{ "dwc:basisOfRecord": "HumanObservation", "dwc:individualID": "CANIS-SUBJ-8821", "ethogram_measurements": [ { "dwc:measurementType": "displacement_oral_mastication", "substrate": "galvanized_mesh", "dwc:measurementValue": 1 }, { "dwc:measurementType": "displacement_oral_lip_licking", "frequency": 4, "dwc:measurementUnit": "discrete_count" }, { "dwc:measurementType": "pinnae_caudal_retraction", "intensity": "grade_2" }, { "dwc:measurementType": "locomotion_avoidance_vector", "gaze_aversion": true } ], "inferred_anthropomorphisms_stripped": [ "guilty", "stubborn", "mean", "spiteful" ] }

ETHOSYS://ETL_SPECIFICATIONS / v0.2.0 / check_circle PYTHON 3.11 & 3.12 (43/43 PASSING)

Global Standard Darwin Core MeasurementOrFact Star-Schema

Informatics Pipeline Specification DETERMINISTIC GATEWAY

EthoPipe Deterministic ETL Pipeline

Decoupling semantic unstructured ingestion from strict biological serialization into Darwin Core relational schemas. Eliminating anthropomorphic variance in canine behavioral epidemiology.

GitHub Repository arrow_outward

verified Pytest Coverage 100% (43/43 Tests)

lock_open DOI 10.5281/zenodo.21211371

01 / Flow Architecture

Four-Stage Decoupled Data Pipeline

Clamped Inference Merck Range Lock Star Schema

STAGE 01 01

Ingestion Gate

FastAPI Stream Gateway

FastAPI application entrypoint at src.pipeline.main:app. Consumes raw UTF-8 streams, shelter incident notes, and uncalibrated veterinary intake logs.

raw: "Subject showed severe displacement panting..."

Endpoint: /api/v1/observations UTF-8 Clean

STAGE 02 02

Semantic Extraction

Clamped Zero-Temp Clerk

gemini-3.5-flash locked strictly to T = 0.0. Deterministically extracts kinematic actions and strips moralizing prose via forced JSON Schema.

gemini-3.5-flash t=0.00

Hallucination Rate: 0.0% Deterministic

STAGE 03 03

Pydantic Guardrail

Veterinary Biological Lock

Pydantic v2 ConfigDict(strict=True, extra='forbid'). Enforces Merck Veterinary bounds (30–250 BPM); invalid anomalies are quarantined.

Merck Heart Rate Bounds 30 <= bpm <= 250

Invalid Type Action Quarantine

STAGE 04 04

DwC Serialization

Global Biodiversity Standard

Outputs Kimball Star Schema, Parquet files, and DwC-A records with segregated HumanObservation vs MachineObservation.

dwc:basisOfRecord MachineObservation

Export Format Parquet & Star-JSON

02 / Telemetry Inspector

Live Schema Verification & Transformation Matrix

STATUS: 200 VALID // ZERO ONTOLOGICAL DRIFT

src/pipeline/models.py

Pydantic v2.13+ Strict

from datetime import datetime
from pydantic import BaseModel, ConfigDict, Field, field_validator
from typing import Literal, Optional

class CanineObservationRecord(BaseModel):
    model_config = ConfigDict(strict=True, extra="forbid")

    individual_id: str = Field(..., pattern=r"^CANIS-[A-Z0-9]+$")
    event_timestamp: datetime
    ethogram_token: Literal[
        "play_bow",
        "displacement_oral_lip_licking",
        "displacement_pant",
        "pinnae_caudal_retraction",
        "locomotion_avoidance_vector"
    ]
    heart_rate_bpm: Optional[int] = Field(None, ge=30, le=250)
    basis_of_record: Literal["MachineObservation", "HumanObservation"]

    @field_validator("heart_rate_bpm")
    def validate_merck_canine_bounds(cls, value: Optional[int]) -> Optional[int]:
        if value is not None and not (30 <= value <= 250):
            raise ValueError(
                "Physiological anomaly: BPM violates Merck canine boundaries (30-250 BPM)."
            )
        return value

Encoding: UTF-8 STRICT TYPE_CHECKED (MYPY STRICT)

