Task 4 validation — numeric TaxID to real STRING network

Date: 2026-07-21
Target: planai5090:/home/xzq/5090/PlantAI-dev4090-7/PlantAI-dev
Gate result: PASS

Read-only implementation review

The reviewed Task 4 implementation keeps the execution TaxID separate from the local gene-ID normalization label. Numeric execution taxa must be members of the fixed STRING snapshot; legacy species codes remain compatible. The TaxID survives preflight, API request normalization, worker parameters, backend CLI construction, skill execution, and manifest generation.

skills/ppi/run.py::StringAPI discovers the official stable API address, rejects untrusted redirects, maps inputs through get_string_ids, attributes rows by queryIndex, validates TaxID and full STRING identifiers, and submits only validated IDs to network with add_nodes=0. Network rows are checked for TaxID, submitted endpoints, score range, and threshold. No Task 5 scheduling or UI behavior was changed during this Gate.

CodeGraph was unavailable on the real target (codegraph: command not found, exit 127), so the review used targeted source reads after recording the limitation.

Regression evidence

The first test command referenced a nonexistent tests/unit/test_string_rate_gate.py and exited before collection. The corrected focused command ran only files present in the repository:

tests/unit/test_ppi_numeric_taxon_task4.py
tests/unit/test_ppi_id_resolution.py
tests/unit/test_string_ppi_production_wiring_red.py
tests/unit/test_ppi_string_policy.py

43 passed in 13.90s

Official identifier discovery

Initial official get_string_ids probes for TaxID 3708 were FLC, FT, SOC1, AP1, and CO. STRING mapped FT to 3708.D0QX04. It also mapped CO, but the returned annotation described a magnesium transporter; this homonym was rejected as biologically unsuitable. FLC, SOC1, and AP1 did not map. A second symbol-only probe (LFY, AG, SVP, GI, LHY, TOC1, TFL1, FUL, SEP1, SPL3, SPL9) returned zero mappings.

The official species-specific STRING v12 aliases file was then downloaded from:

https://stringdb-downloads.org/download/protein.aliases.v12.0/3708.protein.aliases.v12.0.txt.gz

It records 3708.A0A078GRX0 with the product description MADS-box protein SOC1. The protein ID was not submitted directly to the network. The current StringAPI.get_string_ids first resolved the aliases-file token A0A078GRX0, together with FT, producing two validated mappings:

Query	STRING ID	TaxID	Preferred name
`FT`	`3708.D0QX04`	3708	`FT`
`A0A078GRX0`	`3708.A0A078GRX0`	3708	`BnaC04g50370D`

Real StringAPI Gate

The Gate imported and executed the current skills.ppi.run.StringAPI; the HTTP recorder only observed the requests and did not bypass or replace its validation.

Step	Method and endpoint	HTTP
version	`GET https://string-db.org/api/json/version`	200
mapping	`POST https://version-12-0.string-db.org/api/json/get_string_ids`	200
network	`POST https://version-12-0.string-db.org/api/json/network`	200

Version response:

{
  "string_version": "12.0",
  "stable_address": "https://version-12-0.string-db.org"
}

Validated request summary:

TaxID: 3708
Mapping input: FT, A0A078GRX0
Mapping result: 2/2
Network input: 3708.D0QX04, 3708.A0A078GRX0
required_score=200
network_type=functional
add_nodes=0
caller_identity=PlantAI

The real network request returned one validated edge:

3708.A0A078GRX0 (BnaC04g50370D)
    -- score 0.467 -->
3708.D0QX04 (FT)

The response TaxID was 3708 and both endpoints were members of the submitted validated ID set. Therefore Task 4 passes the real STRING stable-v12 Gate.

No service was restarted during the real-network Gate above, and no Task 5 work was performed.

Deployment Gate on port 9010

Deployment Gate completed at 2026-07-21 06:04:42 CST. Only the 9010 Web process and the five plantai_4090_9010 worker parents were restarted after a directed inspect active reported all five workers empty.

Process	Old PID	New PID	Queue
Web 9010	2470234	2479563	—
default worker	2470133	2479460	`plantai_4090_9010.default`
heavy worker	2470139	2479470	`plantai_4090_9010.heavy`
medium worker	2470146	2479477	`plantai_4090_9010.medium`
light worker	2470155	2479484	`plantai_4090_9010.light`
rnaseq worker	2470163	2479492	`plantai_4090_9010.rnaseq`

The Web process and every worker reported cwd /home/xzq/5090/PlantAI-dev4090-7/PlantAI-dev. Every worker reported BBAV5_QUEUE_PREFIX=plantai_4090_9010 and BBAV5_REDIS_URL=redis://localhost:6379/11. Redis DB11 returned PONG; a directed Celery ping returned pong from all five nodes; and /api/celery/health returned ok=true, the expected five queues, and all five project workers online.

Port 19000 retained PID 2466042; port 9000 had no listener before or after. Neither was signalled or restarted. The final socket check showed exactly one 9010 listener, PID 2479563.

Deployed request contract

The live GET /openapi.json schema for ExecuteRequest exposes both ppiSpecies and normalizationSpeciesLabel as strings. A fresh deployed Pydantic model probe constructed:

{
  "ppiSpecies": "3708",
  "ppiSpecies_type": "str",
  "normalizationSpeciesLabel": "Bn"
}

model_dump() retained both values exactly, so numeric-looking TaxID 3708 is not converted, truncated, or discarded by Pydantic.

Fresh worker-environment import probe

A separate fresh Python process used the same target cwd, test4py311 interpreter, queue prefix, and DB11 environment as the workers. It imported:

skills.ppi.run:
  /home/xzq/5090/PlantAI-dev4090-7/PlantAI-dev/skills/ppi/run.py
gfa_core.mas.tools.stage_tools:
  /home/xzq/5090/PlantAI-dev4090-7/PlantAI-dev/gfa_core/mas/tools/stage_tools.py

StringAPI.resolve_species_id("3708") returned "3708". A separate fresh process rejected 999999999 with ValueError: STRING TaxID 999999999 is not in the fixed STRING snapshot registry and exit code 1.

No formal PPI task was submitted; end-to-end task submission remains reserved for Task 7. No Task 5 work was performed.

6.2 KiB Raw Permalink Blame History