P3-15′-c-ii: the criterion (c) synteny predicate, its per-candidate producer, and D4's two symmetric diagnostics

[bioedca]

ADR-0006 D4's criterion (c) over the 339 md5-verified NCBI GFFs P3-15′-c-i acquired: the frozen predicate, the per-candidate producer that writes candidate_id → passed|failed|unavailable, and — not optional, D4 asks for them symmetrically — the false-pass and false-FAIL/per-clade-exclusion/pseudogene diagnostics.

The criterion discriminates, and by a wide margin

arm	n	(c) pass rate
round-0 candidates	941	58.0 % (546 passed / 96 failed / 299 unavailable)
clade-matched random leaders	9,600	1.82 %
§9.1 5′UTR decoys (re-instantiated in-corpus)	9,097	4.74 %
§9.1 tRNA-adjacent decoys	1,860	2.69 %
positive-context control	4,778	99.35 %

Margin 0.975 ≥ 0.20 ⇒ the control is powered, so the rates are interpretable rather than merely quiet. The 58 % is not looseness: the round-0 pool is Stage-1 false positives on a masked negative substrate, and a 96-nt element scoring 0.9977 sitting 19 bp upstream of an alanine--tRNA ligase is most plausibly an uncatalogued T-box. ADR-0005 D14's spare rule exists so those are spared rather than trained against.

⚠ The biggest defect was found by disbelieving a number, not by a test

The very first run returned 541 of 941 passing while the per-CDS background density of D4's four classes is 4.14 % — irreconcilable. The downstream walk bounded distance only from above, so a window buried inside a long gene (a bacterial genome is ~88 % coding) found that gene and inherited its identity. D4's class-II case genuinely needs a negative distance, so the fix could not be "reject overlap": the CDS start must lie at or after the element's own 5′ end, which pins no new number. Corpus counts barely moved — the number that exposed the bug was not the number the bug corrupted.

⚠ Two silent zeros the diagnostic found in itself

1. The pseudogene diagnostic reported 0 corpus-wide, because the tandem carve-out hops exactly the population it exists to size. Counts now accumulate across the whole walk (15 pseudogenized, 88 unjudgeable).
2. D4's hypothetical/pseudogene → Pfam/KO route was entirely unreachable: gene_identity_text collects Dbxref/Name as well as product, and on NCBI CDS those are accessions (Dbxref on 29,852 of 30,273 measured CDS), so every hypothetical protein still had non-empty "identity text" and was scored a decided failed. Identifier-shaped values are now filtered, with a positive control asserting the filter removes identifiers rather than removing judgement.

⚠ The vocabulary is precision-first because the permissive version was measured first

An "amino-acid word + enzyme word" rule swallows HAMP domain-containing histidine kinase (38 CDS), serine/threonine protein kinase, glutamine amidotransferase (purine/His/Trp — not tRNA-dependent transamidation), methylated-DNA--[protein]-cysteine S-methyltransferase, peptide-methionine (S)-S-oxide reductase. Every EXCLUSIONS entry was observed on this corpus first; exclusions apply before every include rule; transamidation rules all require the literal tRNA. Two ordering corrections from the audit: the curated product name outranks the gene symbol (hisJ is the His transporter's binding protein in enterics and histidinol phosphatase in Firmicutes), and the symbol route fires only when the product says nothing — otherwise it put PTS galactitol transporter subunit IIB into transamidation via gatB, cytochrome C-552 into transport via cycA, and homocitrate synthase into aaRS via lysS (3 of its 24 matches wrong).

The four functional classes carry a §10.1 gate with four agreeing sources: DOI:10.1002/wrna.1273, PMID:18359782, PMID:19258532, DOI:10.1073/pnas.1220991110 (accessed 2026-08-07).

⚠ The contig bridge is by identity — an index join would be wrong on half the corpus

:c<ci> indexes FASTA record order; a GFF seqid is the NCBI contig accession. 23 of 48 candidate-carrying annotated hosts order their GFF seqids differently from their FASTA records. The id join resolves 134 of 134 needed (host, contig) pairs.

