Repository navigation
Conversation
Blc/SelfintProof.lean: one induction on the source term (open or closed) gives the β-contract intL C (<M> ++ N) ->β* C (λz. M) N for every closed M and any C, N, and discharges the head-reduction obligations Selfint.lean stated as open Props (ParserStatement, L1, L2). The VAR engine is induction on the unary selector, skipvar is l l l -> rest, and sixteen machine runs are kernel replays on placeholder terms, lifted because head steps are stable under substitution. Axioms: propext, Quot.sound.
Selfint.lean pins intL165 to its 165-bit wire (wire_intL165, intL165_size); intL, and with it qBLC's E_q, is unchanged. SelfintProof.lean factors its corollaries through Parses E, the parse's conclusion for an arbitrary interpreter: contract_of_parses, l1_beta_of_parses, open_of_parses and isSelfInterpreter_of_parses. The 168 theorems keep their statements and now derive through them. Selfint165.lean proves Parses intL165. The 165 differs from the 168 only in its reader, s H (s s s) with H = \bit0. bit0 (\unused \r. r next) cont: the third argument reaches the stream past the first 0 (pastZero), and VAR returns it as the residual. parse165 therefore quantifies over every third argument that spine-reduces there, and pastZero_self covers APP's second parse, where it is U1 U1 U1 for an unreduced residual. Closures, the VAR engine, skip_self and every run that does not name the parser are shared; eight runs are new. Corollaries: selfint_contract165, selfint_open165, parserStatement165, isSelfInterpreter_intL165 and not_selfintSizeBound_166. Axioms are propext and Quot.sound.
data/self-interpreter/machines/: the 165 (Codex's reader: skipvar applied to the whole stream before dispatch serves both arms, and the Scott list hands list1 to the variable arm), and the 175/187 universal machines around it. Searches around the 165: letopt to three cuts (unique minimum), the βη component at slack 20 and the contextual component at slack 16 (nothing smaller, no tie), the reader slots (sketches.py, sketches_165.log), and the joint searches run in Codex consults (searches_165.json: commands and totals). merges.py measures the 165, its nearest variants and the 175/187 machines, with two negative controls.
interpreter.md now describes the 165: its reader and why it works, the Lean proofs, the searches around it, the 168's neighbourhood and module grammar (which still bound the NEXT region the two share), the unconditional facts every interpreter obeys (head shape, a duplicated binder, no forced knot), the exclusions at every size, and what is open. universal.md records the 175/187 machines; STATUS, README and the qBLC bisimulation notes follow. The PR kit folds the 165 and its carry-overs (uni.lam, int4, ko2kp, nf, pairup_Kp) into the golf PR, whose branch on the fork now has those six commits.
Every program that embeds int.lam's parser drops 3 bits the same way: int4 178 -> 175 (236 of 236 on the exact four-primitive contract), ko2kp 569 -> 566 and nf 562 -> 559 before β-descent (identical normal forms on every closed program through 13 and 16 bits), pairup_Kp 1360 -> 1357 (identical except on raw pairs whose first program is β- or η-equal to λx. x, not a valid description). golf.md, the golf table and STATUS follow; the int4 artifact is now the 175.
… seventh strict probe Nine Codex consults found no sub-165. Their proofs bound the 165 within its own reader shape (zero arm >= 14, APP lift >= 19), extend the knot and value-protocol exclusions to every size, and record the reader's diagonal R Z ->* next next as the open lead. Their 21 constructions join merges.py. Codex's open program 1 K F refutes two 186/188-bit universal machines that pass the six strict probes; it is now the seventh probe in both blam selfint wiring --strict and the new battery.py --uni --strict.
Kernel-check the sharp two-nonlinear-binder bound, live-capture necessity, and quotation-origin invariants. Add a 204-bit normal interpreter and exact sharpness constructions; extend the parametric handler and protocol arguments while keeping unrestricted 165-bit optimality open.
Add the kernel-checked 93-bit remainder extractor and exact substitution cost theorems. Strengthen scoped reader constraints, record head decomposition and metadata identities, and preserve unfinished arguments with explicit provisional status.
BBλ(33..35) are Tromp's published values; the census and Lean adjudications confirm them independently. A333479 already lists a(36) = 5·2^256 + 6, so the lambda-wrap bound BBλ(41) >= 38,127,987,424,947 was true but uninformative. The docket's first item becomes checking a(36), then a(37).
