Skip to content

Mutation triage: v1.2.0 candidate

This is an adjudication of recorded experiments, not a claim that the original program contains every injected defect. A surviving mutation proves that its executed tests did not distinguish that changed program; a timeout proves neither survival nor semantic correctness. No production source was edited for this review.

Frozen scope and accounting

The adjudicated sample records 135 completed mutants: 80 caught, 43 missed, 7 unviable and 5 timeouts, selected by round-robin shard 1/40 across eight crates. Initial adapter/proxy baselines failed before any mutant ran; their completed retry snapshots supersede those two initial rows only. The original source manifest and retry source manifest identify the frozen sources. The original enumerated inventory was 5,226 possible mutations; the full space was not executed. The latest per-crate inventories in the adjudicated mixed-snapshot rows sum to 5,241; this is inventory drift between snapshots, not additional executed coverage. These samples do not establish a population mutation score.

The later directed critical run, with its separate source manifest, records 59 experiments: 30 caught and 29 missed, with no recorded timeout/unviable outcome. Its per-crate selected counts are core 30, HDC 13, bus 10 and codec 6. Twelve full recorded names also occur in the original sample; three directed misses also occur among the original misses. Line movement creates additional semantic overlap (for example frame-fit line 456 → 462). Do not add 135+59, caught totals, or inventories to report unique tested mutations. No deduplicated cross-snapshot coverage total is asserted here.

Stable recorded names below include path, frozen location, function and mutation. Guard snippets/follow-ups identify semantic intent when lines move. Source links point to the current function location; manifests/outcomes, rather than current line numbers, establish the experiment's provenance. Subsequent source/tests may differ. A later ordinary test addition is never retroactively counted as CaughtMutant.

Classification: G = meaningful test/oracle gap for an observable contract or safety boundary; P = source-proven returned-value equivalence with a possible cost difference; N = no defect/equivalence established under the documented heuristic/fixture contract; T = recorded timeout requiring bounded direct follow-up. G is not an allegation of a defect in unmutated production source. Two P classifications below are narrowly scoped; no blanket equivalence waiver is given to other survivors.