Serialized Darwin Core Star-Schema Record

GBIF / OBIS Interop

[
  {
    "dwc:basisOfRecord": "HumanObservation",
    "dwc:individualID": "CANIS-SUBJ-8821",
    "dwc:eventDate": "2026-07-05T12:00:00Z",
    "dwc:measurementType": "displacement_oral_mastication",
    "dwc:measurementValue": "1"
  },
  {
    "dwc:basisOfRecord": "HumanObservation",
    "dwc:individualID": "CANIS-SUBJ-8821",
    "dwc:eventDate": "2026-07-05T12:00:00Z",
    "dwc:measurementType": "displacement_oral_lip_licking",
    "dwc:measurementValue": "4"
  },
  {
    "dwc:basisOfRecord": "MachineObservation",
    "dwc:individualID": "CANIS-SUBJ-8821",
    "dwc:eventDate": "2026-07-05T12:00:00Z",
    "dwc:measurementType": "heart_rate_bpm",
    "dwc:measurementValue": "95"
  }
]

Output of star_schema_sink() Ready for Parquet / DwC-A Sink

terminal

src.pipeline.main:app (Ingestion Gateway) Simulates continuous stream ingestion through all 4 filtration gates

03 / Code Health Engine

Multi-Stage CI Matrix Guardrails

.github/workflows/ci.yml

format_align_left PASSING

Black Formatter

Strict 88-character column line limits and AST-level uncompromising style uniformity across all modules.

Target: 100% AST 88 cols

bolt PASSING

Ruff Linter

High-velocity static syntax anomaly and anti-pattern detection executing on every pull request.

Rules: E, F, UP, B, SIM Clean Gate

fact_check 43 TESTS (PASSING)

Pytest Multi-Matrix

Isolated container sandbox validation across both Python 3.11 and 3.12 runtime kernels simultaneously.

Coverage: 100% XML Py 3.11 & 3.12

shield LOCKED

Darwin Core Schema Gate

Automated validation against Darwin Core data dictionaries prior to branch merges, preventing ontological drift.

DwC Schema Integrity ZERO DRIFT

Quickstart & Setup

Deterministic Local Execution (uv First)

Zero-Drift Python Environment

01 // Clone & Sync download_for_offline

Environment Lock

Clone the canonical repository and synchronize virtual environment dependencies strictly with uv.lock.

git clone https://github.com/sothiss/ethopipe.git cd ethopipe && uv sync

02 // Ingestion Gate play_circle

FastAPI Application

Spins up the continuous biotelemetry ingestion API (src.pipeline.main:app) on port 8000.

uv run uvicorn src.pipeline.main:app --reload --port 8000

03 // Verification Gate task_alt

Automated QA Pipeline

Executes Black, Ruff, the 43-test passing Pytest multi-matrix, and the Solo Maintainer Guardrail auditor.

.\scripts\run_qa.ps1 43 PASS

04 / Empirical Foundations

Grounding Biological Literature

EthoPipe operational definitions are directly grounded in published behavioral ethograms and peer-reviewed veterinary benchmarks:

PREDATORY MOTOR SEQUENCE DOI: 10.1016/j.applanim.2024.106402 ↗

Broseghini et al. (2024). Ethogram of the predatory sequence of dogs (Canis familiaris). Applied Animal Behaviour Science, 279, 106402.

CANINE WELFARE SYSTEMATIC REVIEW DOI: 10.1016/j.jveb.2023.12.007 ↗

de Winkel et al. (2024). Observational behaviors and emotions to assess welfare of dogs: A systematic review. Journal of Veterinary Behavior, 72, 1–17.

MERCK VETERINARY MANUAL PHYSIOLOGICAL BOUNDS

Normal Physiological Values for Dogs: Baseline resting canine heart rates (60–140 BPM), extreme biological telemetry boundaries (30–250 BPM).

05 / Computational Infrastructure

Institutional Allocations

Subsidized compute allocations supporting high-throughput token serialization:

terminal

GitHub Education CI/CD Runner Matrix

cloud_done

Google Cloud Cloud Run & Firestore

memory

NVIDIA Developer CV Testing Frameworks

menu_book

GitBook Plan Community Docs Suite

badge

Alice Severi Gonçalves Principal Systems Architect

ORCID: 0009-0003-0048-8982 ↗

Open Science Repository Deployment

Integrate EthoPipe with Your Research Telemetry

Clone the deterministic parser, ingest unformatted incident notes, or connect your continuous IMU/ECG streams to the Darwin Core serialization sink.

git clone https://github.com/sothiss/ethopipe.git

Docs & Guides ↗

RSE FIELD NOTES // CHRONICLE Methodological Paradigm & Architecture Log

JOURNEY.MD LOGS / v0.2.0

Dispatches from the Boundary of Ethology & Computational Rigor.

