Skip to content

Audit a Real AiZynthFinder Route with RENKIN

Keep AiZynthFinder. Audit its routes with RENKIN.

RENKIN Bridge audits route JSON from a competitor tool through the same tool-neutral pipeline it applies to its own routes: structural integrity, stock, and declared-reaction forward-replay validation, each reported independently, rolled up into a route-level pass/fail/partial verdict. This page walks through it end to end against real captured aizynthcli 4.4.1 output — not a hand-authored example — so every command and every line of output below is something you can reproduce yourself from a checkout of this repo.

The fixture

tests/fixtures/aizynthfinder/v4.4.1/single_trees.json is real output from a real aizynthcli --smiles "CCOC(=O)c1ccc(N)cc1" ... run (benzocaine), captured once and committed for CI to use without needing AiZynthFinder installed. Full capture provenance — exact command, model/stock file SHA-256 hashes, ZINC stock size — is in the sibling PROVENANCE.md.

Step 1: audit without a configured stock

renkin audit-route tests/fixtures/aizynthfinder/v4.4.1/single_trees.json \
  --format aizynthfinder \
  --output human

3 routes audited — 0 pass, 0 fail, 3 partial
route 1/3: PARTIAL
  - stock: StockNotProvided
route 2/3: PARTIAL
  - stock: StockNotProvided
route 3/3: PARTIAL
  - stock: StockNotProvided

Every route comes back PARTIAL, not PASS. This is correct, not a weaker result. No --stock was given, so RENKIN has nothing to check the leaf molecules against — it reports not_evaluable (stock_not_provided) rather than silently treating "we didn't check" as "it's fine." partial means at least one check couldn't reach a verdict and nothing outright failed; it is a distinct, three-valued outcome from pass, never collapsed into a boolean.

Step 2: audit with a configured stock

renkin audit-route tests/fixtures/aizynthfinder/v4.4.1/single_trees.json \
  --format aizynthfinder \
  --stock data/building_blocks.smi \
  --output human

3 routes audited — 1 pass, 2 fail, 0 partial
route 1/3: FAIL
  - LeafClaimedStockNotMatched
route 2/3: PASS
route 3/3: FAIL
  - LeafClaimedStockNotMatched

Now RENKIN checks every leaf molecule against data/building_blocks.smi (RENKIN's own default building-block set, unrelated to AiZynthFinder's own in_stock claims). Route 2 passes outright — its two precursors (ethanol and 4-aminobenzoic acid) are both in RENKIN's stock, and every step's declared reaction replays correctly. Routes 1 and 3 fail because a precursor AiZynthFinder's own policy proposed isn't in this particular stock file — a real, informative disagreement about what counts as "available," not a parsing or adapter defect.

--output json gives the same verdicts as a machine-readable report (schema_version, an audit_manifest recording what was audited and under what conditions — RENKIN version, source format, input/stock content hashes, audit policy — for reproducing the same audit later, per-route status/stock_validation/steps/findings, and a route-level summary) — pipe it to jq/python -m json.tool/whatever you already use for the RENKIN-native report shape.

What --format auto does here

Both commands above pass --format aizynthfinder explicitly. Omit it and --format auto (the default) detects the same result on its own — reading single_trees.json's top-level shape (an array whose first element has type: "mol") as AiZynthFinder's single-target export, distinct from RENKIN-native JSON's {"target": ..., "routes": [...]} shape or a batch export's Pandas {"schema": ..., "data": [...]} shape. If the input doesn't match any of those three recognized shapes, audit-route hard-errors rather than guessing.

Compatibility

Source Verified version Input
RENKIN 0.26.0 native JSON
AiZynthFinder 4.3.2, 4.4.0, 4.4.1 single-target JSON, batch JSON, gzip-compressed batch JSON
Other planners / other AiZynthFinder versions unverified unknown fields are tolerated (forward-compatible); a shape RENKIN doesn't recognize is a hard error, never a guess

Syntheseus and SynPlanner are also verified adapters (--format syntheseus / --format synplanner) — see the Syntheseus audit demo and SynPlanner audit demo for their own walkthroughs and compatibility rows.

"Verified against" is deliberate phrasing, not "supported" — this adapter is confirmed against real, individually captured aizynthcli output from three separate versions specifically (4.3.2, 4.4.0, 4.4.1 — see each version's own tests/fixtures/aizynthfinder/vX.Y.Z/PROVENANCE.md), not claimed to work with every AiZynthFinder release. All three versions were run against the identical public model/stock data bundle (confirmed byte-identical by SHA-256 across all three captures) and the identical target molecules, so any behavioral difference found is attributable to the aizynthfinder package itself, not to different inputs — tests/aizynthfinder_version_matrix.rs asserts all three produce identical audit verdicts for the same real routes. One confirmed, harmless cross-version JSON difference was found in the process: 4.3.2's route scores object carries an extra "average template occurrence" field absent from 4.4.0/4.4.1's output — outside the tree structure RENKIN's normalizer reads, so it doesn't affect any verdict (see v4.3.2's own PROVENANCE.md for the full finding). This is one instance of a general rule every adapter follows — see the Audit Reproducibility and Compatibility Contract for the full set, including what audit_manifest guarantees and how a new adapter or fixture is added.

Hit a compatibility problem?

If audit-route rejects real output from AiZynthFinder (any version) or another planner, or a real route gets flagged not_evaluable in a way that looks wrong, please open an adapter-compatibility issue — it's set up to capture exactly what turns a real report into the next test fixture.