Three gaps reported, not papered over

1. The shipped §9.1 decoy rows carry coordinates on 0 annotated production hosts — decoys_v0's three non-structured pools have no accession, its 2,999 structured_rna accessions are Rfam contigs, mining_pool_v0's 45,988 windows sit on TBDB source genomes. That arm is available: false with its counts; the §9.1 leader class is re-instantiated inside the annotated corpus instead.
2. The false-FAIL arm has 2 TBDB records on 1 assembly — the production substrate is a negative substrate by construction. Reported with its n, never as a rate.
3. The joint a ∧ b ∧ c arm is withheld — (b) has no backend until P3-15′-d.

Plus: the Pfam/KO fallback DB is an unmet §10.2 acquisition, so HMM_FALLBACK_AVAILABLE = False, hypothetical/pseudogenized ORFs resolve unavailable, and ADR-0005 D14 spares them (fail-closed). The population is counted rather than assumed.

Per-clade exclusion — the cap D4 commissioned, and it is large

31.8 % of candidates unavailable (286 host-unannotated, 11 unjudgeable, 2 contig-absent). Five phyla at 100 % (Chloroflexota, Desulfobacterota, Fibrobacterota, Thermotogota, Bdellovibrionota_C); Bacillota 38.5 %; Atribacterota / Sysuimicrobiota / Fusobacteriota 0 %. That is the recall cap the commissioned bakta/prokka arm exists to lift. ⚠ CORRECTED (review round 1). This description originally said "passing distances p50 117 / p95 269 / p99 364 / max 454 bp" — that was a superseded run, and the committed artifact said max 1652 over n 554. Two coupled defects: the statistic did not filter on the strand detail's own status, and the tandem carve-out re-anchored the search but not the decision. Both fixed; five candidates move failed → passed. Decision distance (what criterion_c judges): p50 117 / p95 266 / p99 356 / max 454 bp, entirely inside the pad. Element-relative distance: p50 117 / p95 272 / p99 389 / max 1652 bp, with n_passed_via_tandem_carve_out: 5 explaining the tail.

Validation

• full unit+golden 4,621 passed / 13 failed vs a matched baseline worktree at 95d0149 (4,492 / 13) — failure sets identical, comm empty both ways
• 28 sabotages / 28 RED against their NAMED test
• both committed reports re-derive byte-identically from the final code apart from provenance
• ruff 0.15.15 + black 25.11.0 clean; no .smk touched; no envs/* edit; LOCAL only (48 hosts / 120,089 CDS in 5.7 s)

⚠ n_mined is still structurally 0

Sparing is a disjunction so mining is a conjunction: relaxed_architecture remains the one unavailable backend, so max_possible_yield caps the round at 0 until P3-15′-d lands. This step removes one of the two blockers, not both.

https://claude.ai/code/session_017VRtJL7TgZioPxpCUuENmM

Summary by CodeRabbit

• New Features

  • Added synteny-based candidate evaluation using downstream gene context, strand awareness, distance limits, and configurable intervening-gene rules.
  • Added synteny status tracking across mining and re-mining workflows, including unavailable-state handling and fail-closed validation.
  • Added shard processing, status-table merging, supply inspection, and command-line reporting.
  • Added false-pass and exclusion diagnostic reports with clade, pseudogene, availability, and sensitivity metrics.

• Tests

  • Added comprehensive coverage for synteny evaluation, diagnostics, validation, configuration, and workflow integration.

---

⚠ Review round 1 — 20 findings (app 5, CLI 15), all real, 8 major, one in common

The major one: the tandem carve-out reached the search but not the decision. resolve_downstream_gene selected its target with the re-anchored window and then returned that target's element-relative distance, which synteny_status judged against the 500 bp pad — so D4's carve-out fired only where it was not needed and turned exactly the tandem loci it exists for into false FAILs. DownstreamGene now carries both distances and criterion_c sees decision_distance_bp. My own test could not see it: it asserted the walk reached the right gene, which was always true. The search was never wrong; the decision was.