All 43 sampled survivors and five timeouts

ID Frozen outcome / full semantic name Class Source proof or counterevidence Follow-up / later evidence
S01 MissedMutant: crates/icx-core/src/cas.rs:77:9: replace CasStore::unique_chunk_count -> usize with 0 (log) G Public deduplication count can return zero after insertion. source Current integrity_v120 asserts one unique chunk before/after duplicate insertion; later test, not a rerun outcome.
S02 MissedMutant: crates/icx-core/src/cas.rs:565:21: replace \|\| with && in CasStore::from_parts (log) G Line-index byte_offset guard can be weakened independently of chunk identity/length. source Current integrity_v120 independently corrupts all five line-index fields; frozen directed run already catches some byte guards, not every later test.
S03 MissedMutant: crates/icx-core/src/chunker.rs:111:23: replace < with > in chunk_bytes (log) G Disables the relaxed-mask phase when pos is below limit; reconstruction can still pass with different boundaries. source Add a deterministic nontrivial input that reaches phase 2 and assert independently calculated CDC boundaries, not only round-trip/min/max.
S04 MissedMutant: crates/icx-core/src/dag.rs:324:22: replace match guard *n_agent == agent && n_role == role && *n_raw == raw_artifact with false in MerkleContextDag::find_turn (log) G Deletes the uncached serde-restored find_turn fallback match. source Directed run catches this semantic mutant; current integrity_v120 exercises fallback identity. Preserve sampled Missed outcome.
S05 MissedMutant: crates/icx-core/src/dag.rs:976:21: replace \|\| with && in MerkleContextDag::verify_merkle_integrity_with_ancestors (log) G Turn verification combines digest/prefix/map-ID checks; one false conjunct can hide another mismatch. source Current integrity_v120 separately tampers data and jointly forges map/internal IDs; remutate each independent guard to confirm.
S06 MissedMutant: crates/icx-core/src/dag.rs:1112:25: replace \|\| with && in MerkleContextDag::verify_merkle_integrity_with_ancestors (log) G L3 epoch verification combines root/prefix/map-ID/frame checks. source Current integrity_v120 tampers epoch root and map/internal IDs; remutate each condition without relying on a simultaneous frame mismatch.
S07 MissedMutant: crates/icx-core/src/epoch.rs:456:21: replace <= with > in CacheEpochController::assemble_window (log) G Whole-frame fit selection is reversed; earlier fixtures did not observe missing/oversized frames. source Directed run catches the shifted frame-fit mutant (line 462); do not identify it by line number alone or revise frozen result.
S08 MissedMutant: crates/icx-codec/src/ahtl.rs:207:28: replace && with \|\| in escape_clause_field (log) P Only broadens needs-escaping detection in the fast-return predicate; the unchanged escaping loop still decides what to escape. source Source proof: A&&B implies A||B, so fast return is taken for a subset of previously safe strings. Output is unchanged; extra allocation/scan is observable only as cost.
S09 MissedMutant: crates/icx-codec/src/ahtl.rs:734:31: replace \|\| with && in AhtlWirePacketV1::parse (log) G Hex length and alphabet are separate malformed-input cases; && can accept one bad property. source Directed parser run catches this exact mutant; current codec_v120 independently tests alphabet and length.
S10 MissedMutant: crates/icx-codec/src/redact.rs:157:19: replace + with * in SecretRedactor::sensitive_continuation (log) G Changes the UTF-8-aware split after the last delimiter to multiplication. source Exercise indented/YAML continuation keys with multibyte prefixes; assert secret removal, harmless-prefix preservation and no panic. Current redactor evolved after snapshot.
S11 MissedMutant: crates/icx-codec/src/redact.rs:372:40: replace + with * in SecretRedactor::redact_line (log) G Changes cursor advance after a vendor-prefix replacement; nonzero offsets can skip later candidates. source Add two separated vendor tokens with nonzero first offset, then assert both redacted and operation terminates; map onto current scanner before rerun.
