board+plan: R2IL macro-transfer finding, the machine-semantic-contract plan, and the reversible-crystal doctrine (§7) - #1027
Merged
Conversation
…ct v1 Two artifacts, one commit, per the board-hygiene rule. EPIPHANIES: E-R2IL-MACRO-VOCABULARY-TRANSFERS-ACROSS-COMPILER-AND-LANGUAGE-1 — the 33-macro vocabulary trained on two gcc binaries fires in an unseen gcc program at -0.6% density and in an unseen rustc binary at -4.7%, both outside a column-marginal-preserving shuffle null (20 seeds, multiset-asserted per draw). Deflations stated in the entry: 33/33 is partly sample size, the Rust harvest is capped at 200/548 functions, coverage saturates and density carries the finding. The entry also records that my own pre-registration missed in the hypothesis's favour. Plan: r2il-machine-semantic-contract-v1.md — answers the sibling session's 'how should R2IL be stored' question in five rules derived from shipped code (no private object graph; a V4 tenant physically identical to V3; the 12-byte register carved per le-contract section 3 and read by reinterpret; a shared palette, not FnIndex-per-macro; behaviour by address). Separates documentation (cited) / plan (falsified) / status (D-ids) as registers, reproduces the operator's dependency-layering sketch with its correction (r2sleigh uses R2IL as its contract; R2IL is backed by SoA V4; libsla stays as the oracle and is removable LAST), and sequences six waves each with a kill condition, W0 being the Custom(n) content-vs-reading census that gates the 0xC4 mint.
…o a testable overlay, demotion gate Records the converged framing as plan-register material, with the discipline the document itself demands: the six-layer vocabulary names only shipped structure (each line marked shipped/measured/unbuilt); the hexagonal reading stays FALSE/net-new for storage per the #1023 audit and is re-proposed strictly as an optional learned neighbourhood overlay with a defined A/B probe (metrics + kill condition), plus the honest gate that its cheap first version runs on the existing 33-macro co-occurrence graph before any new tissue is built; and crystallization (Explore->Learned->System) gains its measured admission shape from today's transfer finding plus the missing demotion path, with a two-sided can-fire / can-stay-silent falsifier, so the vocabulary cannot become a guard that only ever grows.
…or-converged System is long-term memory, never truth; truth stays with white matter (CLAIM boundary, B4 expansion, contract veto). Crystal is a reversible trust STATUS, with symmetric promote/demote criteria under one epistemic yardstick, required hysteresis (promote threshold above demote threshold, anti-flutter not leniency, values to be measured as policy pins), and the sharpened hex burden of proof: beat Morton measurably as a compute topology or leave — never a retroactive reading of the register.
…ened A corpus may falsify a crystal only after it has falsified the hypothesis that it is an incompetent instrument. Records the defect in my own first draft (the established macros deciding whether the corpus may demote them — a self-immunizing loop) as a correction rather than a silent fix, then the six-point corpus qualification with both control directions (known signal seen, known noise not seen), the leave-one-out form that lets an established macro actually die, the frozen CAL+/CAL- panel target (B4's corruption demo named as CAL- seed material), and demotion-by-accumulated-evidence under the 7.4 hysteresis so one strange binary cannot empty the library.
… task Supersedes 7.2's cheap first version as the primary form of PROBE-GREY-TOPOLOGY-AB: fold intra-ontology structure only, with the operator's leakage fence extended past explicit crosswalks to any derived feature that betrays them; score emergent basins against the withheld curated mappings; four nulls including the is_a-depth- preserving leaf permutation that kills common-root attraction; the 7.5 instrument qualification applied to the oracle itself so a broken oracle gets no veto; and a two-phase sequencing where novel basins become admissible only after known withheld relationships are recovered, entering as Explore-status crystals rather than findings. The data half stays in the consumer repo's own ledger.
…atim The plan defines the experiment's form and falsifiers, nothing else; corpus, bake state, withheld mapping lists and receipts stay in the consumer repo's own ledger. Stated as a tripwire for future edits: any ingestion or crosswalk-list text appearing here means the cut failed.
Bugbot couldn't run - usage limit reachedBugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit. A user or team admin can review and increase usage limits in the Cursor dashboard. (requestId: serverGenReqId_16ffb86f-0b8b-4d61-b7f7-080dbb8278c9) |
|
Important
This repository does not receive automatic reviews because it has fewer than 10 stars. ⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Plus Run ID: Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
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
2 participants
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.
Summary
Docs/board only — no source change, no mint, no layout bump. Two artifacts:
1.
EPIPHANIES.md:E-R2IL-MACRO-VOCABULARY-TRANSFERS-ACROSS-COMPILER-AND-LANGUAGE-1The 33-macro BPE vocabulary from
E-BPE-OVER-DEFUSE-CHAINS-BEATS-LINEAR-AND-FITS-LOCO-1, trained on the two gccstress_testbinaries only, measured against two unseen corpora with two pre-registered shuffle nulls (20 seeds each, multiset/marginal preservation asserted on every draw):vuln_test)serde_jsonbuild script, rustc/LLVM)The vocabulary is not a gcc idiom; the pre-registered kill condition (REAL inside the column-null range ⇒ split the palette per toolchain) did not fire. The entry carries its own deflations: 33/33 macro hits on Rust is partly sample size (density is the honest statistic), the Rust harvest is capped at 200/548 functions, and coverage saturates. It also records that my own pre-registration missed in the hypothesis's favour. Measured by instrumenting the shipped
probe_bpe_r2il_loco_microcode.rsand reverting — its 10 gates stayed green throughout, and its B7 fence correctly failed the moment a third binary entered the corpus.2.
.claude/plans/r2il-machine-semantic-contract-v1.md(new, 445 lines)Answers a sibling session's live question — how should a session store R2IL — once, from shipped code and measurement:
file:line-cited), plan (every item with a falsifier), status (D-ids), evidence (board).r2sleighuses R2IL as its machine-semantic contract, R2IL is backed by SoA V4; Ghidra keeps its Java skin over a lazy compatibility facade, not a compatibility data model.VarnodeFacetpacksSpaceId::Custom(n)— a per-binary ordinal — into classid, against the ratifiedCONTENT NEVER TRAVELS IN CLASSIDlaw (fixed spaces are a legitimate reading;Custom(n)is content). CONJECTURE, falsifier in §6/W0.le-contract.md§3 and read by reinterpret; a shared palette, not FnIndex-per-macro; behaviour by address).Custom(n)content-vs-reading census) gates the0xC4mint; libsla stays as the byte-parity oracle and is removable last.Also carried on the branch: the V4-ruling storno on the name-collision objection (
e76b88a, already on main via the rebase) — this PR's four new commits areffef98f..35f10ae.Test plan
git statusclean, verified.🤖 Generated with Claude Code
https://claude.ai/code/session_01AGVLyRZNEKKBSfBDJfbY3V
Generated by Claude Code