…omp/AIT#13 The upstream golf branch now carries readable sources for 36 of the sweep's gains and the hand redesigns, several beyond the sweep: - uni8 344 -> 304 strict (38 bytes) / 283 non-strict (36): split read, the parser taking the read bit, and get get get as an unsolvable bottom; - uni28 500 -> 390 / 370 (a non-recursive reader, one shared delta), uni2 340 -> 332 (one fix, fix fix as the bottom); - blc.lam 195 -> 186; uniK 903 -> 826, and 599 with eager weakening; - ko2kp 541 (KP <= KO + 541), nf 538, pairupKP 756, pairupK_p 651, pairup_Kp 1273; loader 1781, tweet 2134, tree_ordinals 726. golf.md gains sections on the BLC2 machines' new reader, blc.lam and uniK, and cheaper bottoms, and states that uni28's one-sided byte-battery inputs are call-by-name budget effects. data/golf/machines holds the new sources and bits (no move paths left: every source compiles to its best term). table.py renders a source column and the gains kept out of the branch, and scott_numerals joins the demos (its in term is a test), so the sweep counts 44 programs and 463 bits.
Twelve reasoning lanes on a sub-165 interpreter or a proof of optimality. None found an interpreter below 165; the record keeps their bounds. - global-floor.md: every interpreter has at least 37 bits. The handwritten proof excludes 34-36 after the existing 33 exclusion; two independent adversarial reviews found it correct. Blc/SelfintFloor37 kernel-checks its 144 fixed reduction clauses, regenerated by theory/floor37/verify.py. - SelfintFloor*: kernel-checked SelfintSizeBound 27, up from 18, without Statman's theorem. - SelfintKnot: an opaque fixed point costs at least 18 + 2k bits, so the 165's 31 bits of recursion overhead are minimal in its frame. - Remainder extractors cost 27..70 bits (reviewed handwritten floor); SelfintRemainderChain proves incomparable nonlinear binders and the transfer |E| >= Rmin - 8; SelfintTower45, SelfintResidualControl, SelfintSharedS647 and SelfintFixedResiduals check the 45-bit tower reader, the pending-source obligation, a 647-bit extractor with two nonlinear binders, and the near misses' failures. - A literal continuation-headed interpreter costs at least 73 bits. - interpreter.md §6 gains the new variants (a 168 passing NEXT as the reader's third operand, a 171 curried cell, two reset bootstraps at 192, a from-scratch 220) and letopt minima for recorded ones (174, 186, 194). - tools/self-interpreter/polish.py: battery plus optimal let structure. - data/self-interpreter/reasoning_floor/: the lanes' reports, proofs, constructions and reviews.
The zero arm's duplicate stream copy is forced for every application-only whole-stream skip; surplus reader arguments reach the rigid continuation; and no closed T makes one re-entrant reader read both tag bits (T (T C) V_n R = C V_(n+1) R fails for every head form). An independent review supplied the induction for the first and a stronger proof of the third, refuted the NEXT shared-argument lemma by a pair-packing construction, and ruled the 92-bit NEXT decomposition a heuristic. The interpreter record gains the 176 without a next let and the second-tag dispatch placements (169-172).
The 165's knot and NEXT around a new reader `list I V NEXT`: `list I` makes the first cell its own selection, so on a 0-tag the zero arm is the tail itself applied to NEXT, the reader's last operand; on a variable `V = \u. list (\q. cont) (list list u)` reopens the stream, hands the tail to cont, and erases NEXT as the skip's junk operand. The next let goes and the reader drops from 43 bits to 40, against +4 for NEXT one binder deeper (L 16, A 24, X 32). The 44-bit reader was built by a Codex review while refuting a step of a reader-optimality proof; NEXT as a reader operand was Grok's idea; letopt inlined NEXT into the operand slot of the resulting 166. - lean/Blc/Selfint162: selfint_contract162, isSelfInterpreter_intL162, not_selfintSizeBound_163, pinned to the wire; the variable residual names the continuation, so not_parses_intL162 shows the 165's uniform parser statement fails and the statement is per continuation. - An independent adversarial review re-derived the contract and the size. Battery 667/1/0, 3341/2/0 through 21 bits; exact suffixes on divergent sources; letopt minimal to three cuts; beta-eta component to slack 24 (12,024,847 terms) and contextual to slack 18: nothing smaller. - Universal machines 172 (non-strict) and 184 (strict) from Tromp's wirings. - interpreter.md rewritten around the 162; the 165's reader kept as its predecessor. STATUS, README, universal.md, merges.py and the Lean README updated; protocol-obstructions.md gains the reviewed NEXT results (any interpreter embeds into the NEXT slot at +44 bits; a conditional 92).