S12 MissedMutant: crates/icx-codec/src/redact.rs:416:45: replace \|\| with && in SecretRedactor::redact_value_span (log) G Quoted-value scanning can stop at the wrong boundary when quote alternatives become conjunctive. source Add both quote styles with trailing public fields, escaped quotes and unterminated values; verify redaction plus preserved boundaries.
S13 Timeout: crates/icx-codec/src/redact.rs:563:40: replace + with * in SecretRedactor::redact_colon_assignments (log) T Skipped colon syntax no longer advances when colon_idx is zero: zero*1 stays zero. Log records tests running >60s then Timeout. source For a frozen input beginning with a skipped colon, source shows a no-progress loop in the mutant. Keep Timeout; add bounded direct regression if rerunning, not Caught.
S14 MissedMutant: crates/icx-codec/src/redact.rs:620:51: replace - with + in SecretRedactor::redact_cli_flags (log) G CLI flag boundary checks the following byte instead of preceding byte. source Test alphanumeric-prefixed --secret false positives and valid standalone flags; current redactor must be targeted separately.
S15 MissedMutant: crates/icx-codec/src/redact.rs:803:78: replace == with != in SecretRedactor::redact_jwt_tokens (log) G JWT token alphabet changes hyphen equality to inequality, widening scans across delimiters and excluding hyphens. source Test base64url segments containing hyphens plus punctuation/quotes after complete JWTs; assert exact redacted span.
S16 MissedMutant: crates/icx-codec/src/redact.rs:937:13: replace \|\| with && in SecretRedactor::is_high_entropy_secret_candidate (log) G Entropy exemption for dot-containing tokens can be weakened by a neighboring false predicate. source Test dotted code/path literals and eligible entropy strings independently; broad fuzz success alone does not adjudicate this mutant.
S17 MissedMutant: crates/icx-codec/src/skeleton.rs:228:45: replace < with <= in StructuralSkeletonizer::skeletonize_diff (log) G Diff skeleton file limit changes from four to five. source Add a five-distinct-file diff and assert the documented compact file-clause cap; lossy skeletons need a shape oracle, not exact-source recovery.
S18 MissedMutant: crates/icx-hdc/src/lib.rs:406:76: replace && with \|\| in HdcAssociativeIndex::compute_subject_bloom32 (log) N Changes subject tokenization and therefore approximate subject bloom ranking. source No universal heuristic-recall oracle establishes a production defect. Add path/colon/underscore token fixtures and benchmark quality/cost before deciding whether exact tokenization is a contract.
S19 MissedMutant: crates/icx-hdc/src/lib.rs:638:13: replace \|\| with && in HdcAssociativeIndex::query_associative (log) N Changes full-scan eligibility versus coarse candidate selection, especially legacy/no-postings paths. source Candidate membership may change, not just cost. Exact lexical union remains separate; compare seeded old 03/current 04 corpora and tie order. No general equivalence or bug is established.
S20 MissedMutant: crates/icx-hdc/src/fractal.rs:203:25: replace * with / in FractalCoord::morton_z64 (log) G Positive coordinate binning divides by eight instead of multiplying. source Directed run catches this exact mutant using literal Morton vectors; later negative-half vector addresses a different directed survivor.
S21 MissedMutant: crates/icx-hdc/src/fractal.rs:248:24: replace ^ with & in FractalCoord::digest_u32 (log) G FNV byte mixing becomes bitwise AND. source Directed run catches this exact mutant; numeric_vectors_v120 includes literal expected digest.
S22 MissedMutant: crates/icx-hdc/src/fractal.rs:388:23: replace \|\| with && in FractalTrajectoryArchive::query_shared_suffix (log) P Removes an early return only when one of top_k==0 or empty collection is true. source Source proof: empty iterator stays empty; otherwise truncate(0) returns empty. Nonempty/top_k>0 takes the same path. Returned values equivalent; zero-k work/allocation can increase.
S23 MissedMutant: crates/icx-hdc/src/hv.rs:155:25: replace >> with << in Hypervector::from_u64_seed (log) G Changes the SplitMix recurrence shift direction. source Directed run catches exact mutant with independent literal seed words; retain original Missed.
S24 MissedMutant: crates/icx-hdc/src/hv.rs:513:24: replace ^= with &= in Hypervector::sketch_256 (log) G A zero-initialized sketch changes XOR accumulation to AND, collapsing sparse information. source Directed run catches exact mutant with independent sparse rotating-sketch vector.