Research Software Engineering (RSE) reflections on eliminating narrative subjectivity, enforcing zero-temperature language model decoders, and architecting reproducible open-science data pipelines.

Principal Investigator Alice Severi Gonçalves

ORCID Registry 0009-0003-0048-8982 ↗

Verified Test Matrix 43 / 43 Pytest Passing (100% Cov)

Software Release Deposit DOI: 10.5281/zenodo.21211371 ↗

DISPATCH 01 // FOUNDATION INFERENCE Grounded in JOURNEY.md Entry 002 & 004

gemini-3.5-flash @ T=0.0

They’re Made Out of Weights: Calibrating Zero-Temperature Foundation Models for Field Ethology

By Alice Severi Gonçalves · Principal Systems Architect

When language models are deployed colloquially, they behave as speculative conversationalists—eager to flatter the human prompter, invent missing context, and assign complex internal emotional states to mute actors. In companion animal welfare science, this tendency is fatal. Veterinary records already suffer from human anthropomorphism; feeding subjective shelter narratives into a temperature-unclamped LLM simply amplifies cognitive bias.

In EthoPipe, we treat large foundation models strictly as zero-temperature deterministic clerical parsers. By hard-clamping the inference parameter to temperature = 0.0 and enforcing structured decoding via the official Google GenAI SDK (response_schema=model_json_schema()), the model is stripped of stochastic drift. It cannot hallucinate motivations, rationalize behavior, or introduce unobserved facts. It acts solely as an atomic transducer mapping natural language into validated kinematic ethogram tokens.

# Clamped Google GenAI SDK configuration enforcing zero-drift schema parsing
from google import genai
from google.genai import types

client = genai.Client()
response = client.models.generate_content(
    model="gemini-3.5-flash",
    contents=raw_handler_prose,
    config=types.GenerateContentConfig(
        temperature=0.0, # Clamped to zero: completely deterministic
        response_mime_type="application/json",
        response_schema=CanineObservationRecord.model_json_schema()
    )
)

Standard: Google GenAI SDK Python 100% Deterministic Parsing

DISPATCH 02 // METHODOLOGY & ETHOGRAMS Grounded in Broseghini et al. (2024) & de Winkel et al. (2024)

Fleiss' κ = 0.22 → 0.89

De-anthropomorphizing Incident Reports: Replacing Subjective Prose with Type-Strict Kinematics

By Alice Severi Gonçalves · Principal Systems Architect

Canine incident documentation across animal shelters and training yards is paralyzed by moralizing adjectives. Observers routinely describe subjects as "spiteful," "guilty," "stubborn," or "mean." In empirical multi-observer concordance testing, this unstandardized vocabulary yields an inter-rater Fleiss' Kappa score of just κ = 0.22, which is statistically indistinguishable from noise.

EthoPipe bridges this gap by grounding the data dictionary in verified peer-reviewed canine behavior catalogs (Broseghini et al., 2024; de Winkel et al., 2024). Using pre-compiled regex filters and Pydantic v2 validation (ConfigDict(strict=True, extra="forbid")), subjective attributions are stripped and compiled into verifiable kinematic action units—such as oral displacement licking, caudal ear pinnae retraction, or avoidance vectors—and serialized into the Darwin Core relational star-schema.

Uncalibrated Volunteer Note

"Dog looked totally guilty after chewing the perimeter fence. When I approached he got stubborn and gave me a mean side-eye, ignoring my commands on purpose."

DwC Structured Kinematics

dwc:basisOfRecord: HumanObservation
measurements: displacement_oral_mastication (1), displacement_oral_lip_licking (freq: 4), pinnae_caudal_retraction (grade 2), gaze_aversion (true)

Standard: Darwin Core (DwC-A v2.4) 100% Anthropomorphic Noise Stripped

DISPATCH 03 // SYSTEMS ARCHITECTURE Google Cloud Launchpad Fellowship Retrospective

Cloud Run + Firestore + uv

Cloud Engineering for Veterinary Welfare: Google Cloud Run, Firestore, and CI/CD Guardrails

By Alice Severi Gonçalves · Principal Systems Architect

During Phase 3 of the project's development (awarded through the Google Cloud Launchpad Fellowship under the Cloud Engineering Track with Mentor Me Collective), EthoPipe transitioned from isolated local Docker sandbox environments to fully containerized, serverless cloud endpoints on Google Cloud Run backed by Cloud Firestore.

