RXRE engine v1.9.0 — basename bounded-property milestone

RXRE engine v1.9.0 — basename bounded-property reconstruction (2026-11-13).

New sixth bounded target: GNU coreutils 9.10 /usr/bin/basename (sha256 1ef9913961b3006d8d5f41718035a8ef7aedcf5f81da46100dffa61421989621; BuildID eb755c71c52c5282f84d01ae80ba24e6f0a5e060). Scoped to a single path operand with NO flags (-a, -s, -z) and NO suffix operand: property = last component after trailing-slash strip (all-slash input maps to / when leading-slashed, otherwise .). . and .. are treated as literal components.

Corpus 1,987/2,013 cases (seeds 20261114/15; UTF-8 wide chars + boundary buckets) sealed before reconstruction; independent C11 reconstruction (basename_20261114/basewrap.c); 0 divergences on both generations; fail-closed identity pin; manifest 65 artifacts; verify.sh 16 stages; pytest gate 80 passed, 2 skipped.

test_discovery_on_basename.py added as integration coverage. Discovery still does NOT validate semantics; P0 remains OPEN. No general-basename claim. Third-party verification remains NOT_EXECUTED.

Epistemic status: PARTIAL. MIT.

ZIP SHA-256: see rxre-engine-1.9.0.zip.sha256 (fd7d56cdf689e87bdba0201f78bf0ac9c418b3f5ba90f95a762a53436edad9c9).

RXRE engine v1.8.0 — MENTE synthesis end-to-end loop

RXRE engine v1.8.0 — MENTE synthesis end-to-end loop (2026-11-13).

First complete loop on a previously-unknown target: test_discovery_synthesis.py — binary compiled at test time in tmp; discover_unknown_function runs without any access to the source; rodata literals alpha,beta recovered; sentinels extracted by invoking binary with the visible literals; C regenerated from those literals; 0 mismatches byte differential against the original on a deterministic trial corpus.

Honest limitations: this target was deliberately trivial (-O0 -fno-inline, strong literal hints; both literals visible). P0 remains OPEN for harder targets; no general claim.

verify gate: 78 passed, 2 skipped; 15 stages; manifest 60 artifacts; VERIFY: all steps passed.

Epistemic status: PARTIAL. Third-party verification remains NOT_EXECUTED. MIT.

ZIP SHA-256: see rxre-engine-1.8.0.zip.sha256.

RXRE engine v1.7.0 — MENTE multi-round discriminating cycle

RXRE engine v1.7.0 — MENTE multi-round (2026-11-13).

What changed:

• mente_cycle() runs an iterative discriminating loop, not just one shot. Each round picks the input maximizing rival disagreement; refutes the contradicted hypotheses; stops when a single survivor remains or when no disagreeing input remains. The lab-hard test_discovery_loopbench.py now shows this does refute EXACT,LEN5,NULL and keeps only the ground-truth PREFIX in a synthetic binary compiled fresh by the test.
• Honest unintended limitation remains: this is a lab bench, not end-to-end synthesis. Reports it doesn't claim PASS for general minds.
• verify gate: 77 passed, 2 skipped (count reported by pytest itself); 15 stages; manifest 60 artifacts; VERIFY: all steps passed.

Epistemic status is PARTIAL. Third-party verification remains NOT_EXECUTED. MIT.

ZIP SHA-256: see rxre-engine-1.7.0.zip.sha256.

RXRE engine v1.6.1 — P0 on dirname (gate 76 passed/2 skipped)

RXRE engine v1.6.1 — P0 on a real binary (2026-11-13).

• test_discovery_on_dirname.py: discovery runs over /usr/bin/dirname end-to-end (rizin -A + izj-based literal filter from .rodata only). Produces a FUNCTION_CANDIDATE with cfg_blocks>0, rivals from section-scoped literals, _no_profile_imported == True, no semantic content asserted — semantics_status == NOT_DETERMINED.
• verify gate: 76 passed, 2 skipped (count reported by pytest itself).
• Manifiesto 60 artefactos; 15 stages.

Epistemic status: P0 partial. Discovery does NOT yet solve the security-critical question — rivals don't select discriminating inputs and no byte-exact reconstruction of dirname is produced. Third-party verification remains NOT_EXECUTED. MIT.

ZIP SHA-256: see rxre-engine-1.6.1.zip.sha256.

RXRE engine v1.6.0 — discovery module + critical latent bug fix

RXRE engine v1.6.0 — P0 (discovery) partial + critical latent bug fix (2026-11-13).

What changed under the hood:

• tools/rxre/semantic/discovery.py (new): discover_unknown_function(binary_path) — picks largest FUNCTION_CANDIDATE from the reconstructor, enumerates rivals from literals in rodata (prefix/exact/first-byte), marks _no_profile_imported = True. Lab tests: test_discovery_unknown.py — 2 tests, both PASS.
• Critical latent bug fixed in reconstructor.py: the CFG pump never produced any CFG (_json was never aliased before use, subprocess was missing at module scope, re similarly) — every analyzed binary decayed to ANALYSIS_FAILED. With it fixed, FUNCTION_CANDIDATE count balloons as expected.
• verify.sh: 15 stages, expected pins 74 passed / 2 skipped.