S25 MissedMutant: crates/icx-hdc/src/surprise.rs:221:18: replace - with / in TitansSurpriseGate::observe_with_exit_code (log) G Novelty 1-similarity becomes 1/similarity and then clamps; repeated-fingerprint tests can bypass this branch. source Directed run still missed; current numeric_vectors_v120 adds a near vector with a new fingerprint and hand-calculated novelty. Later addition is not Caught evidence.
S26 MissedMutant: crates/icx-bus/src/ring.rs:280:30: replace >> with << in unpack_meta (log) G Surprise metadata unpacking shifts left, losing the encoded byte. source Directed bus run catches exact mutant and all selected metadata shifts.
S27 MissedMutant: crates/icx-bus/src/ring.rs:945:49: replace * with + in SeqlockTelemetryBus::poll_since (log) G File-backed sketch channel slice offset uses addition instead of eight-byte stride. source Directed bus run catches exact mutant; file-backed patterned channels are required, not only in-process publish tests.
S28 MissedMutant: crates/icx-adapters/src/antigravity.rs:424:41: delete ! in AntigravityAdapter::handle_hook (log) G Subagent summary filter accepts empty summaries and rejects nonempty summaries. source Test missing/empty/nonempty Antigravity completion summaries and resulting fold content.
S29 MissedMutant: crates/icx-adapters/src/hermes.rs:98:42: replace <= with > in HermesAdapter::sanitize_embedded_ahtl_line (log) G Embedded AHTL takes fallback sanitization rather than delimiter-aware segment handling. source Test valid embedded packets with secrets in individual fields and oversized input; require packet/framing and redaction invariants, not merely no secret output.
S30 MissedMutant: crates/icx-adapters/src/mmu.rs:200:25: replace -= with /= in InfinityContextMmu::extract_hook_subject (log) G UTF-8 subject prefix retreat divides rather than decrements; can shorten subject to byte zero. source Use a command crossing byte 48 inside a multibyte codepoint; assert maximal safe prefix and stable hash, not just valid UTF-8.
S31 MissedMutant: crates/icx-adapters/src/mmu.rs:314:40: replace != with == in InfinityContextMmu::truncate_agent_suffix (log) G Suffix truncation retains the wrong agent occurrences. source Interleave two agents then rewind one; assert only that agent suffix is removed and replay identity is retained.
S32 MissedMutant: crates/icx-adapters/src/mmu.rs:377:25: replace == with != in InfinityContextMmu::ingest_or_reuse_at_index (log) G Exit-code replay identity uses inequality, allowing reuse/re-ingest decisions to invert. source Replay same occurrence with changed exit code and observe new verdict/state; separately assert unchanged event reuse.
S33 MissedMutant: crates/icx-adapters/src/mmu.rs:897:47: delete ! in InfinityContextMmu::build_ephemeral_tail_slot_inner (log) G Deletes negation in punctuation trim; alphanumeric content can be trimmed away. source Independent punctuated-query mutated-source regression exits 101 as expected (tail-ppr-mutant.json). This later focused rejection does not change sampled Missed.
S34 MissedMutant: crates/icx-adapters/src/workspace.rs:172:5: replace require_absolute -> Result<(), TrustError> with Ok(()) (log) G Private workspace absolute-path guard is removed. source Exercise public workspace setup with relative HOME/XDG roots in an isolated environment; assert failure before creating state. Tests using only absolute temp paths cannot reject this mutant.
S35 MissedMutant: crates/icx-proxy/src/wire.rs:1039:24: delete ! in WireProxyTransformer::transform_sse_line_stateful (log) G SSE template capture can select the wrong pending stream state. source Interleave pending tool streams with delayed template arrivals; assert both identifiers and emitted buffered data. Single-stream round-trip is insufficient.
S36 Timeout: crates/icx-proxy/src/wire.rs:1216:9: replace WireProxyTransformer::strip_provider_route_prefix -> String with String::new() (log) T Route rewriting becomes an empty path; log shows the proxy monolithic test >60s then Timeout. source Malformed/empty upstream route can strand an integration exchange. Add a direct route-prefix input/output test plus bounded mock transport; do not infer exact hang cause solely from timeout.
