Restructure public documentation for release

.agents/skills/evidence-status-discipline/SKILL.md

@@ -34,7 +34,7 @@ These are the specific overclaims this architecture invites. Use the right-hand
 | exactly-once execution | at-least-once delivery with idempotent effect handling; at most one accepted outcome per intended attempt | The worker protocol is at-least-once (ADR-002). Uniqueness constraints deduplicate *effects*; they do not make *execution* exactly-once. |
 | deterministic execution | deterministic state reconstruction from the ordered event stream | Invariant 6 promises replay reproduces logical state. Infrastructure timing may differ between executions. |
 | guaranteed no data loss | no accepted observation lost across the tested process-termination boundaries | A guarantee is universal; a test covers the boundaries it injected. Name them. |
-| production-ready | deployed and operated under the documented runbook, with the recovery evidence in `<artifact>` | `docs/ROADMAP.md` Slice 6: a cloud URL is not production proof. |
+| production-ready | deployed and operated under the documented runbook, with the recovery evidence in `<artifact>` | `docs/ROADMAP.md` open operational gaps: a cloud URL is not production proof. |
 | demonstrates / proves | implements, when there is no artifact | Reserve for `demonstrated`. |
 | fault-tolerant | fault-aware: every failure produces a durable outcome and a classified failure code | Tolerance implies continued correct service; awareness is what the design provides. |
 | calibrated uncertainty | model uncertainty from `<policy>`, uncalibrated | `AI_CONTRACT.md` §11 forbids calibrated-uncertainty claims without a defined calibration procedure and evaluation. |

.agents/skills/her-source-discipline/SKILL.md

@@ -6,7 +6,7 @@ paths: scripts/fetch_her.py, scripts/inspect_her.py, src/labbridge/infrastructur
 
 # HER source discipline
 
-Authority: `AI_CONTRACT.md` invariant 11 and §7; `docs/DATA_STRATEGY.md` §2; `docs/ROADMAP.md` Gate 0.
+Authority: `AI_CONTRACT.md` invariant 11 and §7; `docs/DATA_STRATEGY.md` §2, including the §2.3 source, licence, and schema gate.
 
 Pinned sources: Zenodo DOI `10.5281/zenodo.20439519` (dataset, authoritative for archive contents) and
 arXiv `2606.00779` (preprint, authoritative for method and interpretation).

.agents/skills/migration-and-schema-evolution/SKILL.md

@@ -7,7 +7,7 @@ paths: alembic/**, migrations/**, src/labbridge/infrastructure/postgres/**, src/
 # Migration and schema evolution
 
 Authority: `AI_CONTRACT.md` §6 and §9; `docs/SPEC.md` §4.1, §5; `docs/FAILURE_MATRIX.md` F-019, F-039,
-F-042; `docs/ROADMAP.md` Slice 3 and Slice 6.
+F-042; `AI_CONTRACT.md` §10 and `docs/ROADMAP.md`.
 
 Two distinct kinds of evolution live here. Do not conflate them.
 
@@ -58,7 +58,7 @@ migration upgrade tests and downgrade tests where safe and supported.
 
 ### Deployment
 
-`docs/ROADMAP.md` Slice 6 requires a migration exercised against production-like data and a documented
+`AI_CONTRACT.md` §10 requires a migration exercised against production-like data and a documented
 application rollback procedure, plus the interrupted-migration recovery path. Until that has been run
 and recorded, migration safety is `implemented`, never `demonstrated`.
 

.claude/agents/reliability-reviewer.md

@@ -19,8 +19,8 @@ skills:
 You are the reliability and failure-injection reviewer for `labbridge`.
 
 Your authorities are `docs/FAILURE_MATRIX.md`, `docs/SPEC.md` §15 (proof obligations PO-01 to PO-10),
-`AI_CONTRACT.md` §3 (invariants 2, 5, 6, 9) and §6, and `docs/ROADMAP.md` (which slice must satisfy
-which scenarios).
+`AI_CONTRACT.md` §3 (invariants 2, 5, 6, 9) and §6, and `docs/PROJECT_STATUS.md` (what each
+capability's evidence already covers).
 
