memory/: scrub contextual leaks per the new public-fork lesson

Applied 'grep is necessary but not sufficient' to all five files in
linuxcnc/memory/ via human review (per the lesson added to
feedback_public_fork.md in the previous commit).

reference_fork_setup.md: clean. All content is fork-side build setup
or generic Linux/LinuxCNC knowledge.

reference_teleop_velocity_bug.md: scrubbed multiple contextual leaks.

- Symptom paragraph: removed 'With Z at HOME=500, re-homing after a
  jog landed at ~498 mm' (private machine config — Z homing position
  reveals downstream hardware) and 'Test 13 re-home consistency'
  (private test name). Generalized to 'sub-millimeter tolerance' and
  'millimeters of deviation'.

- Removed 'Why the test didn't catch this before' paragraph entirely.
  It described 'Phase B of this branch' (private session phasing)
  and the asymmetry between Z=0 (MIN_LIMIT) and Z=500 (MAX_LIMIT) —
  both leak private hardware configuration. Replaced with a
  generalized 'Why earlier tests miss this' explanation: symmetric
  jog/home test setups coincidentally cancel the drift; only
  asymmetric pairs surface it. Same lesson, no private leak.

- Removed 'How to verify' section entirely. It listed assertions in
  test_logic.sh (Test 8b/8c, Tests 13/17/21, sim_lin_enc.so refs)
  that are all private cncmill1 test infrastructure not present in
  upstream LinuxCNC.

reference_upstream_pr_workflow.md: scrubbed two minor leaks.

- Tool intro: removed 'commit added on branch fizzy-forging-stardust'
  (private session branch name from the upstream-pr-prep helper's
  development history). The merge SHA stays — that's public on the
  fork. Branch names that don't exist on the public fork after merge
  are private context.

- Code block comment: changed '/tmp is tmpfs on this machine, override'
  to 'override if /tmp is tmpfs on the host'. Removes the
  work-computer-specific environmental detail.

Both files remain leak-free per a comprehensive grep including the
new contextual-leak terms (fizzy, stonebraker, Phase B, Test 13,
test_logic, HOME=500, sim_lin_enc) plus the original literal-path
patterns.

The five remaining linuxcnc/memory/ entries are now all fully
fork-relevant with no private project structure leaks:
- feedback_public_fork.md (public-fork hygiene rule + contextual-leak lesson)
- reference_fork_setup.md (fork build setup)
- reference_teleop_velocity_bug.md (upstream bug + fix description)
- reference_upstream_pr_workflow.md (upstream PR prep procedure)

Author: igor

memory/reference_teleop_velocity_bug.md

@@ -8,9 +8,11 @@ originSessionId: 4b988890-e736-4841-9234-018b1517d7ab
 **https://github.com/LinuxCNC/linuxcnc/pull/3933** (`motion: axis_sync_teleop_tp_to_carte_pos resets velocity/acceleration`). PR branch `fix-teleop-tp-velocity-reset` on the fork is based on `upstream/master@a2a929ccaa` and contains exactly one commit (`47abd6ac66`), authored as `is-primary-dev <215415441+is-primary-dev@users.noreply.github.com>`. Fork master has the same fix as `f40e8e64bc` (via merge `e4a81b19a5`). If the upstream PR gets reviewer feedback requesting changes, push fixup commits to the same fork branch; it's still checked out locally for that purpose.
 
 **Symptom.** After jogging an axis and then re-homing it, the homed joint
-lands slightly BELOW `HOME` instead of exactly at `HOME`. With Z at HOME=500,
-re-homing after a jog landed at ~498 mm, deterministic per jog-settle
-timing. Test 13 "re-home consistency" fails with ~1-3 mm max deviation.
+lands slightly off `HOME` instead of exactly at `HOME`. The deviation is
+deterministic per jog-settle timing and proportional to the residual jog
+velocity at the moment mode transitions from FREE (homing) to TELEOP.
+Re-home consistency tests with sub-millimeter tolerance fail with
+millimeters of deviation.
 
 **Root cause** (in `linuxcnc/src/emc/motion/axis.c`):
 `axis_sync_teleop_tp_to_carte_pos()` updates `teleop_tp.curr_pos` and
@@ -38,17 +40,11 @@ with the function's documented intent. The fix is on fork master as
 commit `f40e8e64bc` (via merge `e4a81b19a5`) and submitted upstream as
 PR #3933.
 
-**Why the test didn't catch this before.** Before Phase B of this branch,
-Z homed at 0 (MIN_LIMIT). The test jog direction and re-home direction
-coincidentally cancelled the drift. When Z started homing at 500
-(MAX_LIMIT) the asymmetry exposed the bug.
-
-**How to verify.** The assertions added in test_logic.sh:
-- Test 8b/8c: `joint.{2,3}.motor-offset` within ±10mm (proves
-  HOME_INDEX_NO_ENCODER_RESET special case fired, catches stale
-  sim_lin_enc.so bug independently)
-- Tests 13/17/21: re-home consistency <1µm over multiple cycles
-  (catches teleop_tp velocity leak on mode transition)
+**Why earlier tests miss this.** A symmetric test setup (jog direction
+equal-and-opposite to the homing search direction) coincidentally
+cancels the drift. The bug only surfaces when jog and re-home
+directions are asymmetric. Re-home consistency assertions need to
+exercise asymmetric jog/home pairs to catch it.
 
 **Related lesson.** When a function is documented as "initialize X to a
 rest state," check whether it resets velocity/acceleration state, not

memory/reference_upstream_pr_workflow.md

@@ -4,14 +4,14 @@ description: Use agent-helpers/upstream-pr-prep.sh to cherry-pick a fork commit
 type: reference
 originSessionId: 795ab1b7-ad70-4ca0-8fe0-8a6903e238a7
 ---
-**Tool.** `agent-helpers/upstream-pr-prep.sh` on the fork (merged to fork master 2026-04-12, commit added on branch `fizzy-forging-stardust` via merge `89f87b5a40`). **Always invoke via the wrapper subcommand** so the PreToolUse auto-approval hook covers the call. Direct invocation of `upstream-pr-prep.sh` bypasses the hook and prompts per call.
+**Tool.** `agent-helpers/upstream-pr-prep.sh` on the fork (merged to fork master 2026-04-12 via merge `89f87b5a40`). **Always invoke via the wrapper subcommand** so the PreToolUse auto-approval hook covers the call. Direct invocation of `upstream-pr-prep.sh` bypasses the hook and prompts per call.
 
 ```
 ./agent-helpers/safe-agent-action.sh upstream-pr-prep \
     --commit <fork-sha> \
     --branch <pr-branch-name> \
     --message /tmp/upstream-commit-msg.txt \
-    --worktree /var/tmp/linuxcnc-pr-<branch>   # /tmp is tmpfs on this machine, override
+    --worktree /var/tmp/linuxcnc-pr-<branch>   # override if /tmp is tmpfs on the host
     [--build] [--push] [--cleanup] [--dry-run]
 ```