S37 Timeout: crates/icx-proxy/src/wire.rs:1293:38: replace == with != in WireProxyTransformer::read_next_http_chunk (log) T Trailer blank-line terminator equality becomes inequality; log shows proxy test timeout. source Exercise empty and nonempty trailers and EOF with a bounded reader; outcome remains unresolved timeout, not semantic-equivalence evidence.
S38 MissedMutant: crates/icx-proxy/src/wire.rs:1500:57: replace \|\| with && in WireProxyTransformer::relay_upstream_response_with_eof (log) G HEAD/204/304 bodyless-response alternatives become conjunctive around one status check. source Assert HEAD, 204 and 304 independently against a mock upstream that supplies no body and clean EOF; protect status/header handling.
S39 MissedMutant: crates/icx-proxy/src/wire.rs:2222:21: replace && with \|\| in WireProxyTransformer::handle_http_request_concurrent (log) G Transfer-Encoding classification accepts unrelated headers or non-chunked values. source Test unrelated header containing chunked, TE identity, valid mixed-case TE chunked, and framing conflicts; assert request consumption/rejection.
S40 MissedMutant: crates/icx-proxy/src/wire.rs:2744:26: replace += with *= in WireProxyTransformer::virtualize_item_text (log) G Proxy occurrence index multiplies rather than increments; repeated item identities can collapse. source Transform multiple tool items and replay/rewind; assert distinct ordered occurrence identities, CAS retention and stable output.
S41 MissedMutant: crates/icx-cli/src/main.rs:369:36: replace < with <= in run_daemon (log) G A terminal --idle-exit flag can pass bounds guard and index args[len]. source Add missing-value CLI case and require normal diagnostic/nonzero exit rather than panic. Valid-value happy path does not exercise boundary.
S42 Timeout: crates/icx-cli/src/main.rs:474:15: replace += with *= in parse_dir_flag (log) T At first --dir index zero, i*=2 does not advance; CLI integration log ends in Timeout. source Source establishes no-progress on leading --dir in the mutant; add subprocess deadline regression. Preserve Timeout instead of recording Caught.
S43 MissedMutant: crates/icx-cli/src/main.rs:615:28: replace + with * in execute_hook (log) G --event guard uses i*1 rather than i+1; a final flag can index beyond args. source Test terminal/missing --event and normal --agent/--event permutations; require explicit error with no panic.
S44 MissedMutant: crates/icx-cli/src/main.rs:834:33: replace match guard i + 1 < args.len() with true in run_proxy_server (log) G --max-requests always passes value-availability guard. source Test trailing --max-requests with no value using bounded subprocess and assert diagnostic, not panic.
S45 Timeout: crates/icx-cli/src/main.rs:832:19: replace += with -= in run_proxy_server (log) T Proxy-server --dir cursor decreases instead of increases; source can underflow at first index. Log records broader CLI test Timeout. source Direct option-parser subprocess should reject/halt predictably. Arithmetic suggests panic for some positions, but recorded timeout cause is not established; no invented Caught result.
S46 MissedMutant: crates/icx-cli/src/main.rs:913:35: replace + with - in execute_proxy_transform (log) G Proxy-transform --dir uses preceding argument instead of next, or underflows at zero. source Run --dir first and after provider selection; assert selected workspace and no out-of-bounds panic.
S47 MissedMutant: crates/icx-cli/src/main.rs:1395:26: replace * with / in varied_tool_log_fixture (log) N Changes synthetic varied-log word schedule; this is fixture content, not an algorithm correctness oracle. source No runtime defect follows from a different deterministic fixture. Preserve fixture provenance/digest and requalify reported measurements if changed; do not call it equivalent benchmark input.
S48 MissedMutant: crates/icx-cli/src/main.rs:1560:48: replace * with / in run_benchmark (log) G PPR elapsed seconds divided instead of multiplied by 1e6; reported microseconds/SLO checks can be corrupted. source Assert benchmark report units with independently injected elapsed duration or parser/oracle; shape-only report tests are insufficient.

