PROJECT_STATUS.md

LabBridge — project status

Status: canonical statement of current capability and evidence Companion: ROADMAP.md lists what is still open and what is deferred

This document is the single place where a capability's status is decided. Where any other document implies a stronger status than the table below, this document is correct.

How to read a status

Every material capability carries exactly one status.

Status	Means	What it takes to move up
planned	Specified, not implemented	Code
implemented	Code exists and its relevant local automated tests pass	A reproducible artifact or an operational experiment
demonstrated	A committed, reproducible artifact or operational experiment proves it	—
deferred	Intentionally outside the current release	A product decision

Code existing is never sufficient for demonstrated. Tests passing is never sufficient for demonstrated. Only an inspectable artifact that can be named is.

Four further distinctions matter and are never collapsed:

• Completeness — the declared evidence a Package should carry is present. It is not reproducibility.
• Integrity — retained bytes still hash to their recorded checksums. It is not scientific validity.
• Scientific validity — whether a measurement means what it appears to mean. LabBridge does not assess this.
• Reproducibility — whether an experiment repeated would yield the same result. LabBridge does not assess this either.

Package verification proves integrity and reports completeness. It is not an experimental certification.

File-to-Package path

Capability	Status	Evidence and boundary
Opaque source intake, exact-byte retrieval, and integrity verification	demonstrated	artifacts/source-capture. Retention and checksum verification of exact bytes only; assigns no CSV semantics
Explicit generic CV CSV ingestion and closed normalisation lineage	demonstrated	artifacts/cv-ingestion. Parser, mapping, unit, structural, identity, and lineage behaviour for one committed profile; claims no electrochemical validity
Append-only Experiment assertions and deterministic release validation	implemented	artifacts/experiment-passport
JSON/HTML Experiment Passport and independently verified Experiment Package	implemented	Same artifact: initial and superseding Passports, independent CLI verification, and byte-level proof that one user assertion changed neither the source assertion nor the prior release
Bounded Gamry DTA Framework 7.07 CV ingestion and retained parser diagnostics	implemented	artifacts/gamry-dta-cv and its SUPPORT.md. One pinned textual variant; every other Framework version, technique, and table layout fails closed
Galvanostatic electrolysis electrical time series and Package schema 3	implemented	artifacts/galvanostatic-electrolysis. Time, current, and potential series only; chemical analysis reported as unavailable
EchemDB-aligned CV export	implemented	artifacts/echemdb-cv-exchange and its MAPPING.md. Validated against one pinned EchemDB metadata-schema commit and one pinned Frictionless version; not a general EchemDB compatibility claim
Single-user interactive CV Passport demo	implemented	artifacts/cv-passport-demo: browser trace, built-asset manifest, downloaded Package, independent CLI verification. Two human acceptance records remain outstanding — see ROADMAP.md

Every implemented capability above holds a committed candidate artifact. Each is held below demonstrated by outstanding clean-checkout reproduction, not by missing code. The demo is additionally held by the two human acceptance records in ROADMAP.md.

Campaign runtime

Capability	Status	Evidence and boundary
Process-boundary campaign fault injection	demonstrated	artifacts/fault-campaign
Deployment restore and the seeded synthetic-replay fault campaign	demonstrated	Same artifact
HER source acquisition, checksums, and landing provenance	implemented	Code and tests; the acquired archive is git-ignored
Licence and data-use gate, resolved only by a recorded decision	implemented	ADR-009; the licence parser always yields unresolved, so a record that stops declaring the licence reopens the gate
Archive inspection and versioned dataset inventory	implemented	Records actual paths, member tables, and column shapes rather than inferring a schema
Independently generated schema-compatible HER fixture	implemented	Copies no source values
HER replay adapter, with origin decided by the evidence on disk	implemented	A fixture_manifest.json means synthetic + replay; a provenance.json means observed + replay; a root holding both fails
Typed quantities, candidates, observations, outcomes, and derived metrics	implemented	Domain tests
Canonical serialisation and content-derived identity	implemented	Property tests
Campaign, work-item, and attempt state machines	implemented	Transition tests
PostgreSQL schema, constraints, and migrations	implemented	Migration tests
Object storage with read-back checksum verification and a pending/committed lifecycle	implemented	Integration tests against MinIO
Durable jobs with atomic claim, lease fencing, heartbeats, and lease recovery	implemented	Integration tests against PostgreSQL
Constraint-arbitrated idempotency for submission, enqueueing, and outcome acceptance	implemented	Each decided by a database constraint reached through a conflict-safe insert, tested under real concurrency
Worker-startup and CLI reconciliation, including non-deleting object classification	implemented	Unexplained bytes are classified, never deleted
Retry scheduling	implemented	Durable and bounded; on-demand rather than a running scheduler
Typed, version-checked event append with aggregate and campaign ordering	implemented	Unknown types and unsupported schema versions fail explicitly
Explicit legacy/incomplete event-stream boundary and validated replay input	implemented	Pre-contract streams are excluded rather than guessed
Deterministic state reconstruction from the event log	implemented	Replay reproduces logical state; infrastructure timing may differ
Append-only budget ledger, written in the outcome transaction	implemented	Attempt and actual cost retained even for suppressed duplicates
Budget reservation and hard stopping rules	implemented	Transactional under concurrent workers
Evidence bundles, local bundle checks, and full stored-object verification	implemented	--mode bundle-only reports partial; --mode full reports complete only when every referenced object matches
Campaign submission API with idempotency keys	implemented	API tests
Campaign control endpoints, observability, and operator runbook	implemented	OPERATOR_RUNBOOK.md
Enzymatic-biosensor simulator	deferred	Contract retained in SIMULATOR_MODEL.md; no adapter exists
Model-based selection policy beyond a seeded random baseline	deferred	—