Two artifact-pinning tests went green under sabotage — the campaign's best output. The distance-statistic and provenance tests read the committed report, generated by the fixed code, so breaking the source could not reach them. TestDiagnosticsComputedNotJustCommitted now calls the report builders directly, so the rule is under test rather than one frozen output of it.

A test that could not catch the trap it named. test_no_clause_reads_dvc_tracked_data derived against REPO_ROOT, so on any checkout that has run dvc pull a DVC-path clause would read True and it would still pass. It now derives against a tmp_path root holding only the SUPPLY_CLAUSES files.

The produced table was never read by either round — both reviewers found it independently. --synteny-status is wired into both apply-spare-rule legs, and declaring the backend without a table (or supplying one without declaring it) now raises, the same answer apply_remine_spare_rule's probe_set already got for the same reason.

Also fixed. An absolute /home/<user>/… path in committed provenance (out-of-repo inputs are hashed as {name, sha256}); the unavailability reason scanning both strands under a single-strand policy; merge_status_tables passing two inputs through unvalidated and accepting an empty shard; _upstream_window bounding the plus arm but not the minus; combine_strand_statuses raising a bare KeyError; load_contig_ids raising IndexError on a bare > header; synteny._gene_symbols private while the producer's control arm depended on it.

[coderabbitai]

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

Use the following commands to manage reviews:

• @coderabbitai resume to resume automatic reviews.
• @coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

• ▶️ Resume reviews
• 🔍 Trigger review

📝 Walkthrough

Walkthrough

Changes

The PR adds ADR-0006 D4 synteny evaluation, annotation-backed producer workflows, diagnostic reports, and synteny status propagation into mining and re-mining evidence.

Synteny evaluation

Layer / File(s)	Summary
D4 synteny backend and validation src/tbox_finder/mining/synteny.py, tests/unit/test_synteny.py	Adds identity classification, coordinate mapping, downstream CDS traversal, tandem carve-outs, status folding, and unit coverage.

Producer execution and outputs

Layer / File(s)	Summary
Producer execution, diagnostics, and reports src/tbox_finder/mining/synteny_producer.py, reports/p3/synteny_*.json, tests/unit/test_synteny_producer.py	Adds shard status tables, control diagnostics, exclusion diagnostics, supply derivation, CLI commands, provenance output, and validation tests.

Evidence integration

Layer / File(s)	Summary
Round and re-mining evidence wiring src/tbox_finder/mining/mine_round.py, src/tbox_finder/mining/remine.py, tests/unit/test_synteny_producer.py	Propagates per-candidate synteny statuses, defaults missing entries to unavailable, and validates supply declarations and status-table inputs.

Estimated code review effort: 5 (Critical) | ~120 minutes

Sequence Diagram(s)

sequenceDiagram
  participant CLI
  participant SyntenyProducer
  participant SyntenyEvaluator
  participant GFF3Annotations
  participant StatusTable
  CLI->>SyntenyProducer: Run shard evaluation
  SyntenyProducer->>SyntenyEvaluator: Evaluate candidate locus
  SyntenyEvaluator->>GFF3Annotations: Load contig and CDS annotations
  GFF3Annotations-->>SyntenyEvaluator: Return downstream CDS features
  SyntenyEvaluator-->>SyntenyProducer: Return synteny status
  SyntenyProducer->>StatusTable: Build and validate status table
  StatusTable-->>CLI: Write results and diagnostics

Loading

Possibly related PRs

• bioedca/tbox-finder#86: Shares candidate-evidence and manifest wiring patterns.
• bioedca/tbox-finder#111: Shares re-mining supply and preflight integration.
• bioedca/tbox-finder#113: Provides related GFF3 and annotation-input integration.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 52.66% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly identifies the criterion (c) synteny predicate, per-candidate producer, and two diagnostics that form the main changes.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches 💡 1

📝 Generate docstrings 💡

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch p3-15c-ii-synteny

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 5

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

src/tbox_finder/mining/mine_round.py (1)

524-526: 📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

The docstring no longer matches the signature.