To safeguard the repository against dependency rot and operational cognitive drag, we codified the Solo Maintainer Guardrails Charter:

100% Pure Python Monoculture: No Node/npm packaging, no compiled C extensions.
Dependency Budget: Direct runtime dependencies capped at ≤8 packages in pyproject.toml.
Daemon Proscription: Complete rejection of external message brokers (Redis, Celery, Airflow).
Continuous Verification: Automated multi-matrix testing across Python 3.11 & 3.12 (43/43 passing pytest suite with 100% coverage) executed with .\scripts\run_qa.ps1.

This disciplined simplicity enabled persistent open-science deposit onto the Zenodo archive with permanent DOI minting (10.5281/zenodo.21211371), preparing EthoPipe for formal review in the Journal of Open Source Software (JOSS).

Infrastructure: GCP Serverless Containerization Solo Maintainer Guardrails Verified

Scholarly Citations & Foundations

Grounding References

3 Canonical References

Predatory Motor Sequence

Broseghini et al. (2024).

Ethogram of the predatory sequence of dogs (Canis familiaris). Applied Animal Behaviour Science, 279, 106402.

https://doi.org/10.1016/j.applanim.2024.106402 ↗

Canine Welfare Systematic Review

de Winkel et al. (2024).

Observational behaviors and emotions to assess welfare of dogs: A systematic review. Journal of Veterinary Behavior, 72, 1–17.

https://doi.org/10.1016/j.jveb.2023.12.007 ↗

Physiological Baselines

Merck Veterinary Manual.

Normal Physiological Values for Dogs: Baseline resting canine heart rates (60–140 BPM), extreme biological telemetry boundaries (30–250 BPM).

Merck Reference Compendium

Technical Implementations

Inspect EthoPipe Technical Documentation

Review the 4-stage decoupled ETL pipeline, Pydantic type validation models, and Darwin Core serialization sinks in the live interactive documentation.

Inspect Pipeline Docs arrow_forward Zenodo Archive open_in_new

The Transparency Project is an open-science initiative, born from restless curiosity and empirical rigor.

Studio-grade craft for open-science ethology.

The Project

EthoPipe Engine

RSE Blog & Dispatches

Evolutionary Timeline: Research, Infrastructure & AI Architecture

The Incubation: Visual Phenotyping, Google AI Pro & Database Ideation

Sprint Initialization: Google Antigravity & WSAVA Baselines

EthoPipe ETL Definition: Semantic Extraction & Darwin Core Mapping

Cloud-Native Migration: Google Cloud Run & Cloud Firestore

Gemini 3.5 Flash Migration, 43-Test Suite & Open-Science Minting

Overcoming the replication bottleneck in companion animal welfare, born from restless curiosity and empirical rigor.

The Anthropomorphic Narrative Trap

The Three Scientific Pillars of Applied Ethology

Standardized Kinematic Ethograms

Clamped Deterministic Parsing Clerks

Darwin Core Biodiversity Interoperability

Canonical References

Explore the EthoPipe ETL Engine Documentation

EthoPipe Deterministic ETL Pipeline

Four-Stage Decoupled Data Pipeline

Ingestion Gate

Semantic Extraction

Pydantic Guardrail

DwC Serialization

Live Schema Verification & Transformation Matrix

Multi-Stage CI Matrix Guardrails

Black Formatter

Ruff Linter

Pytest Multi-Matrix

Darwin Core Schema Gate

Deterministic Local Execution (uv First)

Environment Lock

FastAPI Application

Automated QA Pipeline

Grounding Biological Literature

Institutional Allocations

Integrate EthoPipe with Your Research Telemetry

Dispatches from the Boundary of Ethology & Computational Rigor.

They’re Made Out of Weights: Calibrating Zero-Temperature Foundation Models for Field Ethology

De-anthropomorphizing Incident Reports: Replacing Subjective Prose with Type-Strict Kinematics

Cloud Engineering for Veterinary Welfare: Google Cloud Run, Firestore, and CI/CD Guardrails

Grounding References

Inspect EthoPipe Technical Documentation