What the campaign reliability artifact measured

The committed run executed 100 seeded campaigns covering all six declared process-termination boundaries: after lease acquisition, after adapter return before upload, during object upload, after upload before the outcome transaction, after commit before acknowledgement, and during evidence export.

It recorded zero lost accepted observations, zero unintended duplicate acceptances, zero hard-budget overspends, zero projection mismatches, and zero Package-verification failures. It also recorded 51 lease_lost failures and 100 suppressed redeliveries. The deployment restore reverified 100 stored objects, replayed 100 campaigns, and fully verified 100 restored campaign Packages.

Campaign cancellation with a leased job is exercised across the same kind of process boundary: an already-leased job may finish, while new claims and retries are rejected and any received bytes remain evidence (FAILURE_MATRIX.md F-034).

These are software-behaviour measurements on generated synthetic bytes in replay mode. They are not scientific results, not observed-data results, and not live-instrument results. Observed HER replay and live execution were not run.

Known limitations

• No scientific number in this repository comes from a released physical measurement run.
• The end-to-end campaign path runs on the generated fixture. labbridge demo her completes and its bundle verifies, but every record it produces is synthetic.
• At-least-once delivery with idempotent effect handling is the worker protocol. Duplicate submission, enqueueing, and acceptance are each suppressed by a database constraint, and that suppression is tested under real concurrency. An accepted work item can still be re-executed by a redelivery, costing an adapter call and an object upload before the acceptance claim refuses it. Suppression is proven; avoiding the wasted execution is not attempted.
• A suppressed duplicate retains its received bytes as a non-accepted observation under its own attempt, so identical and divergent reads stay distinguishable.
• Experiment Passport validation reports declared evidence completeness. A released Package may retain warnings and unknowns; release asserts neither scientific validity, data quality, nor experimental reproducibility.
• Gamry support covers one pinned Framework 7.07 CV layout. It is not general Gamry support. The parser retains source-field locations and diagnostics without inferring a reference scale, a potential correction, or a working-electrode area, and it does not normalise current to current density.
• The EchemDB export is validated against pinned external versions. It is not general EchemDB compatibility, and it is not EchemDB submission.
• Galvanostatic electrolysis covers generic CSV electrical time series. The Passport and Package report no conversion, selectivity, yield, product assignment, or Faradaic efficiency, and the capability excludes instrument control and chromatography ingestion.
• The acquired HER archives and the generated fixture are git-ignored. A clean checkout has no data until labbridge fetch-her or labbridge build-her-fixture produces it.
• The integration suite requires PostgreSQL and MinIO and skips loudly when they are absent rather than passing vacuously.
• Nothing here has been deployed or operated outside local Docker Compose.

Scientific boundaries

• Observed and synthetic data are never conflated. Every observation and derived artifact carries data_origin, execution_mode, environment_id, and provenance linking it to source files or to a seeded simulator configuration.
• Synthetic output may be exported and visualised, but every human-readable and machine-readable representation identifies it as synthetic. It is never described as measured, experimental, or observed.
• User-supplied metadata stays identified as user-supplied. Unknown metadata stays unknown. An inferred value carries its method, version, evidence, and confidence representation, and cannot be relabelled as source_file or user_supplied.
• Structural validation is not scientific validation. LabBridge checks that a value resolves to a source column, a declared transformation, and a unit. It does not check that the value is physically correct.
• The demo's RHE value is an operator declaration. LabBridge does not infer it, validate it as physically correct, or convert potential values to that reference scale.

Full definitions are in DATA_STRATEGY.md.