The text says the other two disjuncts stay unavailable because no synteny backend exists at P2. downstream_aaRS_synteny is now produced from synteny_status. Update the bullet to describe the new third mode, and keep the absent-id fail-closed rule stated.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/tbox_finder/mining/mine_round.py` around lines 524 - 526, Update the
docstring around the SpareRuleEvidence handling to reflect that
downstream_aaRS_synteny is now derived from synteny_status, documenting the new
third mode instead of claiming no synteny backend exists at P2. Preserve the
statement that an absent identifier fails closed by being spared rather than
mined.

🧹 Nitpick comments (4)

src/tbox_finder/mining/synteny.py (2)

78-98: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

__all__ omits two public names that other modules already use.

combine_strand_statuses and excluded_by are used by src/tbox_finder/mining/synteny_producer.py and by the tests, but they are not exported. Add them so the public surface matches the actual contract.

♻️ Proposed export addition

     "classify_gene_identity",
+    "combine_strand_statuses",
     "contig_seqid",
     "criterion_c",
     "downstream_cds_on_strand",
+    "excluded_by",
     "load_contig_ids",

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/tbox_finder/mining/synteny.py` around lines 78 - 98, Update the __all__
declaration in synteny.py to include the existing public symbols
combine_strand_statuses and excluded_by, preserving the current exports and
ordering style.

---

520-528: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

_gene_symbols is private but is called from another module.

src/tbox_finder/mining/synteny_producer.py line 710 calls synteny._gene_symbols(c). Either promote this helper to a public name, or expose a public wrapper that classifies a CdsFeature directly. That would also remove the duplicated "identity text plus symbols" call pattern in the producer.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/tbox_finder/mining/synteny.py` around lines 520 - 528, Expose the
symbol-extraction behavior from _gene_symbols through a public helper, or rename
it to a public symbol, and update synteny_producer’s call site to use that API
instead of accessing the private helper. Preserve the existing lower-casing,
attribute keys, length limit, and frozenset result while consolidating the
producer’s duplicated identity-text and symbol classification flow.

tests/unit/test_synteny_producer.py (1)

338-348: 🎯 Functional Correctness | 🔵 Trivial | ⚡ Quick win

Add a pin for the passing-distance block.

No test checks that passing_distance_sensitivity describes passing rows only. That gap let max_bp: 1652 ship against window_bp: 500. Add an assertion that max_bp is at most window_bp and that n does not exceed the passed count. This test would go red against the current committed report, which is the point.

💚 Proposed test

+    def test_the_passing_distances_are_inside_the_window(self, exclusion: dict) -> None:
+        block = exclusion["passing_distance_sensitivity"]
+        assert block["max_bp"] is None or block["max_bp"] <= block["window_bp"]

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/unit/test_synteny_producer.py` around lines 338 - 348, Add assertions
to the passing_distance_sensitivity test coverage ensuring max_bp is no greater
than window_bp and n does not exceed the number of passed rows. Anchor the
checks to the existing exclusion fixture/report symbols and preserve the current
reconciliation assertions.

src/tbox_finder/mining/synteny_producer.py (1)

479-500: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Contig extent is approximated by the last CDS end.

extents uses the maximum CDS end on the contig as the contig length. Random control windows therefore never fall in the region after the last annotated CDS. State this bias in the docstring, or read the real contig lengths from the genome FASTA that HostAnnotation already opens.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/tbox_finder/mining/synteny_producer.py` around lines 479 - 500, Update
the function containing the extents calculation to avoid treating the maximum
CDS end as the contig length: use the real contig lengths from the FASTA already
opened by HostAnnotation when available, and preserve the existing fallback only
if necessary. If the approximation remains, explicitly document this sampling
bias in the function docstring.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@reports/p3/synteny_exclusion_diagnostic.json`:
- Around line 251-254: Update the diagnostic provenance generation for the
inputs recorded in _provenance so committed keys use repository-relative paths
instead of absolute personal temporary paths. Rerun the diagnostics with
synteny_status.json located under a repository-relative location, or normalize
the existing inputs keys before writing the report while preserving their
checksums.

