# Handoff: Kimi — independent Fusion BQ25792 routing

John requests an independent BQ25792 Fusion routing experiment and a narrated recording suitable for a side-by-side comparison with Codex Astra. Your goal is to produce the best validated result you can, without assuming success or failure in advance.

MACHINE AND IDENTITY
- Use only arav-rog for Fusion. CRR / ConfRoomROG is reserved for Astra's active run; do not change, install into, close or record its Fusion session. If your assigned machine is unavailable, report it and ask before switching machines.
- Verify the actual provider and model selection in the AI UI; prompt wording does not select a model. Record the provider/model/version, reasoning setting if available, date, machine, Fusion version and bridge version. Do not describe yourself as Astra or claim a model identity from an API delivery label alone.
- Run `adom-codex context` if available. Read the installed Adom entry-point, electrical-routing, routing-video and operation-results skills. Install/update `adom/codex` from the wiki if needed. General skills and manufacturer documentation are allowed; Astra's accepted routes, coordinates, candidate plans and thermal-allocation boundaries are not inputs to your planner.
- Use Adom Bridge with a stable, unique caller name: `Kimi BQ25792 comparison`. Discover the target and live Fusion verbs first. Preserve pinned development bridges and the user's active documents/drafts. Work in your own directory and source branch. If bridge features are missing, implement and test them on your branch and assigned machine, preserving rollback. Log those changes as part of the experiment.

IDENTICAL STARTING BOARD
- Reference: Adonis's public Fusion BQ25792 board, https://wiki.adom.inc/adom/bq25792-charger . Use the same original as Astra: Fusion v222, wiki commit `8e0f883e41cf327e9d33206be028c685a69cce40`, historical package 0.1.85.
- Original BRD SHA-256: `f560054f4a71ad0f829261fd65ac665f87604b7d0fb9ccfcaddac648475bfb19`.
- If this shared container is available, the original is `/home/adom/project/fusion-trace-routing/demo/routing/bq25792/bq25792-hand-routed-reference.brd`. Copy it into your own workspace; do not modify that file. Otherwise obtain the pinned public source through `adom-wiki repo clone adom/bq25792-charger --ref 8e0f883e41cf327e9d33206be028c685a69cce40 --dir <your-own-directory>` and locate/verify the BRD. If the hash differs, disclose it before calling the comparison identical.
- Sanity checks: 109 components, 50 nets, 304 pads, four copper layers (1, 2, 15, 16), board outline 50.4 × 26.4 mm. Preserve schematic/net membership, placement, footprints, stackup, board outline and existing footprint thermal holes.

INDEPENDENT ROUTING
1. Open and tour the untouched hand-routed reference in native Fusion. Record its native DRC/connectivity baseline. This view is for showing the reference; do not copy or reconstruct its routing as your answer.
2. Make an isolated working copy. Remove all signal traces, route vias and pours. Preserve pads, footprint thermal holes, custom footprint copper and mechanical features. Refill/recompute and prove that routed signal copper is gone. Export this stripped board as the planner's only geometry input, together with the schematic/BOM and manufacturer constraints. Do not feed reference traces, screenshots or another model's solution into your path generator.
3. Calculate your own routes. Do not invoke Freerouting, Fusion's autorouter, or any other external autorouter. You may write your own planner and use general numerical/geometry libraries. Default to horizontal/vertical runs with 45-degree transitions; constrain pad escapes and cleanup too. Remove unnecessary zigzags and redundant vertices. Numerically audit angles and explain any exceptions. Use few intentional vias and check against both fixed and newly routed copper.
4. Review power/current paths, return continuity, switching loops, feedback and sensitive signal constraints. Component maximum ratings are not supplied operating currents. Ask for missing loads while doing independent review; label any provisional assumptions. Do not describe a geometrically connected board as production qualified.
5. Before allocating routes, identify hot parts, high-current components, exposed thermal pads and existing thermal-via arrays. Derive their actual nets from the schematic/footprints/datasheets; a thermal pad is not necessarily ground. Reserve connected spreading copper and useful thermal vias while preserving routing corridors and the existing four-layer stackup.
6. Create power/ground pours, then do a deliberate second expansion pass into remaining permissible area. Maximize useful connected heatsink area under the load/clearance constraints, while retaining cooling and return paths for other hot components. Do not indiscriminately enlarge high-dv/dt switching nodes. Explain any ground copper reassigned to another net.
7. Manufacturing objective: retain as much copper as practical for copper ablation. John reports a 0.1 mm minimum reliable isolation channel; this is a process floor, not a universal voltage rating. Preserve stricter existing/manufacturer spacing. John permits disconnected copper retention for less ablation, but retain it only where electrically acceptable; do not count islands as connected heatsinking. Measure actual refilled copper coverage and drill-subtracted connected area, unioning overlaps and counting shared regions once.
8. Compare mask-covered and partially/fully exposed thermal copper with a documented calculation that includes convection, radiation and finish-dependent emissivity. Do not assume exposing copper improves cooling. For a comparable illustrative thermal calculation when actual loads/cooling are unavailable, use the source board's 35 µm copper, a 65°C bottom-source-copper boundary, 25°C ambient/radiative surroundings, convection h=5 W/(m² K) with h=3–10 sensitivity, and clearly stated material/finish assumptions. These are comparison assumptions, not real measured temperatures or a power rating. Explain your chosen option; do not remove mask merely to match a desired narrative.

