STATE: Wave 46.C — autonomath.db → jpcite.db symlink (entrypoint §1.4)

Date: 2026-05-12 Branch: feat/jpcite_2026_05_12_wave46_rename_46c_symlink Worktree: /tmp/jpcite-w46-rename-46c Memory anchors: project_jpcite_internal_autonomath_rename / feedback_destruction_free_organization / feedback_no_quick_check_on_huge_sqlite / feedback_dual_cli_lane_atomic

Scope

Implement axis 46.C of the plan in docs/_internal/W46_autonomath_to_jpcite_rename_plan.md §3.2 (Symlink overlay). Add a compatibility symlink between the legacy /data/autonomath.db and the canonical /data/jpcite.db so both env consumers (AUTONOMATH_DB_PATH and the future JPCITE_DB_PATH) resolve to the same inode. Zero destructive op, zero PRAGMA probe, additive only.

Files touched

File	Change	Δ LOC
entrypoint.sh §1.4 (new block, before §1.5)	+35 LOC inserted	+35
tests/test_w46c_symlink.py	new file, 10 cases	+207
docs/research/wave46/STATE_w46_46c_pr.md	this STATE doc	+80
Total		~322

Net LOC delta against main HEAD 2658c116: entrypoint.sh +35, test file +207, STATE doc +80.

entrypoint diff (full §1.4 block)

@@ -25,6 +25,41 @@ if [ ! -d /data ]; then
   exit 1
 fi

+# 1.4. jpcite ⇄ autonomath compatibility symlink (Wave 46.C).
+# Brand rename strategy: jpcite.db is the canonical name going forward; old
+# AUTONOMATH_DB_PATH consumers still resolve through the same inode. Per
+# `project_jpcite_internal_autonomath_rename` and
+# `feedback_destruction_free_organization`: never delete or rename the
+# physical autonomath.db, only create the symlink when the new path is
+# absent. Per `feedback_no_quick_check_on_huge_sqlite`: zero PRAGMA /
+# integrity probe here — symlink ops are O(1) inode-only so boot stays well
+# under the 60s Fly grace window. Per `feedback_dual_cli_lane_atomic`:
+# additive overlay (`ln -sf` only when target missing) — safe against
+# concurrent boot.
+JPCITE_DB="${JPCITE_DB_PATH:-/data/jpcite.db}"
+AM_DB="${AUTONOMATH_DB_PATH:-/data/autonomath.db}"
+
+if [ -f "$AM_DB" ] && [ ! -e "$JPCITE_DB" ]; then
+  ln -sf "$AM_DB" "$JPCITE_DB"
+  log "[W46.C] symlink created: $JPCITE_DB -> $AM_DB"
+elif [ -f "$JPCITE_DB" ] && [ ! -e "$AM_DB" ]; then
+  # Inverse case (post-eventual-rename world): jpcite.db is the real file
+  # and autonomath.db is missing. Symlink the legacy path so old code paths
+  # remain transparent. Still no destructive op on either side.
+  ln -sf "$JPCITE_DB" "$AM_DB"
+  log "[W46.C] reverse symlink created: $AM_DB -> $JPCITE_DB"
+elif [ -e "$AM_DB" ] && [ -e "$JPCITE_DB" ]; then
+  # Both exist. If they resolve to the same inode (either is symlink to
+  # the other, or both are bind mounts of the same file), nothing to do.
+  # Otherwise split-brain — log a warning and continue; downstream §2
+  # bootstrap still operates on $DB_PATH (= AUTONOMATH_DB_PATH default).
+  am_inode=$(stat -L -c%i "$AM_DB" 2>/dev/null || stat -L -f%i "$AM_DB" 2>/dev/null || echo "?")
+  jc_inode=$(stat -L -c%i "$JPCITE_DB" 2>/dev/null || stat -L -f%i "$JPCITE_DB" 2>/dev/null || echo "?")
+  if [ "$am_inode" != "$jc_inode" ] || [ "$am_inode" = "?" ]; then
+    err "[W46.C] split-brain: $AM_DB (inode=$am_inode) and $JPCITE_DB (inode=$jc_inode) are distinct files — manual reconcile required; continuing with $AM_DB as canonical"
+  fi
+fi
+
 # Helper: compute SHA256 of a file (portable across Linux + macOS).

Bugs-not-introduced verify

Gate	Command	Result
bash syntax	bash -n entrypoint.sh	SYNTAX_OK
pytest new file (10 cases)	pytest tests/test_w46c_symlink.py	10/10 PASS
pytest pre-existing entrypoint suite	pytest tests/test_entrypoint_vec0_boot_gate.py	14/15 PASS (1 pre-existing failure on test_autonomath_boot_manifest_exists_and_is_empty_allowlist_by_default reproduces on main HEAD 2658c116 — unrelated)
boot-time budget (Fly grace 60s)	runtime test wall clock measured	<3s on macOS runner (block itself ≈ 0.0s — no disk read, only inode ops)
destructive op grep	grep -E '(^| )(rm|mv|unlink|DROP|DELETE FROM) ' §1.4	0 hits
sqlite probe grep	grep -iE 'PRAGMA|quick_check|integrity_check|sqlite3 ' executable lines of §1.4	0 hits
forbidden brand	"税務会計AI" in §1.4	0 hits
LLM API import	n/a — bash only	0

Boot-time argument: §1.4 calls [ -f "$AM_DB" ], optionally ln -sf, and optionally stat -L. All three are O(1) inode-level syscalls; the 8.29 GB file body is never read. This is structurally incapable of triggering the 9.7 GB quick_check 15-minute hang documented in feedback_no_quick_check_on_huge_sqlite.

Test cases (10)

Static / structural:

1. test_w46c_block_present_in_entrypoint — anchor string + env var names
2. test_w46c_block_uses_ln_sf_only_and_no_destructive_op — no rm / mv / unlink / DROP / DELETE FROM
3. test_w46c_block_runs_before_seed_sync_and_r2_bootstrap — 1.4 < 1.5 < 2. ordering
4. test_w46c_block_has_no_quick_check_or_integrity_probe — comments may reference the rule, but executable lines hold no PRAGMA / sqlite3

Runtime (bash entrypoint.sh true against tmp_path-rewritten /data):

1. test_w46c_creates_jpcite_symlink_when_autonomath_present_and_jpcite_absent
2. test_w46c_creates_reverse_symlink_when_jpcite_present_and_autonomath_absent
3. test_w46c_is_noop_when_both_paths_are_same_inode
4. test_w46c_logs_split_brain_when_two_distinct_files
5. test_w46c_is_noop_when_neither_path_exists
6. test_w46c_block_boots_well_under_fly_grace_window — wall clock <10s

Out of scope

• Physical mv /data/autonomath.db /data/jpcite.db — explicitly forbidden (plan §3.2, plan §9).
• AUTONOMATH_* → JPCITE_* env bridge — that is axis 46.D.
• SQL am_* → jc_* view overlay — that is axis 46.B.
• R2 backup key migration — Wave 60+, plan §10 OQ4.

PR target

main (admin merge after CI green). No automatic Fly deploy required because §1.4 is no-op for the current prod inode topology (autonomath.db exists, jpcite.db absent → first deploy creates symlink, cost ≈ 0 ms).