zer0pa-biomolecular-explorer 0.1.0

Zer0pa Bio-Molecular Explorer — falsification-engine pipeline for molecular and biological-system exploration. Research use only. Not for diagnosis, treatment, cure claims, prescribing, clinical deployment, regulatory compliance, or drug-safety certification.

Zer0pa Bio-Molecular Explorer

Live window into the Zer0pa lab. Bio-Molecular Explorer is a research-only evidence pipeline, not a medical product or safety certification.

What This Is

Research use only. Not for diagnosis, treatment, cure claims, prescribing, clinical deployment, regulatory compliance, or drug-safety certification. Every artifact you produce carries this boundary verbatim.

Zer0pa Bio-Molecular Explorer turns molecular research inputs into auditable, falsifier-governed biological evidence packets. The first authority wedge is cardiac: dofetilide, verapamil, ranolazine, and held-out comparators through KCNH2, SCN5A, KCNQ1, and CACNA1C under multi-current CiPA framing. FDA E14/S7B are regulatory-science anchors only; this repo does not claim regulatory compliance.

The repo currently exposes a CPU-first scaffold with contracts, audit trails, KG records, falsifier state, Runpod-sim adapters, Pathway 1 R&D front-end stubs, and an H100 completion plan. The live top gate is not a green test count. The top gate is an L6-governed authority path where scientific evidence, audit state, KG state, falsifier ledger state, and packet export all agree.

Pipeline Mechanics

Field	Value
Architecture	FALSIFICATION_ENGINE_PIPELINE
Primary Wedge	Cardiac evidence: dofetilide / verapamil / ranolazine plus held-out comparators
Targets	KCNH2, SCN5A, KCNQ1, CACNA1C
Execution	Local-first L2-L6 CPU scaffold; L1 REST stubs; Runpod/H100 work replaces stubs only after authority hardening
Mechanics	universal layer envelopes, append-only audit log, KG, falsifier ledger, L6 router/backedge propagation, blind PubMed-reader gate
Open Gate	WP0 authority hardening before authoritative Runpod cutover

Key Metrics

Metric	Value	Baseline
CPU scaffold tests	768 passing at last recorded verification	unit + integration + falsification + runpod-sim surface
Cardiac stub-state run	9 / 9 seed + held-out compounds pass in recorded scaffold run	not authority until real evidence and blind scoring pass
Falsifier surface	16 cardiac classes + 13 Pathway 1 R&D classes	registry + negative fixtures
Minimum non-toy H100 completion	60-120 H100 hours + 40-80 engineering hours	docs/H100_COMPLETION_PLAN.md

Source: docs/H100_COMPLETION_PLAN.md, docs/RUNPOD_READINESS.md, docs/execution-report.md, PRD.md, and the 2026-05-02 blocker review reflected below.

Repo Identity

Field	Value
Identifier	Bio-Molecular Explorer
Repository	https://github.com/Zer0pa/Bio-Molecular-Explorer
Portfolio	Bio-molecular research infrastructure
Visibility	INTERNAL
Default Branch	main
Authority Source	PRD.md; docs/H100_COMPLETION_PLAN.md; this README blocker register
License	No public license file in this snapshot

Readiness

Field	Value
Evidence posture	Work-in-progress research scaffold; not a deployable medical product
Current state	CPU-governed scaffold with runpod-sim cutover tests and explicit H100 completion plan
Top acceptance gate	L6-governed packet export with blind scientific correctness, falsifier ledger, audit, and KG reconciliation
H100 status	ready to begin completion work; not ready for authoritative Runpod cutover
Next gate	WP0 authority hardening, then H100 real-artifact replacement and falsification wave

Honest Blocker

The current review findings control readiness. The sovereign authority metric is not met while PubMed-reader lift is synthetic or fixture-shaped, while the user-facing cardiac run can bypass L6 authority behavior, while packet assembly can read canned panels instead of validated L1-L5 evidence envelopes, while the cardiac path can write audit rows without populating the governing FalsifierLedger, or while runpod-precheck can pass invalid Runpod states. These must close before the repo can be called authoritative on H100.

What We Prove

• The repo has a concrete CPU-side implementation surface for L1-L6 contracts, Pathway 1 R&D contracts, audit records, KG validation, falsifier registries, and packet assembly.
• Runpod-sim adapters demonstrate the intended plug-replaceability mechanism: backend swaps should preserve envelope shape, falsifier semantics, audit shape, and downstream contracts.
• The cardiac wedge is scoped around concrete compounds and ion-channel targets rather than generic drug-discovery copy.
• The workstream has an explicit H100 plan, time budget, agent topology, and decision mandate for completing the real pipeline instead of demoing a command path.
• The public-facing posture preserves blockers and non-claims rather than converting scaffold progress into a medical or regulatory readiness narrative.