In `@src/tbox_finder/mining/mine_round.py`:
- Around line 508-511: Update candidate_evidence and both spare-rule call paths
to propagate the loaded synteny status map whenever synteny_available=True: pass
it into read_fp_manifest and read_remine_manifest, while preserving None when
synteny is unavailable so existing fail-closed behavior remains correct.

In `@src/tbox_finder/mining/synteny_producer.py`:
- Around line 816-824: Update the distance collection in the loop over
per-strand details, and the corresponding logic around the referenced second
occurrence, to include distances only when the detail’s own status indicates it
passed and its distance is within the configured window. Keep class-count and
other counters unchanged, then regenerate both affected reports so their
passing-distance statistics reflect only passing candidates.
- Around line 291-295: Update the unavailability-reason logic in the
status-folding function so pseudo detection examines only the strand selected by
config.strand_policy when it is plus or minus; retain scanning across both
strands for the policy that considers both, ensuring reason matches the strand
that determined status.

In `@src/tbox_finder/mining/synteny.py`:
- Around line 795-837: Update the downstream-gene selection flow around the
re-anchoring of anchor in the relevant mining function to preserve both the
original element-relative distance and the re-anchored distance. Use the
re-anchored distance when evaluating or passing the target to criterion_c, while
retaining the original distance for the reported DownstreamGene result. Add a
synteny_status regression test covering a tandem carve-out target beyond 500 bp
from the original element but within the re-anchored window.

---

Outside diff comments:
In `@src/tbox_finder/mining/mine_round.py`:
- Around line 524-526: Update the docstring around the SpareRuleEvidence
handling to reflect that downstream_aaRS_synteny is now derived from
synteny_status, documenting the new third mode instead of claiming no synteny
backend exists at P2. Preserve the statement that an absent identifier fails
closed by being spared rather than mined.

---

Nitpick comments:
In `@src/tbox_finder/mining/synteny_producer.py`:
- Around line 479-500: Update the function containing the extents calculation to
avoid treating the maximum CDS end as the contig length: use the real contig
lengths from the FASTA already opened by HostAnnotation when available, and
preserve the existing fallback only if necessary. If the approximation remains,
explicitly document this sampling bias in the function docstring.

In `@src/tbox_finder/mining/synteny.py`:
- Around line 78-98: Update the __all__ declaration in synteny.py to include the
existing public symbols combine_strand_statuses and excluded_by, preserving the
current exports and ordering style.
- Around line 520-528: Expose the symbol-extraction behavior from _gene_symbols
through a public helper, or rename it to a public symbol, and update
synteny_producer’s call site to use that API instead of accessing the private
helper. Preserve the existing lower-casing, attribute keys, length limit, and
frozenset result while consolidating the producer’s duplicated identity-text and
symbol classification flow.

In `@tests/unit/test_synteny_producer.py`:
- Around line 338-348: Add assertions to the passing_distance_sensitivity test
coverage ensuring max_bp is no greater than window_bp and n does not exceed the
number of passed rows. Anchor the checks to the existing exclusion
fixture/report symbols and preserve the current reconciliation assertions.

🪄 Autofix

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 5c1efa48-0a48-45b3-8ba2-7df6dcadf81d

📥 Commits

Reviewing files that changed from the base of the PR and between 95d0149 and ba04aea.

