Integer interval decode reads the module's data rows; the 2^n ladder and digit primitives are carried tables (stage 1 of #11741's rewire program)

[briansrls]

What

Stage 1 of the rewire program behind #11741 (option-A ruling): the top within-claim recurrence on the required floor with the eval-frame memo retired.

integer_standard_integer_type_interval_spec called integer_machine_interval_spec on every demand — a second way to compute a bound beside the int8_interval_spec … uint128_interval_spec data rows that already carry each spec once per frame (their own annotation says the fn "remains the single derivation authority; these rows are its applications" — the fn had never been pointed at them). integer_standard_integer_type_from_node folded that fn over all ten standard types per decode, so one integer-node decode re-derived 2^127 and its nine siblings (a 127-step decimal-doubling ladder each). std.materialization_ladder rule 1: one computation identity, carried, never re-derived.

Now: the projection matches the type to its row; the decoder compares against a once-per-frame table of (type, interval node) derived by list_map from the one roster. integer_machine_interval_spec stays the single derivation authority; nothing new computes a bound.

One more carry in the same ladder, found by re-measuring after the first: integer_pow2_magnitude walks a once-per-frame ladder of 2^0..2^128 (built by the same doubling the recursion performs; index n IS 2^n by construction) so the ten widths share one chain's prefix instead of climbing ten ladders. The walk is structural — the nth Cons is 2^n, running off the end continues doubling from the last carried power — so there is no indexed lookup and no miss arm inside the carried range (review 68908). The digit primitives (decimal_digit_value and its inverse, decimal_digit_double) are deliberately untouched: making the inverse direction a second hand-authored table was refused as a §3 re-mint, and the right shape — one digit↔value row set with both directions derived — is its own change.

Semantics-preserving on its face

Same value by construction — each row's body is integer_machine_interval_spec(signedness: S, width_bits: W) with the literal S/W the two total matches produce for that type — and by execution: interval_derivation_matches_all_legacy_spelled_bounds (all ten types against hand-spelled bounds, an oracle independent of both the old and the new route) and from_node_round_trips_all_standard_types hold on the memo-on and the memo-retired binary. A direct old-vs-new equality claim was written, measured at 1.87M eval steps (it re-runs the ladder it exists to retire), and not enrolled: the legacy-bounds claim is the independent receipt. Memo-on cost of the first-touching claim per frame rises slightly (41k → 49k steps) because the ladder table is built to 2^128 once regardless of which widths a frame demands; every other claim falls.

Matched before/after on the claims it names

claim_batch --claim-run, same three modules, same session host; "before" is origin/main at dfa4644000, "after" is this head; measured under both realizations (memo-on = today's floor; memo-retired = the #11741 head).

claim	memo-on before → after (eval_steps)	memo-retired before → after (eval_steps)
interval_derivation_matches_all_legacy_spelled_bounds	41,146 → 48,318	941,846 → 379,036
derivation_discriminates_planted_off_by_one	1,631 → 258	2,465 → 319
pow2_digit_shapes_hold	34,407 → 1,814	337,172 → 1,826
interval_min_not_above_max_for_all_standard_types	37,728 → 275	933,618 → 275
from_node_round_trips_all_standard_types	43,189 → 6,286	7,631,294 → 16,334
well_formed_unbounded_carrier_still_decodes	238 → 238	244 → 244
decode_refusals_carry_distinct_reason_symbols	43,764 → 51,083	942,502 → 379,839
producer_fault_refusals_retain_the_lookup_cause	42,612 → 537	941,183 → 561
cross_language_compile_python_to_typescript_round_trip_holds	136,926 → 136,926	487,690 → 487,690
translate_refuses_underived_conj_holds	2,476 → 2,476	4,590 → 4,590
translate_refuses_underived_disj_holds	2,443 → 2,443	2,532 → 2,532
translate_refuses_underived_value_holds	2,446 → 2,446	2,535 → 2,535
translate_refuses_underived_bodied_arrow_holds	22,158 → 22,158	99,762 → 99,762
translate_refuses_underived_grammar_match_holds	16,009 → 48,567	45,734 → 365,936
translate_refuses_mixed_derived_root_underived_child_holds	37,564 → 37,564	3,852,779 → 3,852,779
translate_refuses_absent_parent_underived_child_holds	15,989 → 15,989	32,859 → 32,859
translate_refuses_bodyless_emission_root_with_poison_child_holds	15,906 → 15,906	1,502,339 → 1,502,339
translate_refuses_bodied_arrow_shortcut_underived_descendant_holds	7,666 → 7,666	17,369 → 17,369
translate_refuses_projection_absent_fold_bodied_arrow_underived_descendant_holds	7,908 → 7,908	17,745 → 17,745
translate_accepts_synthetic_empty_facts_tree_holds	9,462 → 9,462	14,687 → 14,687
translate_accepts_derived_corpus_module_holds	92,537 → 64,597	1,837,523 → 954,561
translate_derived_emit_byte_identical_to_before_holds	78,833 → 78,833	790,717 → 790,717
rust_binop_vep_emit_core_catalog_holds	88,462 → 95,684	1,711,850 → 1,149,090
rust_binop_vep_emit_discriminates_from_add	41,334 → 41,334	646,283 → 646,283
rust_binop_vep_emit_wrong_token_discriminates	29,803 → 29,803	60,588 → 60,588
rust_binop_vep_emit_missing_catalog_rejects	1,144 → 1,144	1,318 → 1,318
rust_binop_producer_emit_sub_holds	117,846 → 117,846	309,054 → 309,054
rust_binop_producer_emit_sub_discriminates_from_add	116,031 → 116,031	293,064 → 293,064
rust_binop_producer_emit_mul_holds	116,034 → 116,034	292,950 → 292,950
rust_binop_producer_emit_eq_holds	115,974 → 115,974	293,065 → 293,065
totals (30 claims)	1,319,666 → 1,195,600	24,047,357 → 11,969,947