What We Don't Claim

• No diagnosis, treatment, cure, prescribing, clinical deployment, regulatory compliance, or drug-safety certification.
• No completed end-to-end evidence pipeline yet.
• No authoritative Runpod/H100 cutover yet.
• No sovereign PubMed-reader lift until blind scientific-correctness scoring replaces packet-shape or fixture scoring.
• No governing morphology evidence until real PTB-XL+/extractor-backed fiducial artifacts are source-manifested, audited, KG-linked, and falsified.
• No Pathway 1 authority until the cardiac authority gate passes; Pathway 1 remains an R&D front-end surface.

Verification Status

Code	Check	Verdict
V_01	Canonical live-lab README spine applied with Pipeline Mechanics as Zone 02	PASS
V_02	Research/medical boundary preserved verbatim	PASS
V_03	Visibility unchanged; repo remains INTERNAL	PASS
V_04	CPU scaffold verification recorded at 768 tests before this README-only alignment	PASS
V_05	Authority-path blocker remains the governing front-door gate	BLOCKED
V_06	Authoritative H100 cutover remains gated until completion evidence exists	BLOCKED

Proof Anchors

Path	State
PRD.md	AUTHORITY DESIGN
docs/H100_COMPLETION_PLAN.md	H100 COMPLETION PLAN
docs/RUNPOD_READINESS.md	RUNPOD SCAFFOLD STATUS
docs/execution-report.md	EXECUTION HISTORY
runpod.config.yaml	CUTOVER CONFIG
src/zer0pa_biomolecular_explorer/	IMPLEMENTATION SURFACE

Repo Shape

Field	Value
Proof Anchors	6 display anchors
Portfolio	Bio-molecular research infrastructure
Authority Source	PRD.md plus current blocker register
Python Package	zer0pa-biomolecular-explorer
Pipeline Code	src/zer0pa_biomolecular_explorer/
Test Surface	tests/
Fixtures / Seeds	fixtures/; kg/; schemas/
Support Sections	H100 Plan; Blocker Register; Agent Handoff; Repository Layout; Historical Notes; Provenance

H100 Plan

The H100 work stream is led by an Opus Max chief-engineer agent with Sonnet-level subagents at minimum, escalating to Opus-level subagents where cross-layer scientific or systems reasoning is required. The lead agent has authority to make engineering decisions that move the pipeline toward more performant, more dataful, and more powerful outcomes, provided the boundary, interface contracts, falsifier gates, audit/KG integrity, and GitHub handoff discipline are preserved.

Budget on one H100 plus engineering effort:

Scope	GPU / engineering budget	Calendar wall clock	Meaning
Final authority hardening	0-2 GPU hours	4-8 engineering hours	Fix blockers below; mostly CPU/code work
H100 environment bring-up	1-3 GPU hours	2-4 hours	Container, drivers, repo, secrets, artifact paths, smoke run
Minimal governing cardiac cutover	60-120 GPU hours + 40-80 engineering hours	3-5 days continuous	Replace priority L1/L2/L5/reasoner stubs for seed + held-out wedge; rerun falsification
Enterprise cardiac pass	120-250 GPU hours + 80-160 engineering hours	5-10 days continuous	Broader MD/FEP sampling, repeated seeds, artifact deltas, source/audit/KG reconciliation
Full Pathway 1 plus cardiac authority	180-400 GPU hours + 120-240 engineering hours	10-21 days continuous	Add generation/screening/structure adapters after cardiac authority passes
Exhaustive channel/compound sweep	500+ GPU hours	Multi-week	Scale-out phase, not first cutover

Detailed plan: docs/H100_COMPLETION_PLAN.md.

Blocker Register

These are current authority blockers for the next engineering pass:

1. PubMed-reader lift must become a blind scientific-correctness gate, not a hard-coded or packet-shape proxy.
2. The user-facing cardiac CLI path must prove L6 governs packet generation and downstream promotion.
3. Packet assembly must consume validated L1-L5 evidence envelopes, not canned fixture panels.
4. The cardiac run must populate the governing FalsifierLedger, not only write audit rows.
5. runpod-precheck must fail invalid real-cutover states, including endpoint, GPU SKU, backend, and quarantined Pathway 1 cases.
6. Terminal L6 failures, unresolved REROUTE, late-layer FAIL state, orphan KG falsifiers, and HAS_FALSIFIER divergence must all block authority promotion.

Agent Handoff

Read in this order when taking over the lane:

1. PRD.md - governing product and architecture requirements.
2. docs/H100_COMPLETION_PLAN.md - H100 budget, WP0-WP5, agent topology, and acceptance gates.
3. HANDOFF-TO-OVERNIGHT-EXECUTOR.md and OVERNIGHT-EXECUTOR-STARTUP-PROMPT.md - execution role and startup behavior.
4. docs/RUNPOD_READINESS.md and docs/runpod-migration.md - stub-swap and Runpod readiness surface.
5. docs/DECISIONS.md and docs/CONVENTIONS.md - decision ledger and cross-layer conventions.
6. source-briefs/, briefing-pack/, and synthesis/ - inherited research context and cardiac wedge briefing.