Epistemic status: P0 remains largely OPEN. Discovery produces rival hypotheses from the visible rodata, but semantic hypotheses have NOT been driven to a verified reconstruction of an unknown function. reconstruct() remains a structural scaffold; _discovery itself does NOT produce byte-exact reconstruction yet. Third-party verification NOT_EXECUTED. MIT.

ZIP SHA-256: see rxre-engine-1.6.0.zip.sha256.

RXRE engine v1.5.3 — coherence/content fix

RXRE engine v1.5.3 — coherence/content fix release (2026-11-13).

Content correct w.r.t. claims: this tag's zip contains the P1 strengthening code (test_5b/5c2/5d, test_replay_uses_original_request_payload, evaluate_contract semantic gate, replay executes with request_json payload, test_integridad_adv_final and test_replay_mision aligned) — the same code that was documented but not tagged in v1.5.0..v1.5.2. Doc counts now consistent with the executed gate: 72 passed, 2 skipped.

Verification content: 60-artifact manifest (60 SHA-256s bound), 15 stages, five bounded targets sealed, independent third-party verification remains NOT_EXECUTED. MIT.

Supersedes v1.5.0/v1.5.1/v1.5.2 (tag-rewritable convenience releases with stale README content pointers). All evidence is sealed by manifest of this tag; v1.5.3 README is the authoritative coherence anchor.

ZIP SHA-256: see rxre-engine-1.5.3.zip.sha256.

RXRE engine v1.5.2 — convenience supersede (transparency coherence)

RXRE engine v1.5.2 — convenience supersede (2026-11-13).

No verification content changed. The v1.5.1 tag/release carries the same sealed evidence as v1.5.2; v1.5.2 only fixes the README snapshot pointer (v1.5.0 -> v1.5.2) so the v1.5.1 release notes can state this coherently. All seals remain byte-exact; verify.sh unmodified.

Verification gate: 72 passed, 2 skipped. VERIFY: all steps passed.

Epistemic status unchanged: no general claim; third-party verification remains NOT_EXECUTED. MIT.

ZIP SHA-256: see rxre-engine-1.5.2.zip.sha256.

RXRE engine v1.5.0 — FIVE sealed bounded targets

RXRE engine v1.5.0 — FIVE bounded targets sealed (2026-11-13).

With filenvercmp, wc, fold and tr, the package adds a fifth bounded property verified the same way:

• dirname (/usr/bin/dirname — sha256 aac6a889dcfe1bcaabdcfea17b89c643c29abee5efce3fafbbb0c34b70e864a7, GNU coreutils 9.10-5.fc44): property reconstruction (parent directory computation) in C11 vs sealed corpus (seeds 20261113/14, 1.917/1.901 cases) — byte-exact, 0 divergences.
• verify.sh → 15 stages with /usr/bin/dirname fail-closed identity check; manifest extended to 60 artifacts; verify gate 69 passed, 2 skipped.

Epistemic status unchanged: no general claim; third-party verification remains NOT_EXECUTED; issue #1 stands. MIT.

ZIP SHA-256: see rxre-engine-1.5.0.zip.sha256.

RXRE engine v1.4.0 — FOUR sealed bounded targets

RXRE engine v1.4.0 — FOUR sealed bounded targets (2026-11-13).

After filenvercmp and wc, the fold milestone (cell width, multi-width) and now tr -s x are verified with the same discipline:

• tr milestone (evidence/r3/tr_20261113/): independent C11 reconstruction (trwrap.c): greedy squeeze de runs consecutivos de 0x78 ('x'); 2 x 2.513 = 2.513×2, 0 divergencias (seeds 20261113/14); corpus sellado sesgo a runs de x; verify.sh 14 etapas con /usr/bin/tr fail-closed pin (b1d1bfbe… coreutils-9.10-5.fc44).
• Four targets: /usr/bin/ls (filenvercmp), /usr/bin/wc (11 corpora), /usr/bin/fold (6 regeneraciones), /usr/bin/tr (2 generations).
• Manifest extended to 56 artifacts; verify gate 69 passed, 2 skipped.

Epistemic status: no general claim; third-party verification remains NOT_EXECUTED; issue #1 stands. MIT.

ZIP SHA-256: see rxre-engine-1.4.0.zip.sha256.

RXRE engine v1.3.10 — convenience supersede (README pointer fix)

RXRE engine v1.3.10 — convenience supersede (2026-11-13).

Content identical to v1.3.9, except README now correctly points to v1.3.10 (v1.3.9's README points to v1.3.8 — pointer lag; no code or evidence changed).

Verification gate: 69 passed, 2 skipped, 13 stages, VERIFY: all steps passed.

Epistemic status unchanged: no general claim; third-party verification remains NOT_EXECUTED. MIT.

ZIP SHA-256: see rxre-engine-1.3.10.zip.sha256.