⛔ Files ignored due to path filters (1)

• analyses/phase3_log.qmd is excluded by !**/*.qmd

📒 Files selected for processing (8)

• reports/p3/synteny_exclusion_diagnostic.json
• reports/p3/synteny_false_pass.json
• src/tbox_finder/mining/mine_round.py
• src/tbox_finder/mining/remine.py
• src/tbox_finder/mining/synteny.py
• src/tbox_finder/mining/synteny_producer.py
• tests/unit/test_synteny.py
• tests/unit/test_synteny_producer.py

[coderabbitai]

Actionable comments posted: 1

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (2)

src/tbox_finder/mining/mine_round.py (1)

1304-1314: 🗄️ Data Integrity & Integration | 🟠 Major | ⚡ Quick win

Add the synteny supply evidence gate to _cmd_apply_spare_rule.

When args.synteny_available is true, call derive_synteny_supply_available(). Return exit code 4 when the derivation is unavailable, before calling apply_spare_rule. The current path can otherwise pass an unevidenced criterion-(c) backend to run_mine_round.

src/tbox_finder/mining/remine.py (1)

771-797: 🩺 Stability & Availability | 🟡 Minor | ⚡ Quick win

Wire the synteny status table into slurm/p3/stage1_remine.sbatch. The remine CLI defaults synteny to available, but this driver omits --synteny-status; the apply step therefore raises before writing the report. Catch this configuration error and return the existing distinct error code. Apply the same guard to the P2 driver if it passes --synteny-available.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/tbox_finder/mining/remine.py` around lines 771 - 797, Update the
apply-spare-rule invocation in slurm/p3/stage1_remine.sbatch to provide the
required --synteny-status input whenever --synteny-available is enabled, and
catch the resulting configuration error so the driver returns the existing
distinct error code instead of aborting before writing the report. Apply the
same argument and error guard to the P2 driver wherever it passes
--synteny-available, while preserving behavior when synteny is unavailable.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@tests/unit/test_synteny_producer.py`:
- Around line 453-456: Replace the `--help`-based assertion in the
`apply-spare-rule` test with a meaningful option-recognition check: expose
parser construction through `build_parser()` and inspect the subparser’s option
strings, or parse a complete command and verify an unknown option exits 2 while
`--synteny-status` reaches `_cmd_apply_spare_rule`. Ensure the test fails if
`--synteny-status` is removed.

---

Outside diff comments:
In `@src/tbox_finder/mining/remine.py`:
- Around line 771-797: Update the apply-spare-rule invocation in
slurm/p3/stage1_remine.sbatch to provide the required --synteny-status input
whenever --synteny-available is enabled, and catch the resulting configuration
error so the driver returns the existing distinct error code instead of aborting
before writing the report. Apply the same argument and error guard to the P2
driver wherever it passes --synteny-available, while preserving behavior when
synteny is unavailable.

🪄 Autofix

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 78d59d76-5d9d-411f-90d2-cd4fbcb6dda1

📥 Commits

Reviewing files that changed from the base of the PR and between ba04aea and 4b49352.

⛔ Files ignored due to path filters (1)

• analyses/phase3_log.qmd is excluded by !**/*.qmd

📒 Files selected for processing (8)

• reports/p3/synteny_exclusion_diagnostic.json
• reports/p3/synteny_false_pass.json
• src/tbox_finder/mining/mine_round.py
• src/tbox_finder/mining/remine.py
• src/tbox_finder/mining/synteny.py
• src/tbox_finder/mining/synteny_producer.py
• tests/unit/test_synteny.py
• tests/unit/test_synteny_producer.py

🚧 Files skipped from review as they are similar to previous changes (5)

• reports/p3/synteny_false_pass.json
• reports/p3/synteny_exclusion_diagnostic.json
• src/tbox_finder/mining/synteny_producer.py
• src/tbox_finder/mining/synteny.py
• tests/unit/test_synteny.py

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@reports/p3/synteny_false_pass.json`:
- Around line 13-30: The class-excluded diagnostic must be derived by filtering
the wide-arm candidate population so its row count cannot exceed the wide arm;
regenerate reports/p3/synteny_false_pass.json (lines 13-30) or revise its
wording so the rates are not presented as one arm and its subset. Add a
validation in tests/unit/test_synteny_producer.py (lines 317-325) asserting
strict["n"] <= wide["n"] to catch future inconsistencies.