Terminal results: 30/30 PASS before and after under both realizations — floor population unchanged, no claim moved its verdict. The translate_* / rust_binop_* rows are unchanged by design: their memo-retired cost has a different root (the formal_production_* grammar lookup family, ranks 2–10 of the floor census), which is stage 2.

Instrument for re-derivation: required_floor_cross_claim_demand.tsv of the run, within-claim repeats = evals − claims per producer row; the ranking is on #11741.

Generated-artifact side effects

The two non-.dag hunks (.gitattributes, docs/design-rung-drops.md) are the CI auto-heal regenerating projections that were stale on this branch's merge base. The removal of the POSTSCRIPT, 2026-09-20 line in docs/design-rung-drops.md is not this PR dropping operator text: that line was a hand-edit to the generated file (never in the .dag row), main itself removed it in #11789, and origin/main carries zero occurrences today; this PR's file is byte-identical to main's.

🤖 Generated with Claude Code

[gunbai-bot]

Reply to review 68856 (REQUEST_CHANGES), both findings verified and fixed in b4e9aa8: the annotation now names the instrument (the floor's demand census, within-claim repeats = evals − claims) and carries no transcribed figure; the decode table is list_map over integer_standard_integer_type_roster, so the closed set is enumerated once (an eleventh variant reaches the decoder through the roster, and the exhaustive match refuses if the row is missing). The PR title was the dashboard's session title, now replaced. Re-measuring after the first fix surfaced three more carries in the same ladder (pow2 table shared across widths; digit primitives as tables) — before/after table on the PR updated; the first-touching claim per frame is now 87k eval steps memo-retired (was 941k) and 49k memo-on (was 41k).

[gunbai-bot]

Reply to review 68885 (REQUEST_CHANGES), which reviewed the auto-heal head b227fd9e4c. (1) The transcribed figures were already removed in b4e9aa8c39 (the annotation names the floor's demand census and carries no count or run id) — grep 22,715 src/v2/std/integer.dag is empty on the current head. (2) The rung-drop postscript: verified, and it is not this PR's deletion to carry. The POSTSCRIPT was a hand-edit to the generated docs/design-rung-drops.md on main (added by 77b62fc, never in the .dag row), and main's own regeneration in #11789 (d482860844) removed it the same way the auto-heal here did; after merging current main this PR's docs/design-rung-drops.md is byte-identical to origin/main's (git diff origin/main -- docs/design-rung-drops.md is empty). Restoring it belongs in the row gunbc.rung_drop.namespace_wave_admission_wall_removed on its own change by that lane; I am flagging it to the parent rather than folding an unrelated authority edit into stage 1. — sent from merry-fox-260

[gunbai-bot]

Reply to review 68908 (REQUEST_CHANGES), all three findings verified and fixed in f5ca0df. (1)+(2) The digit primitives are reverted to main's derivations: the twenty-row doubling table and the hand-written inverses re-minted the digit↔value authority beside decimal_digit_value while decimal_digit_sub_borrow / decimal_nonzero_double still derived from it — agreed, a §3 fork; the one-row-set-both-directions shape is its own change and is not attempted here. (3) The 2^n ladder is now walked by structure (nth Cons is 2^n; running off the end continues doubling from the last carried power), so there is no list_at_optional and no miss arm inside the carried range — nothing to absorb. Re-measured: first-touching claim 941k → 379k eval steps memo-retired, 41k → 48k memo-on; 30/30 PASS both sides; before/after table on the PR updated.