) as distribution channel — QGIS for analysis, Earth/Cesium for client-facing.

1. 1. Star github.com/opengeos/GeoAgent and watch for public release
2. 2. Stand up plugin on Mac Studio when shipped
3. 3. Test on Cameron Parish nest-buffer + cypress-overlap analysis

Replaces Blender for terrain/aerial-context renders in proposal phases. Scriptable, batch-friendly, $0.

1. 1. Install forge3d on Mac Pro
2. 2. test Cameron Parish marsh DEM + Wallace site DEM before next proposal cycle

Adds 'native app companion' as near-zero-cost upsell on every website proposal. Submission/maintenance still chargeable.

1. 1. Test Shipper against an existing CAVU website (e.g. tchefunk.com)
2. 2. draft an 'app companion' line item for next website quote

Compounding-improvement value across every CAVU dev project. Workflow upgrade.

1. 1. Watch the 24-min video before next major dev session
2. 2. update claude-code-2026-features.md with newly-learned features

Click-event for actually using Higgsfield from CAVU's Claude-centric stack. Marketing+Cinema Studio now agent-accessible (not just gen models).

1. 1. Install Higgsfield connector in Claude Desktop on Mac Studio
2. 2. verify web-credit billing applies
3. 3. test Adam+Emily save-the-date loop

Forward-looking value-add for any client e-commerce/paywall/subscription engagement. Strategic, not tactical.

1. 1. Add 'Stripe Link agent support' as a no-cost differentiator line item in next monetization-related proposal

Articulated focus is novel vs Anthropic-Blender connector (which has shown text→scene + image→donut, but not articulated mechanisms). Watch.

1. 1. Visit moonlakeai.com
2. 2. capture pricing/access model + export formats
3. 3. re-evaluate when SaaS access path is clear

Long-term replacement for COLMAP step in dynamic-scene captures (animals, water, machinery).

1. 1. Bookmark for code release
2. 2. head-to-head vs COLMAP on existing CorbeauSplat data when public

De-occlusion could fill splat dropouts behind objects without re-capture.

1. 1. Clone repo
2. 2. test on Building C interior shot with known occlusion

Reference-driven design starter for any client web project; clone-then-modify flow.

1. 1. Locate repo from author profile/replies
2. 2. install on Mac Studio
3. 3. test on SMG-tier industrial site clone

Compounds time savings across every dev session via reduced permission prompts.

1. 1. Diff against current ~/.claude/settings.json
2. 2. cherry-pick missing safe allows

Pair with Stripe Link agent wallet — completes the buyer+infra story for agent commerce.

1. 1. Reference in next agent-feature design for client tools
2. 2. CF Pages already default deploy target

Native app companion' SKU at near-zero gen cost; charge for app-store ops not generation.

1. 1. Test against an existing CAVU client website to validate output fidelity.

Could replace large parts of Blender Cycles render chain for slurry-pit / drone-DEM workflow.

1. 1. Install on Mac Studio
2. 2. benchmark vs current PTFI slurry-pit DEM render

16K resolution = poster-print viable. Replaces Blender Cycles for slurry-pit overview posters.

1. 1. Pair with sibling forge3d entry
2. 2. one install covers both bookmarks

Pair with prior Anthropic vibe-coding talk for ~1hr Claude Code mastery series.

1. 1. Watch back-to-back with prior 30-min talk before next major dev session.

Different use case from existing notebooklm-py-setup memory (which is podcasts/videos/quizzes). Counter-intuitive: bigger docset = more savings vs in-context. Adjacent to Context7 (public packages) — NotebookLM covers private/closed corpora.

1. 1. Resolve actual MCP repo URL from tutorial
2. 2. Confirm not officially Google before installing
3. 3. Pilot one large corpus (Cesium docs)
4. 4. Measure answer quality + latency + fair-use
5. 5. If clean, add to standard CC stack

Auto-annotation pattern relevant for Building C Phase 5 Track 1 interior tagging — could automate label suggestion. Dataset can train/eval splat scene understanding. Pure research today; productization expected 3-6 months out.

1. 1. Bookmark GitHub repo
2. 2. Re-evaluate when an inference/labeling tool ships on top

Quality-of-life infra. Eliminates manual checking on long-running CorbeauSplat training, RealityScan alignment, Building C exterior renders, AME batch exports, PTFI doc gens. Particularly valuable for cloud-VM watchdog patterns (Vagon, TensorDock).

1. 1. Enable push notifications on Mac Studio + Mac Pro + Mac Mini + MacBook Pro Claude Code installs

DECISION RULE for any new tool acquisition: prefer software that already shipped or pre-announced Claude connectors. Polyhaven mention relevant for environmental render work (PTFI slurry pit, Cameron Parish marsh).