🪄 Autofix

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 19781ecd-35fc-4fd9-ad91-4a3283032778

📥 Commits

Reviewing files that changed from the base of the PR and between 4b49352 and b61fd3a.

⛔ Files ignored due to path filters (1)

• analyses/phase3_log.qmd is excluded by !**/*.qmd

📒 Files selected for processing (7)

• reports/p3/synteny_exclusion_diagnostic.json
• reports/p3/synteny_false_pass.json
• src/tbox_finder/mining/mine_round.py
• src/tbox_finder/mining/synteny.py
• src/tbox_finder/mining/synteny_producer.py
• tests/unit/test_synteny.py
• tests/unit/test_synteny_producer.py

🚧 Files skipped from review as they are similar to previous changes (4)

• reports/p3/synteny_exclusion_diagnostic.json
• tests/unit/test_synteny.py
• src/tbox_finder/mining/synteny_producer.py
• src/tbox_finder/mining/synteny.py

[coderabbitai]

Actionable comments posted: 2

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@reports/p3/synteny_false_pass.json`:
- Around line 75-79: Reconcile the positive-context control metric in the
report: regenerate it so the stored counts, pass rate, objective, and
control.margin all derive from the same source data. Update the report’s
objective and margin if the 99.414% count-derived rate is correct, or correct
the counts and rate consistently if 99.35% is required.
- Around line 22-30: The strict-arm report must not claim to use the same
sampled windows when false_pass_report independently selects spans with
rng.choice(lengths). Reuse the wide-arm (CDS, span, window) records when
constructing the strict arm, then regenerate the report; alternatively, if
independent window sampling is intentional, revise the note to state that only
the nested CDS population is shared.

🪄 Autofix

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 11049879-fa3a-4744-97b7-bd9059fe006a

📥 Commits

Reviewing files that changed from the base of the PR and between b61fd3a and c7beb0a.

⛔ Files ignored due to path filters (1)

• analyses/phase3_log.qmd is excluded by !**/*.qmd

📒 Files selected for processing (4)

• reports/p3/synteny_exclusion_diagnostic.json
• reports/p3/synteny_false_pass.json
• src/tbox_finder/mining/synteny_producer.py
• tests/unit/test_synteny_producer.py

🚧 Files skipped from review as they are similar to previous changes (3)

• reports/p3/synteny_exclusion_diagnostic.json
• tests/unit/test_synteny_producer.py
• src/tbox_finder/mining/synteny_producer.py

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@tests/unit/test_synteny_producer.py`:
- Around line 772-780: Update the strict-arm assertions in
test_an_all_non_d4_sample_keeps_every_window and the preceding strict-window
test to validate object identity rather than tuple value equality: use is-based
checks for each corresponding window, while preserving the existing ordering and
coverage expectations.

🪄 Autofix

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: fdae0a46-b98a-4d3f-b026-f61c3e98580d

📥 Commits

Reviewing files that changed from the base of the PR and between c7beb0a and b6234ec.

⛔ Files ignored due to path filters (1)

• analyses/phase3_log.qmd is excluded by !**/*.qmd

📒 Files selected for processing (4)

• reports/p3/synteny_exclusion_diagnostic.json
• reports/p3/synteny_false_pass.json
• src/tbox_finder/mining/synteny_producer.py
• tests/unit/test_synteny_producer.py

🚧 Files skipped from review as they are similar to previous changes (3)

• reports/p3/synteny_false_pass.json
• reports/p3/synteny_exclusion_diagnostic.json
• src/tbox_finder/mining/synteny_producer.py