The table preserves the recorded 43 MissedMutant and five Timeout outcomes. Source reasoning establishes two narrow returned-value-equivalent fast paths (S08/S22); three heuristic/fixture rows remain N (S18/S19/S47). The remaining survivors have concrete oracle follow-ups. This classification does not relabel historical outcomes as passed.

Directed critical survivors

All 29 directed misses are listed here, including the three already listed above; overlap labels prevent treating them as new independent gaps. The directed outcomes also retain all 30 caught names. In particular, later directed experiments reject the sampled turn-cache fallback, positive Morton binning, digest/sketch/seed arithmetic, both selected bus families, Profile A hex parser guards, and the relocated frame-fit selection guard. Those are distinct recorded observations, not changes to the earlier sample.

ID Full recorded semantic name Class Source guard / evidence and follow-up
D01 crates/icx-core/src/cas.rs:529:17: replace \|\| with && in CasStore::from_parts G Frozen guard token: \|\| verified_chunks.contains_key(&cid). Chunk digest/ID/duplicate guards: forge each independently, including valid digest with wrong short ID, wrong full digest, and duplicate entries. source
D02 crates/icx-core/src/cas.rs:528:17: replace \|\| with && in CasStore::from_parts G Frozen guard token: \|\| computed_digest != cdigest. Chunk digest/ID/duplicate guards: forge each independently, including valid digest with wrong short ID, wrong full digest, and duplicate entries. source
D03 crates/icx-core/src/cas.rs:541:17: replace \|\| with && in CasStore::from_parts G Frozen guard token: \|\| store.artifacts.contains_key(&id). Artifact digest-prefix/line-count/duplicate guards: test each malformed field and duplicate artifact independently. source
D04 crates/icx-core/src/cas.rs:540:17: replace \|\| with && in CasStore::from_parts G Frozen guard token: \|\| art.chunk_ids.len() != art.line_index.len(). Artifact digest-prefix/line-count/duplicate guards: test each malformed field and duplicate artifact independently. source
D05 crates/icx-core/src/cas.rs:552:50: replace \|\| with && in CasStore::from_parts G Frozen guard token: if computed_bytes != art.total_bytes \|\| computed_bytes > ICX_MAX_INGEST_BYTES {. Test wrong declared total bytes separately from over-limit computed size; max-size allocation test needs an explicit resource bound. source
D06 crates/icx-core/src/cas.rs:568:21: replace \|\| with && in CasStore::from_parts G Frozen guard token: \|\| entry.end_line != end_line. Single start_line/end_line corruption; current integrity_v120 now includes both. No post-addition directed rerun is recorded here. source
D07 crates/icx-core/src/cas.rs:567:21: replace \|\| with && in CasStore::from_parts G Frozen guard token: \|\| entry.start_line != start_line. Single start_line/end_line corruption; current integrity_v120 now includes both. No post-addition directed rerun is recorded here. source
D08 crates/icx-core/src/cas.rs:583:17: replace \|\| with && in CasStore::from_parts G Frozen guard token: \|\| art.total_lines != expected_total_lines. Test wrong total_lines independently from wrong full digest; current five line-entry tests do not by themselves cover total_lines. source
D09 crates/icx-core/src/dag.rs:976:21: replace \|\| with && in MerkleContextDag::verify_merkle_integrity_with_ancestors G Frozen guard token: \|\| *node_id != id. Turn digest-prefix/internal-ID/map-key checks: current integrity_v120 adds isolated data and jointly forged map/internal IDs; rerun each guard. Already S05 above; same recorded name. source
D10 crates/icx-core/src/dag.rs:975:21: replace \|\| with && in MerkleContextDag::verify_merkle_integrity_with_ancestors G Frozen guard token: \|\| ArtifactId::from_digest(&recomputed) != *node_id. Turn digest-prefix/internal-ID/map-key checks: current integrity_v120 adds isolated data and jointly forged map/internal IDs; rerun each guard. source
D11 crates/icx-core/src/dag.rs:987:21: replace \|\| with && in MerkleContextDag::verify_merkle_integrity_with_ancestors G Frozen guard token: \|\| tool.id != id. ToolArtifact digest, internal ID and map-key checks need independent serde-restored corruptions; turn/epoch-only tests do not prove coverage. source
D12 crates/icx-core/src/dag.rs:986:21: replace \|\| with && in MerkleContextDag::verify_merkle_integrity_with_ancestors G Frozen guard token: \|\| ArtifactId::from_digest(&recomputed) != tool.id. ToolArtifact digest, internal ID and map-key checks need independent serde-restored corruptions; turn/epoch-only tests do not prove coverage. source
D13 crates/icx-core/src/dag.rs:1006:21: replace \|\| with && in MerkleContextDag::verify_merkle_integrity_with_ancestors G Frozen guard token: \|\| folded_ahtl_commit.len() < expected_prefix.len() + expected_suffix.len(). SubTrajectoryBranch prefix/suffix/minimum-length checks: independent malformed frame cases; correlated framing failures must not mask each other. source
D14 crates/icx-core/src/dag.rs:1005:21: replace \|\| with && in MerkleContextDag::verify_merkle_integrity_with_ancestors G Frozen guard token: \|\| !folded_ahtl_commit.ends_with(&expected_suffix). SubTrajectoryBranch prefix/suffix/minimum-length checks: independent malformed frame cases; correlated framing failures must not mask each other. source