1. 1. Apply connector-readiness as a tool-selection criterion going forward

Drone-engineering community traction. Chester is at droneforge — if drone-hardware teams adopt Claude+Fusion, expect rapid template/example library growth that CAVU work can leverage.

1. 1. Watch droneforge / chesterzelaya for Fusion-connector example workflows

Synthetic-360 mockups become a cheap pre-visualisation tool for proposals and Phase 5 Track 1 Kuula-overlay UX testing — author overlay logic against synthetic panoramas before re-capturing real interiors. Codex+Three.js path remains valid for fully custom viewers.

1. 1. Use as proposal-phase pre-viz path
2. 2. Test Phase 5 Track 1 overlay logic against synthetic panoramas
3. 3. Verify whether OpenAI exposes the rendered 360 as downloadable equirectangular asset

Replaces web-UI clicking with scripted agent calls. CAUTION: previous research (`higgsfield-integration-research.md`) documented that web-subscription credits don't transfer to API. Confirm before scaling — if MCP is web-credit-billed, fully usable; if separate API tier, costs need review.

1. 1. Confirm Higgsfield MCP credit / billing model
2. 2. Pilot wedding promo asset gen via MCP
3. 3. Update higgsfield-integration-research.md with MCP findings

Validates parent connector announcement (a3483511) is real and image-grounded. Reference-image → matched-3D-prop is the pattern needed for splat cleanup, architectural massing context shots, NOA pontoon Rev-C iteration, and Sentinel custom enclosures.

1. 1. Install Blender connector before next CorbeauSplat or pontoon iteration
2. 2. Spec a reference-image-to-geometry stress test against existing Building C reference photo

Alternative to Higgsfield MCP for short-form social/promo. Different UX (agent-as-character vs Claude orchestration). Same Seedance 2.0 model under the hood.

1. 1. Spin up a Pika Agent for CAVU brand
2. 2. cross-eval vs Higgsfield MCP on a sample wedding teaser

Direct workflow win. Replaces manual Premiere scrub-and-mark with text-query → exported clip → drop into bin. For Live Stream Rip, turns multi-day stream archives into queryable libraries. Sentinel could index Reolink RTSP archives without YOLO retraining.

1. 1. Get the actual repo name from oliviscus replies / pinned thread
2. 2. Install on Mac Studio
3. 3. Test against a Coachella raw-footage folder

Geometric-fidelity reference. Phase 5 Track 2 measurement / callout overlays now have a quality bar to hit. For Cameron Parish, marsh-nest survey could deliver localizable splat overlaying GPS-tagged nest data via drone-mounted Insta360.

1. 1. Set CorbeauSplat quality bar at "robot-localization-grade"
2. 2. Watch Niantic Spatial product roadmap (Asim, Azadux, Spatial Map)

HeyGen multi-homing across IDEs (Codex + Claude Design). Confirms going where users are.

1. 1. Use Claude Design version (next bookmark) since Ian's primary IDE is Claude Code.

Design-system rules become portable artifacts. For any client deliverable that includes a design system definition, DESIGN.md is the portable export format that decouples client from any single AI agent ecosystem. Adds optional deliverable line item to design-system contract bids.

1. 1. Track DESIGN.md spec location (Google Stitch repo or independent)
2. 2. Pencil DESIGN.md export as optional deliverable in design-system contracts

Generative-UI direction. Game UIs (Chuck's Quest) are exactly the use case where pixel-streamed model UI could feel native. ACE Studio UI Neon Nocturne theme could explore model-rendered moments. MixMatrix v3 Rethink is research-phase — can absorb new UI paradigms before commitment.

1. 1. Track Flipbook public release (zan2434 / South Park Commons)
2. 2. Creative-direction prompt for next Chuck's Quest session

Meta-skill upgrade. 30 minutes here likely returns hours saved across CorbeauSplat, Sentinel, MixMatrix v3 Rethink, daily Claude Code use. Direct-from-source training.

1. 1. Watch the 30-min video before next major dev session
2. 2. Re-resolve t.co URL to actual hosted source

Watch-only; adjacent to the browser-3D-authoring lane (cf. Pascal Editor). Not on critical path.

1. 1. None (watch).

DeepSeek = low-cost LLM option for GeoAgent/OpenGeoAgent GIS path.

1. 1. None (watch)
2. 2. reference if standing up LLM-assisted QGIS workflows

No direct CAVU pivot. Track underlying Google I/O GEO docs only if NSIT structured-data needs change.

1. 1. Skip consultant blueprint. Bookmark Google I/O 2026 GEO primary docs when they surface on developers.google.com/search.