Repository Layout

Path	Purpose
PRD.md	Medical orchestrator PRD for long-horizon execution
PATHWAY1_PRD.md	Pathway 1 R&D front-end PRD
src/zer0pa_biomolecular_explorer/	CPU-side contracts, layers, orchestration, audit, KG, falsifiers, packets, reasoner, runpod-sim
tests/	Unit, integration, plug-swap, falsification, packet, KG, and Runpod-sim tests
fixtures/	Compound, negative, morphology, PubMed-baseline, route, SBML, and Pathway 1 fixtures
kg/	Cardiac and Pathway 1 KG seed files
schemas/	JSON schemas for envelopes, fixtures, and reasoner tuples
docs/	H100 plan, Runpod readiness, decisions, conventions, and execution report
packets/	Historical generated packet artifacts; non-governing until aligned to current authority path
runtime/	Cloud-lab dry-run defaults with interlocks on

Historical Notes

The previous implementation record reported 768 passing tests, a runpod-sim cutover surface, L1-L6 CPU contracts, Pathway 1 R&D stubs, cardiac packet generation, KG checks, and falsification-wave coverage. Treat those as scaffold evidence, not as final authority. The 2026-05-02 blocker register above controls current readiness.

This repository was previously named Zer0pa/Health (historical). Renamed to Zer0pa/Bio-Molecular-Explorer on 2026-05-03 as a clean pre-public rename. No compatibility aliases remain for zer0pa-health or zer0pa_health.

Provenance

• Initial commit: 2026-04-29.
• Front-door alignment: 2026-05-02, using the orchestration-state lane-agent guidance dated 2026-05-02.
• Rename to Bio-Molecular Explorer: 2026-05-03.
• Upstream cardiac wedge context: Brain Phase 8 / Rosalind Bioelectric Translational Engine briefing pack.
• Medical orchestrator: produced PRD.md.
• Overnight executor: built the CPU scaffold, tests, Runpod-sim adapters, audit/KG/falsifier surface, and H100 completion artifacts.

What's Next

The scaffold is complete and renamed. The next gate is the governing cardiac authority path — replacing CPU-side stubs with real GPU evidence, tying every artifact to the falsifier/audit/KG chain, and running the full falsification wave.

Immediate: WP0 — Authority Hardening (CPU, no H100 required)

Six blockers must close before H100 time is spent. Wall clock: 4–8 engineering hours. Zero H100 cost.

1. L6 terminal FAIL/REROUTE must block packet export — currently passes with unresolved FAIL state.
2. runpod-precheck must return nonzero for invalid real-cutover states (missing GPU SKU, null endpoints, quarantined Pathway 1 claims).
3. Audit / falsifier ledger / KG reconciliation must fail on orphan falsifier nodes and missing HAS_FALSIFIER edges.
4. Morphology stubs must be source-manifested before entering governing evidence.
5. PubMed-reader scoring must be blind scientific correctness, not packet-shape completeness.
6. scripts/generate_cardiac_packets.py must use the governing assembly path or be marked non-governing.

H100 Provisioning: 120-Hour Budget

Provision one H100 SXM 80 GB pod on RunPod after WP0. RunPod is hourly — stop between sessions, do not leave idle.

Phase	H100 hours	Outcome
WP1 — environment bring-up	1–3 h	Repo cloned on H100, tests green, artifact root configured
WP2 — real L1 cardiac evidence	20–50 h	Structure/docking + OpenMM for KCNH2/SCN5A/KCNQ1/CACNA1C seed + held-out wedge
WP3 — real morphology evidence	0–6 h	PTB-XL+ extractor-backed fiducials replace locked morphology stubs
WP4 — L2/L5/reasoner replacement	4–14 h	DeepXDE dissolution, PKPD/QSP stubs, TxGemma 27B (after license check)
WP5 — full governing rerun	8–24 h	L6-governed cardiac pass, falsification wave, cutover report
Minimum governing budget	60–120 h	First real authority path: seed + held-out wedge fully governed

Budget 120 hours to absorb debugging. 60 hours is the floor if everything goes right.

Pathway 1

Non-governing until the cardiac authority gate passes. R&D front-end stubs remain in place.

HuggingFace Dataset Status

Three private HF datasets exist under legacy names (Architect-Prime/zer0pa-health-cardiac-packets, Architect-Prime/zer0pa-health-pathway1-packets, Architect-Prime/zer0pa-health-cardiac-v0). HTTP 401 confirmed — private, accessible under authenticated access. These names are retained pending a separate dataset migration decision; no migration is needed for the scaffold to operate. Code references in runs/cardiac_run.py and runs/pathway1_run.py point to these IDs.

Front Door Status

Repository is aligned to the Zer0pa Lab Front Door canonical spine as of 2026-05-03. Visibility: INTERNAL. This repo becomes a candidate for the Zer0pa website sync cycle when the cardiac authority gate clears and the H100 governing pass is verified.