Peptiter / DiscoveryLab
Workflow
Product workflow

Incremental strategy — gate-by-gate production readiness.

The incremental production strategy used to advance a candidate gate by gate without skipping evidence.

Incremental improvement strategy

Production readiness improves gate by gate, without pausing discovery.

The system runs today with conservative local triage plus mechanism verification, then replaces specific scoring paths with validated production packages, Lean proof hardening, curated pathway importers, and calibration data as they become available.

NowClaim-bounded

Conservative local triage + claim ledger

receptor relevance
pathway reachability
protected-node safety
CalibrateData-staged

Public + wet-lab calibration loop

ChEMBL, BindingDB, PDBbind, and returned LabSpace observations are staged as typed calibration manifests before any score is described as a calibrated probability.

StructureAWS-ready

Boltz/contact-map compute path

The AWS HealthOmics plan, Boltz-1 request schema, deterministic contact-map fixture, and molecular viewer contact application are wired before live GPU execution is provisioned.

LoopExport-ready

Wet-lab export loop

Portfolio selection now exposes objective tradeoffs, then Handoff generates a checksum-bound export package with sample sheet, 96-well plate map, assay protocol, controls, and result template.

Current increment

Scientific claim ledger

The Evidence workspace now shows a claim-boundary ledger spanning formal verification, physics-bearing reduced models, heuristic surrogates, calibrated predictions, wet-lab observations, and orchestration contracts.

Public calibration staging

ChEMBL, BindingDB, and PDBbind rows can be imported into typed manifests, deduplicated, split into train/validation/test cohorts, and previewed as receptor-fit calibration profiles.

Embedding task-head replacement plan

The 221 repository candidates now score through committed 2560-input task heads retrained with 1024-wide MLPs, precompiled Core ML exports for local Macs, scaffold splits, and raw label-space held-out metrics.

Boltz-1 complex prediction preview

Cloud run descriptors, HealthOmics workflow names, Boltz-1 request constraints, and deterministic contact-map evidence are visible before live AWS execution is provisioned.

Active-learning next batch

The portfolio has the active-learning design contract; final W1.4 batch-export wiring remains before this becomes the collaborator-facing next-batch package.

Wet-lab export package

The wet-lab prep directory now contains a 10-target candidate selection report, 10 candidate synthesis specs, 3 epigenome placeholders, 10 tiered assay protocols, and collaborator onboarding.

Seeded evolutionary discovery

Candidates now start from known bioactive peptide seeds, stay inside compatible receptor and motif families, and carry parent IDs, operator history, rationale, and next evidence requirements.

MOTS-c start candidate

DiscoveryLab now features MOTS-c as DL-SEED-MOTS-C: an evidence-backed mitochondrial-derived peptide seed for metabolic-resilience programs, with human efficacy gaps tracked explicitly.

AttractKit production path

DiscoveryLab can consume AttractKit and its Rust-backed scoring contract as the attract backend, while keeping the app-level triage flow auditable.

PEPFOLD Metal receptor-fit workflow

Peptide candidates can be locally folded and refined, then combined with receptor confidence and contact proxies for triage-level fit scoring.

Pathway mechanism verification

Candidates can now carry explicit pathway hypotheses built from source-qualified assertions, with reachability, blockade, safety, conservation, and perturbation-evidence claims checked before wet-lab handoff.

Pathway world-model governance

The local pathway registry tracks source release, license policy, evidence namespace, assertion layer, context overlay, knowledge status, confidence, executable-island links, and blocked claims.

Federated body-twin model cards

Reduced process twins now declare context of use, calibration state, update cadence, uncertainty method, validation endpoints, privacy boundary, allowed claims, and blocked clinical claims.

External Lean verifier CLI

Generated Lean 4 audit artifacts can be sent to peptiter-lean-verify for checksum validation, Lean execution or dry-run receipt generation, diagnostics capture, and CI integration.

PathwayLean package plan

Mechanism artifacts are now surfaced as a reusable Lean package plan with package name, version, generated files, checksum, and caveats for review.

Bedrock/Kendra tutor grounding contract

The peptide tutor has an evidence-grounding request contract for Bedrock Claude and Kendra citations, with local fallback and citation validation for no-evidence cases.