No deliverable shift. Signals viability of an Unreal-runtime kiosk deliverable for investor/exec views if any project pivots to desktop runtime.

1. 1. Watch only. No active action.

Pattern: 3 specialist agents + consensus filter on Claude API. Wallet-scan→copy-trade-with-filter loop

1. 1. Watch only. Treat P&L as marketing. If Trade Lab adds a strategy-selection meta-layer, mirror the consensus-filter shape.

No action change vs parent

1. 1. None — duplicate parent

1. 1. Watch-only

Drafted's audience-crossover from AI-tools-educator audience to web3/founder audience typically precedes vertical-expansion into commercial/industrial AI-architecture by 3-6 months. Watch leading indicator.

1. 1. No action — track as a pacing signal for when AI-architecture tooling hits the commercial/industrial verticals where CAVU plays.

No new technical capability over parent. Watch-only. Confirms Google Earth's elevation-profile language is now the canonical pitch frame for terrain-aware site work — adopt the same three scenarios (route/transport, viewshed, machinery logistics) when describing CAVU site deliverables.

1. 1. Adopt Google Earth's three-scenario framing in proposals where terrain matters
2. 2. No tooling changes — parent action items still govern

Pattern transfers: public-data + polished operator-facing portal as a posture CAVU could adopt for any underserved Louisiana/Gulf Coast vertical. No active deliverable impact today.

1. 1. Track Kyle Walker as a builder reference
2. 2. file pattern as candidate posture if a Louisiana O&G or similar GIS-data-vertical opportunity surfaces

Cluster echo of parent Seed3D 2.0 entry. Nona_xai now appears twice on radar (DESIGN.md 2026-04-22 + this) — track as a recurring strategic-AI signal source. Wait for Seed3D 2.0 to land in Higgsfield / Replicate / fal / a Claude MCP before action.

1. 1. No action change
2. 2. Watch for Seed3D 2.0 in agent-accessible MCP/wrapper
3. 3. When available, head-to-head vs Tripo + Hunyuan3D + Hyper3D + Moonlake

Loose cluster with SceneVerse++ / STream3R / LingBot-Map (3D research tooling). Different signal: the live-training-viz UX pattern (viser triptych) is a reusable training-time inspection idiom.

1. 1. pip install viser for any future CAVU 3D-debugging session
2. 2. Use triptych pattern if a CAVU pipeline ever needs custom segmentation training

Loose cluster with digital-twin-platform watch list + 3D-recon cluster. NEW: APPLICATION-template (traffic, fixed-camera) at $0. Suggests fixed-camera-twin services are now feasible — possible CAVU service-line expansion.

1. 1. Archive repo URL in digital-twin-platform watch list
2. 2. When client RFP asks for live-traffic / live-flow monitoring, fork as starting template

Cluster (parent OpenGeometry + Pascal Editor 71e36366 — AI-architecture/browser-CAD theme, 3+ in 14 days). NEW: consumer-end-product UX (vs developer kernels). Reference-image remix = architecture-domain equivalent of Jerrod Lew's donut-from-image (f73626d9). Residential focus, not industrial — limited CAVU portfolio applicability today.

1. 1. File as adjacent industry signal
2. 2. Watch for commercial/industrial equivalent (re-evaluate then)
3. 3. Watch for glTF/3DGS export (synthetic-residential staging mockups in CorbeauSplat)

Cluster (parent a3483511 Anthropic creative connectors; sibling Mint, Topographical Explorer, Gizmo, Tripo+Claude+Blender, Pascal Editor, OpenGeometry, OpenArt Smart Shot — 10+ in 14d). NEW signal: external/peer validation that the cluster is a real trend, not bookmark-feed artifact.

1. 1. File as trend-validation entry
2. 2. Cite when client/stakeholder asks 'is this a real wave?'

Reference example, not a tool to install. Pattern is directly applicable to VPC stack (Astro 6 + MapLibre + Cloudflare) when port master plan needs animated ship positions / dock states / 3D massings beyond MapLibre native extrusion. GLSL conditional color is cheaper than texture-swap or per-state layer duplication.

1. 1. File link in VPC reference list
2. 2. Re-reference if/when In The Clouds source is opened or @maptheclouds publishes a tutorial

Cross-link to parent memories `1f231598` (giswqs OpenGeoAgent sneak-peek) and the GeoAgent framework memory. Hydrological-model angle is the new useful signal — directly applicable to Cameron Parish water-flow analysis and PTFI slurry-pit drainage. Voice control adds hands-free option for field use.

1. 1. No new action — already on action list from parent (install GeoAgent + OpenGeoAgent on Mac Studio for next Cameron Parish or PTFI proposal cycle)
2. 2. Watch youtu.be/5zkXQlHUsu8 before next install attempt