T2: check-casts and residue gates, kept-cast macro set, standards, task log and summary

docs/gen3-standards.md

@@ -59,6 +59,17 @@ failure modes are different:
 4. **Types come from the code, names from observation.** Structure is recovered from access patterns the compiler locked into
    the bytes (many functions reading a `u16` at one offset from one base); meaning comes from observation. The 1,232 struct
    definitions are unified before any field is named; a duplicate type is a defect, a wrong name is a worse one.
+5. **Every raw cast goes, or names its cause.** (Phase 39 T2, 2026-10-01.) A `*(T *)(p + k)` is a member access the drafter
+   could not spell; finished code spells it as the member. Where the byte gate refuses the member (cookbook §458: `/s` moves
+   the scheduler), the cast is kept through ONE of the kept-cast macros in `include/common.h`, each expanding to the identical
+   cast and naming one cause class — `CAST_ALIAS` SCHED-ALIAS, `CAST_SIGN` SIGN, `CAST_WIDTH` WIDTH (overlap loser),
+   `CAST_MISALIGNED` MISALIGNED, `CAST_NONSTRUCT` NON-STRUCT base, and `LOBU/LOH/HIH/LOHU/HIHU/LOW/LOWU` REINTERP (an lvalue
+   read at another width or sign). A macro site is honest only with its restruct ledger row (rung S, site verdict
+   `KEPT(<class>)`, keyed tu + function per R99); `tools/type_census.py --check-casts` counts raw sites by form, macro sites
+   by name against the ledger, and lying declarations against `config/lying_exceptions.tsv`, and fails on any raw site, any
+   unbacked macro site or any unledgered lie. `M2C_FIELD` is a raw form, never a kept spelling. Each macro's byte proof (one
+   kept site respelled, `make check` BYTE-IDENTICAL) is recorded in the Phase 39 T2 summary. Levers follow the same reading:
+   `tools/lever_census.py --strict --residue config/lever_residue.tsv` is the gate, the residue file the ledger of what stays.
 
 Every step is checked the only way this project checks anything: 218 binaries rebuild identical.
 
@@ -109,7 +120,8 @@ The order, each step gated by the 218 hashes and each sized by a probe before it
    every residual.
 
 **Definition of done for Gen3 (proposed; the Gen3 plan ratifies):** 0 pins or each one `// !FAKE:`-marked with its reason; 0
-macro bodies (**met at Phase 35**); one definition per structure; every renamed symbol with a recorded basis; clang-format clean; and, unchanged,
+macro bodies (**met at Phase 35**); one definition per structure; 0 raw casts, every kept cast a `CAST_*`/`REINTERP` macro
+site with its ledger row (`type_census --check-casts` green, Phase 39 T2); every renamed symbol with a recorded basis; clang-format clean; and, unchanged,
 218 of 218 byte-identical from a clean rebuild.
 
 ## 5. Related

docs/struct-map.md

@@ -1,8 +1,8 @@
 # The struct map — the types the fleet's cast sites need (Phase 37 T1)
 
-> **Generated by `tools/type_census.py` at `0cb2b6b522` (2026-10-01T16:47:07Z); never typed (R75).** Every raw pointer-cast dereference in `src/` was assigned to a base (a parameter, a local, a global by value or by address, an absolute address) and the bases were clustered by EVIDENCE only — the same global symbol, an argument passed to a callee's parameter, an assignment or return, a pointer field read, a shared body's fleet-wide identity — never by a similar offset signature. Each cluster is one type the tree needs. Names are placeholders (`Unkstruct_<addr>`: the lowest global instance's address, else the lowest member function's); fields are placeholders (`unk<HEX>`) until Phase 38 cites evidence.
+> **Generated by `tools/type_census.py` at `7845da8f03` (2026-10-01T17:32:55Z); never typed (R75).** Every raw pointer-cast dereference in `src/` was assigned to a base (a parameter, a local, a global by value or by address, an absolute address) and the bases were clustered by EVIDENCE only — the same global symbol, an argument passed to a callee's parameter, an assignment or return, a pointer field read, a shared body's fleet-wide identity — never by a similar offset signature. Each cluster is one type the tree needs. Names are placeholders (`Unkstruct_<addr>`: the lowest global instance's address, else the lowest member function's); fields are placeholders (`unk<HEX>`) until Phase 38 cites evidence.
 
-**19122 types explain 498863 of 502983 dereference sites (99.2 %); 4120 sites have a base the census could not classify; 857 types carry a width conflict.** Merges by evidence: param 38862, assign 16777, arg 10823, field 9896, ret-of 4334, return 1524, assign& 1183, arg& 270.
+**19123 types explain 498868 of 502988 dereference sites (99.2 %); 4120 sites have a base the census could not classify; 857 types carry a width conflict.** Merges by evidence: param 38862, assign 16777, arg 10823, field 9896, ret-of 4334, return 1524, assign& 1183, arg& 270.
 
 | # | type | sites | bodies | fields | size ≥ | width conflicts (+sign mixes) | instance globals | pointer globals |
 |---:|---|---:|---:|---:|---:|---:|---|---|
@@ -207,4 +207,4 @@
 | 199 | `Unkstruct_76f4299b` | 262 | 131 | 1 | 0x4a | 0 (+0 sign) |  |  |
 | 200 | `Unkstruct_3816caab` | 260 | 130 | 2 | 0x1c | 0 (+0 sign) |  |  |
 
-*19122 types in all; the table shows the 200 largest by site count. The full map with every type's layout (offset, width, sign, site count per field) and conflicts is `.run/P37/census/struct_map.json`.*
+*19123 types in all; the table shows the 200 largest by site count. The full map with every type's layout (offset, width, sign, site count per field) and conflicts is `.run/P37/census/struct_map.json`.*