D15 crates/icx-core/src/dag.rs:1036:21: replace \|\| with && in MerkleContextDag::verify_merkle_integrity_with_ancestors G Frozen guard token: \|\| *raw_tokens_saved != expected_saved. Branch digest-prefix/map key/raw_tokens_saved checks: corrupt each separately while preserving child integrity and canonical fold framing. source
D16 crates/icx-core/src/dag.rs:1035:21: replace \|\| with && in MerkleContextDag::verify_merkle_integrity_with_ancestors G Frozen guard token: \|\| *node_id != id. Branch digest-prefix/map key/raw_tokens_saved checks: corrupt each separately while preserving child integrity and canonical fold framing. source
D17 crates/icx-core/src/dag.rs:1034:21: replace \|\| with && in MerkleContextDag::verify_merkle_integrity_with_ancestors G Frozen guard token: \|\| ArtifactId::from_digest(&recomputed) != *node_id. Branch digest-prefix/map key/raw_tokens_saved checks: corrupt each separately while preserving child integrity and canonical fold framing. source
D18 crates/icx-core/src/dag.rs:1075:29: replace \|\| with && in MerkleContextDag::verify_merkle_integrity_with_ancestors G Frozen guard token: \|\| !frame_matches(. Folded L4 epoch index/root/ID/map-key/raw-token/frame checks: independently corrupt each serde-restored field and verify rejection. source
D19 crates/icx-core/src/dag.rs:1074:29: replace \|\| with && in MerkleContextDag::verify_merkle_integrity_with_ancestors G Frozen guard token: \|\| exp_raw != *raw_tokens. Folded L4 epoch index/root/ID/map-key/raw-token/frame checks: independently corrupt each serde-restored field and verify rejection. source
D20 crates/icx-core/src/dag.rs:1073:29: replace \|\| with && in MerkleContextDag::verify_merkle_integrity_with_ancestors G Frozen guard token: \|\| *node_id != id. Folded L4 epoch index/root/ID/map-key/raw-token/frame checks: independently corrupt each serde-restored field and verify rejection. source
D21 crates/icx-core/src/dag.rs:1072:29: replace \|\| with && in MerkleContextDag::verify_merkle_integrity_with_ancestors G Frozen guard token: \|\| exp_id != *node_id. Folded L4 epoch index/root/ID/map-key/raw-token/frame checks: independently corrupt each serde-restored field and verify rejection. source
D22 crates/icx-core/src/dag.rs:1071:29: replace \|\| with && in MerkleContextDag::verify_merkle_integrity_with_ancestors G Frozen guard token: \|\| exp_root != *merkle_root. Folded L4 epoch index/root/ID/map-key/raw-token/frame checks: independently corrupt each serde-restored field and verify rejection. source
D23 crates/icx-core/src/dag.rs:1112:25: replace \|\| with && in MerkleContextDag::verify_merkle_integrity_with_ancestors G Frozen guard token: \|\| *node_id != id. L3 epoch digest-prefix/map-key/frame checks; current map/internal-ID and root tests added after the snapshot. Rerun independently. Already S06 above; same recorded name. source
D24 crates/icx-core/src/epoch.rs:444:51: replace <= with > in CacheEpochController::assemble_window G Frozen guard token: let pinned_head = if diagnostic.len() <= cap {. Oversized-head fallback diagnostic-fit guard, not whole-frame fit. Test cap below diagnostic length and cap exactly diagnostic length; assert managed view stays within cap. source
D25 crates/icx-hdc/src/fractal.rs:205:29: replace * with / in FractalCoord::morton_z64 G Frozen guard token: (((c + 1.0) * 8.0).floor() as u64).min(7). Negative coordinate binning at -0.5 differs between multiply/divide; current numeric_vectors_v120 adds literal 0x0000ffff00000000. Added after this frozen run, not counted caught. source
D26 crates/icx-hdc/src/surprise.rs:221:18: replace - with / in TitansSurpriseGate::observe_with_exit_code G Frozen guard token: (1.0 - self.bundler.finalize().similarity(hv)).clamp(0.0, 1.0). Near-vector/new-fingerprint novelty arithmetic remains missed in directed snapshot; current hand-calculated third observation was added later, not counted caught. Already S25 above; same recorded name. source
D27 crates/icx-codec/src/ahtl.rs:623:33: replace \|\| with && in AhtlPacket::to_profile_a G Frozen guard token: if fc_clean.len() != 64 \|\| !fc_clean.bytes().all(\|b\| b.is_ascii_hexdigit()) {. Promotion length and alphabet are independent malformed fc/hv cases. Current codec_v120 promotion test adds both for each field after this snapshot; not a recorded rejection yet. source
D28 crates/icx-codec/src/ahtl.rs:630:33: replace \|\| with && in AhtlPacket::to_profile_a G Frozen guard token: if hv_clean.len() != 16 \|\| !hv_clean.bytes().all(\|b\| b.is_ascii_hexdigit()) {. Promotion length and alphabet are independent malformed fc/hv cases. Current codec_v120 promotion test adds both for each field after this snapshot; not a recorded rejection yet. source
D29 crates/icx-codec/src/ahtl.rs:861:46: replace \|\| with && in EpochHeader::parse G Frozen guard token: if !(4..=8).contains(&ep_part.len()) \|\| !ep_part.bytes().all(\|b\| b.is_ascii_hexdigit()) {. Epoch-header ep width 4..8 and hex alphabet guard. from_str_radix separately rejects invalid hex, but valid 1–3-digit epochs can bypass width under this mutation. Add short valid-hex and 9-digit cases; not equivalent. source

