docs: session summary 2026-10-04(c) - L3 attempted, parked green

Records that Durable SymTab.LookupAny/FindAny(*Any) resolution made L3
byte-compare (candiag, t_lower3 MATCH; suite 183/183), but the SymTab
additions broke self-hosting (NthParam used before declaration; stage-3
miscompile) and were reverted, along with the NkProcDecl op=0 and
CalleeName fixes.  Also notes the accidentally committed ASTDUMP (now
removed on HEAD) and rules out option 1 (shared QbeGen state).

Recommendation: do L3 via option 2 with zero SymTab surface (grammar
stashes local/param info into a Lower-owned table).

Gates: suite 182/182; fixpoint OK (3147849 bytes).

docs/summary_two-phase-session-2026-10-04c.md

@@ -0,0 +1,76 @@
+# Session summary — 2026-10-04 (c): L3 attempted to completion; parked green
+
+Branch **`ast-stage-c`**.  HEAD **`e8956b2`**.  Suite **182/182**;
+self-hosting fixpoint **OK** (**3,147,849 bytes**).  Working tree clean
+except the user's `compiler/toto.mod`.
+
+**L3 (procedures) is not enabled** — `Lower.CanLower` still rejects
+procedure-bodied units, so the tree stays green.  L0–L2 remain
+byte-verified.
+
+## What was achieved (then reverted, because it broke self-hosting)
+
+The L3 *physics* were solved during this session and **did** byte-compare:
+
+- **Durable symbol resolution.**  `SymTab.PopScope` keeps symbol nodes
+  allocated (they just become unreachable), so a new read-only
+  `LookupAny`/`FindAny` plus `ProcNParAny`/`ParamTypeAny`/
+  `ParamIsVarAny`/`ProcUidAny`/`ProcDepthOfAny`/`ProcResAny`/
+  `ParamNameAny`/`IsExternalAny` could resolve a name (and a procedure's
+  formals) **after its scope closed** — exactly what an end-of-unit
+  `Lower` needs.
+- **Two real bugs fixed** (also reverted with the rest):
+  - `NkProcDecl` op defaulted to `-1`; `CanProc` requires `0` for a
+    normal body → every `PROCEDURE` was rejected.  (Fix: set `0` at node
+    creation.)
+  - `NkCall` child 0 is an `NkDesignator`, not a bare `NkIdent`; calls
+    were mangling as `$_0`.  (Fix: `CalleeName` unwraps it.)
+- With those, `candiag` and `t_lower3` **byte-compared MATCH** and the
+  suite reached **183/183**.
+
+## Why it was parked
+
+The `SymTab` additions **broke self-hosting** and could not be
+stabilized within the session:
+
+- `NthParam` was used before its declaration (line ~2258) by the new
+  accessors → `invalid call` / `not a RECORD type` cascades under the
+  V3-built compiler.  Inlining the plink walk fixed *that* error, but:
+- stage 3 of the fixpoint still failed (`M2S.mod:46 not supported yet`).
+- Compounding this, an **`ASTDUMP` debug statement was accidentally
+  committed** into `M2.atg` in `v3-lower3-wip`, corrupting later builds
+  and wasting diagnosis time (now removed on HEAD).
+
+Option 1 (concurrent legacy+`L` emit) is also ruled out: `QbeGen` shares
+one buffer and one counter set, so interleaved emitters corrupt the
+legacy path (`docs/summary_two-phase-lower3-finding.md`).
+
+## The path forward (recommended)
+
+L3 does **not** need `SymTab` changes.  Do **option 2 via grammar hooks
+with zero `SymTab` surface**:
+
+- while each scope is open, have the grammar stash `(name, kind, ty)`
+  for locals/params (and procedure signature info) into a **`Lower`-owned
+  table**;
+- `Lower` resolves through that table, so it works at unit end without
+  touching `SymTab` and without concurrent emit.
+
+This reuses all the AST/call/header work already committed in
+`v3-lower3-wip` and avoids the self-host fragility that the `SymTab`
+additions triggered.
+
+## State / tags
+
+- Tags: `v3-lower0`, `v3-lower1`, `v3-lower2`, `v3-lower3-wip`,
+  `v3-lower-eod-2026-10-04`, and this summary's tag.
+- L0–L2 byte-compare tests: `t_exit`, `t_arith`, `t_lower1`, `t_lower2`.
+
+## Resume
+
+```sh
+cd ~/Projets/Projets-Modula2/MyWork/m2compiler-V3
+git checkout ast-stage-c
+cd compiler && ./build.sh && ./run_tests.sh   # 182/182
+cd .. && ./bootstrap/fixpoint.sh              # FIXPOINT OK
+```