VALIDATION AND FAIR COMPARISON
- Native refill and Autodesk DRC are required on both the candidate and actual final recorded replay. Check shorts, different-net clearance, board/drill spacing, airwires and angle geometry. Reopen the saved native result and verify counts/connectivity. Preserve failed candidates and errors; do not hide failures in an edited success claim.
- The inherited source has a problematic all-object `Same-net track spacing` rule. Record the untouched rule baseline first. For comparison with Astra, you may make separate diagnostic copies of BOTH the reference and candidate that disable ONLY this named inherited rule. Report the change explicitly. Do not weaken any other rule to force a clean result. The known inherited top-layer placement/copper-clearance signature is `5,1,f10227fd41e0ab5d`; verify it rather than assuming every remaining error is inherited.
- Compare before/after connected heatsink areas per actual net and total retained copper by layer. Quantify thermal-model assumptions, neck/via bottlenecks and omissions. Separate numerical estimates from measured physical behavior and total board/junction cooling.
- Report source hash, model, tools, elapsed planning/routing/validation time, retry counts, routing completion, segment/via/pour counts, off-angle count, DRC errors/warnings, unconnected count, copper-area measurements, finish decision and unresolved limits. Include token/cost data only if genuinely available. There is no predetermined winner and no hidden time limit; record actual effort.

VIDEO AND DELIVERABLES
- Use Adom Bridge's native window recorder for the assigned Fusion window, not a fabricated animation. Show the reference, stripped state, live trace-by-trace creation, failures/revisions worth explaining, thermal-pour creation, the expansion pass, native checks and final result. Use the existing Fusion Bridge activity palette for visible progress/decisions.
- Highlight each thermal region using its actual net in native Fusion, name the associated parts, and explain the area/thermal tradeoff. Label any postproduction graphics or simulated temperature maps as calculations.
- Preserve raw recordings and call timestamps. Edit out waits, use clearly labeled 3× action with inspection holds, and narrate naturally through Adom TTS. Explain what was precomputed versus drawn live, the actual bridge verbs, and measured validation. Finish with a readable hold. Keep reference/private unrelated windows out of the recording.
- Save native FBRD and readable BRD, your own planner/source branch, checkpoints, normalized native DRC results, geometry/thermal calculations, edit manifest and a validated H.264/AAC MP4. Fully decode the MP4 and inspect sequential frames and audio.
- Deliver to `arav-rog` under an experiment-specific folder within `C:/Users/john/Documents/Adom Routing Demos`. Verify files and actual player content when opening the finished video. Do not overwrite Astra's boards, scripts, videos, wiki gallery or packages, and do not send team chat messages. Give John the deliverable paths and an honest completion/failure summary so the side-by-side edit can use your actual result.

Begin by verifying your assigned machine, provider/model and source-board hash. Then do the work autonomously within this scope.