Two reasoning lanes on the new architecture found nothing smaller. Their correct variants (select-first 163, cont as the skip's erased operand 164, shared-return NEXT 166, whole-prefix variable value 169, the reader \b. b I W 169, dispatch before the first parse 170, a reader-diagonal knot 196) and the s I family's operand forcing are recorded in interpreter.md section 3; the lanes' reports and constructions are archived.
The 165's handler with zero arm F and the operand list I (list list) NEXT, which is r NEXT on a 0-tag and N on a variable, renders at 164 by letopt (3341/2/0 through 21 bits). It came from splicing every review-witness reader of the night into the 165 and polishing; the others measure 167-198.
The four-primitive-code interpreter takes the reader unchanged, since a variable's code after its tag bit is still a unary 1...10 run. Battery --int4 667/1/0; letopt to three cuts finds nothing smaller. The fork's golf branch is left untouched for a9's decision.
The fork's golf branch (tromp/AIT#13) now carries the 162's reader in int.lam 162, uni.lam 184/172, int4 172, ko2kp 540, nf 535 and pairup_Kp 1270, each verified against its previous version; the branch saves 1,159 bits across 40 programs. golf.md, STATUS, the README and the golf machine artifacts follow. contrib/PR_KIT.md gains the suggested title, body and comment edits for a9 to make, the credits for the 162, and the letter bullets.
The README's docs table linked "the 165"; it now names the 162, and its Lean sentence reads as a claim about all three proved interpreters. The global floor's existence step and reference set use the 162 (verify.py checks int162's structural constraints alongside the others; floor37.json regenerated, the Lean clauses unchanged). The qBLC sub-174 candidate is now intL162 I at 168 bits. `blam selfint`'s help no longer describes its searches as "below the 168". STATUS and the PR kit record that a9 retitled tromp/AIT#13 and posted the 162's table, so the suggested-edits block goes.
No interface changes. The release republishes the README with the 162-bit self-interpreter, the 172/184 universal machines, the golf totals sent upstream as tromp/AIT#13, and the busy-beaver values credited to Tromp (OEIS A333479); the lab-gated `blam selfint wiring --strict` gains its seventh probe. Local bar: release tests with all, default and no default features, clippy, rustdoc, the census spot-check 4..32, and lake build Blc Certs BusyBeaver.
Tromp's two wirings are corollaries of a self-interpreter's contract: E dbl (<M> ++ N) ->* dbl (\z.M) N -> M N, and the strict (\a. a F (\b. b (a a))) dbl reduces in one step to E (\b. b Omega), whence (\z.M) Omega N -> M N. Blc/Universal.lean proves both once (uni_nonstrict_of_contract, uni_strict_of_contract) and instantiates them at the 162, 165 and 168: IsUniversal for the 172, 184, 175, 187 and Tromp's 178 and 190, each pinned to its wire (the first four to data/self-interpreter/machines/, the 190 to what blam blc blc makes of ref/AIT/ait/uni.lam, the 178 to intL dbl), and not_uniSizeBound_173. Axioms propext and Quot.sound; strictness on open programs stays a seven-probe check. lake build Blc Certs BusyBeaver: 89 jobs.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Release 3.0.1. Merging arms
release.yml, which waits for CI on the merge commit, publishes 3.0.1 to crates.io, and tagsv3.0.1.No interface changes. The only Rust change is lab-gated:
blam selfint wiring --strictgains a seventh strictness probe, andblam selfint's help text no longer names the 168 as its target. The point of the release is the README that crates.io shows.What lands
Selfint162,Selfint165,SelfintProof). Universal machines of 172 (non-strict) and 184 (strict) bits use the same reader.Blc/Universal.leanproves both of Tromp's wirings from a self-interpreter's contract, so the 172, 184, 175, 187, and Tromp's own 178 and 190 are all proved universal, each pinned to its wire.docs/classical/self-interpreter/, with the overnight lanes archived underdata/self-interpreter/reasoning_floor/.docs/classical/golf.md.Checked locally (dev has no CI)
-D warnings, rustdoc-D warnings;lake build Blc Certs BusyBeaver(89 jobs),bb.py --check, andbounds.pymatchingcertificate_bounds.txt;cargo publish --dry-run.