# PoeTchi — BUILD CARD v1.1 (transforms unification)

2026-10-31 · N8toriousAi's order: "improve and then upload my full zig kernel to
google drive." v1.0 stays frozen as a release; v1.1 is a behavior-preserving
source improvement on top of it, proven below. The boot banner still reads
PoeTchi v1.0 — behavior is v1.0's; only the source is cleaner and truer.

## The improvement

**Transforms unification** — the last open refactor from the 2026-10-19
list. `mesh.zig` and `wells.zig` each carried an identical hand-rolled
`project()` with 4-decimal trig literals (0.8776, 0.4794, 0.9394, 0.3429).
Both now call one shared projection:

- `pyre.zig` gains `Mat4.rot_x` (same column-major pattern as `rot_y`) and
  the shared orbit projection: `ORBIT_VIEW = rot_x(0.35) · rot_y(0.5)` +
  `project_orbit()` — real `@cos`/`@sin`, one place to fix, no literals.
- `mesh.zig` / `wells.zig`: `project()` is now a one-line delegate to
  `pyre.project_orbit`. Call sites untouched.

## Proof

- **Full clean build**: `zig build` from an empty `zig-out` —
  `harakiri-ng` (1,816,720 B) + `harakiri-ng.bin` (531,200 B), zero errors.
- **New parity harness** `src/check_project.zig`: old formula vs the Mat4
  composition over 4,913 points × 3 scales = **19,652 checks, 0 fail**.
  Max rotated-coordinate deviation **0.000086** — exactly the old literals'
  rounding error, nothing else. Max screen deviation 1 px (an
  `@intFromFloat` truncation edge at the largest scale; individual dots
  only, visually indistinguishable, mathematically explained).
- **Host harnesses re-run after the change**: `check_reason` **24 pass,
  0 fail**; `check_bell` **13/13 green**; `check_neural` runs clean.
- **QEMU regression (`v10_regression.sh`, 16 checks): NOT re-run** — QEMU
  is not installed in this VM, and a claim needs the run. The refactor is
  confined to the viewport projection (no REPL, agent, net, or hand-code
  paths touched), and the freeze script stands ready on any QEMU machine:
  `./v10_regression.sh` against this tree. Recorded here as owed, not as
  passed.

## Still open (N8toriousAi's calls, untouched)

- **Specular term** — Blinn-Phong direct vs neural-only: his decision,
  recorded 2026-10-19, still his.
- Kernoid OS (AIRX-DESIGN.md) remains design-complete and dormant; the
  device hunt is parked by his order.

## Archive

`poetchi-zig-kernel-v1.1.zip` — the full tree (src incl. the new harness,
boot, build.zig, roadmaps, both build cards, regression script, fresh
binaries), excluding `.zig-cache` and `__pycache__`. Uploaded to Google
Drive (Harakiri Builds folder) 2026-10-31.


## Addendum — 2026-11-02

**QEMU regression PAID: 16/16 PASS, 0 failures.** The regression owed since 2026-10-31 was run in the Muse VM (QEMU 8.2.2, qemu-system-aarch64) against this exact v1.1 binary (531,200 B); no source under `src/` was modified. Full log: `test/poetchi-v11-regression-2026-11-02.log`. The banner still reads PoeTchi v1.0 (unchanged, as recorded above). This closes the last open item on this card and freezes PoeTchi v1.1 as the proven baseline for the Nimbusis line.