Current follow-up evidence and remaining limits

Core integrity tests now cover all five CAS line-entry fields, deduplicated chunk count, fallback turn lookup, digest tampering and self-consistent forged Turn/CacheEpoch map/internal IDs that disagree with the computed digest prefix. They do not establish every ToolArtifact, branch, folded-epoch, duplicate-entry, total-lines or maximum-size guard mutation is caught. Correlated corruption can trigger an unchanged neighboring guard, so independent field cases matter.

HDC numeric vectors now include negative Morton half-space and near/new-fingerprint novelty cases; codec tests independently test promotion length/alphabet. These latest additions occurred after the directed source snapshot. This document records their existence and intended oracle, not an unrecorded successful mutation rerun. The isolated PPR trim experiment does record a focused later test rejection (expected exit 101); it overlaps S33 and adds no new unique sample count.

The redactor source changed further after these frozen runs while fuzz findings were repaired and original regressions replayed. Frozen redactor locations must be remapped onto the current scanner before further mutation experiments. Passing those later original-code regressions is counterevidence against specific reproduced fuzz failures, not proof that the old redactor survivors are now caught or equivalent. See retained fuzz outcomes and the release qualification record for the latest completed campaign; unfinished/interrupted campaigns remain distinct.

No concrete defect in the unmutated current source was established by this read-only review. No material branch was proven unreachable in the original program; the two P rows describe redundant early-return work, not unreachable public functionality. The observable mutation gaps above remain follow-up work, especially independent integrity checks, noncanonical header widths, CLI missing-value boundaries, and HTTP framing/SSE cases. The original 5,226-entry full mutation space remains unexecuted; later inventory changes do not establish exhaustive coverage. Closing an audit/release ticket must preserve these limits rather than assert exhaustive mutation resistance or convert test additions into historical caught counts.

Later executed evidence

The following separate frozen rechecks supersede statements above that these particular follow-up oracles had not yet been mutation-tested. They preserve every original table and outcome:

  • r2: HDC 13/13 caught, including negative Morton and near-vector novelty; codec 6/6 caught, including promotion length/alphabet and EpochHeader widths. Core recorded 11 caught / 19 missed from 30 selected.
  • r3: later independent CAS metadata tests changed the core recheck to 16 caught / 14 missed from the same 30 selected semantic families.
  • r4: independent chunk ID/full-digest and ToolArtifact/L2/L4 map-ID, content, framing and token-field corruption oracles produced 29 caught / 1 missed from 30 selected. See its adjudication. The remaining miss was the oversized-head diagnostic-fit comparison.
  • r5: a public-constructor test requires an actionable ICX_CONTEXT_BUDGET_EXCEEDED diagnostic for an oversized pinned head. A bounded two-case assemble_window recheck caught both the diagnostic-fit mutation and the frame-allowance mutation. This is later proof for the r4 survivor, not a revision of r4's outcome.

The original directed bus run remains 10/10 caught in the 59-case record. These follow-ups overlap each other and earlier samples; no additional aggregate unique coverage count is asserted. Each linked run retains its own source manifest, full per-crate inventory, selector and outcomes. The directly rechecked critical families are now caught by the described oracles. The broad sample's 43 historical misses remain recorded misses; unrelated oracle gaps, heuristic cases, timeouts and unexecuted mutation space remain subject to their original limits. No exhaustive mutation-resistance claim follows from these bounded rechecks.