 You are read-only. Never edit, never change Git state, never invoke another agent. Use `Bash` for
 inspection and for running existing tests without `--fix`, `-p no:cacheprovider` where available.

.claude/agents/scope-guard.md

@@ -42,7 +42,7 @@ Read the current file contents, not remembered status:
 1. `AI_CONTRACT.md` — invariants (§3), approved stack (§4), boundaries (§5), forbidden patterns (§11);
 2. `docs/SPEC.md` — V1 boundaries (§2), proof obligations (§15), module map (§16);
 3. `docs/ARCHITECTURE_DECISIONS.md` — accepted decisions and their consequences;
-4. `docs/ROADMAP.md` — gate and slice ordering, exit criteria, stop conditions, release blockers;
+4. `docs/PROJECT_STATUS.md` and `docs/ROADMAP.md` — current capability status, open gaps, deferred tracks;
 5. `docs/DATA_STRATEGY.md` and `docs/SIMULATOR_MODEL.md` — scientific and licence boundaries;
 6. `docs/FAILURE_MATRIX.md` — which failure semantics a slice must already satisfy.
 
@@ -56,8 +56,8 @@ a comment claiming a gate passed, or a branch name.
 Roadmap position is evidence-based, not asserted. Establish it from the repository:
 
 - which of Gate 0 and Slices 1–7 have their **deliverables** present on disk;
-- whether the exit criteria of the preceding slice are met by inspectable evidence, not by intent;
-- whether `docs/ROADMAP.md` stop conditions for the preceding slice are cleared.
+- whether the status a task assumes is met by inspectable evidence, not by intent;
+- whether the task would silently start a track `docs/ROADMAP.md` records as deferred.
 
 If you cannot establish the active slice from the repository, say so and treat timing as unresolved
 rather than assuming the slice the request implies.

.claude/agents/verification-auditor.md

@@ -39,8 +39,8 @@ Re-run it. Read the output. Count the failures.
 Restate the claim precisely, and classify it:
 
 - **a change is complete** — the per-change definition of done, `AI_CONTRACT.md` §10;
-- **a slice is complete** — every exit criterion of that slice in `docs/ROADMAP.md`, with no stop
-  condition tripped;
+- **a capability's status is justified** — the evidence `docs/PROJECT_STATUS.md` names for it exists
+  and verifies;
 - **a capability is `implemented`** — code exists and its relevant local automated tests pass;
 - **a capability is `demonstrated`** — a reproducible artifact, manifest, or operational experiment
   proves it;

.claude/commands/audit-repo.md

@@ -22,7 +22,7 @@ exists.
 ## 2. Documentation consistency
 
 - Do `AI_CONTRACT.md`, `docs/SPEC.md`, `docs/ARCHITECTURE_DECISIONS.md`, `docs/DATA_STRATEGY.md`,
-  `docs/SIMULATOR_MODEL.md`, `docs/FAILURE_MATRIX.md`, and `docs/ROADMAP.md` agree? Resolve conflicts by
+  `docs/SIMULATOR_MODEL.md`, `docs/FAILURE_MATRIX.md`, and `docs/PROJECT_STATUS.md` agree? Resolve conflicts by
   the precedence in `AI_CONTRACT.md`, section *"When documents conflict"*, and **report** the conflict —
   never pick the convenient reading silently.
 - Do `CLAUDE.md`, `AGENTS.md`, `.claude/`, and `.agents/` contradict `AI_CONTRACT.md` anywhere?
@@ -48,7 +48,7 @@ Establish the active slice from what exists on disk, not from what the documents
 Gate 0 and Slices 1–7: deliverables present, exit criteria met, stop conditions cleared. Report the
 first slice whose exit criteria are not met — that is where work belongs.
 
-Check the V1 release blockers in `docs/ROADMAP.md` and report which are currently open.
+Check the open gaps and deferred tracks in `docs/ROADMAP.md` and report which are still open.
 
 ## 5. Failure and proof coverage
 

.claude/commands/plan-slice.md

@@ -9,8 +9,8 @@ Plan the following task against the LabBridge roadmap. Do not write implementati
 
 ## 1. Establish the active roadmap slice from evidence
 
-Read `docs/ROADMAP.md`. Determine the active slice from what exists on disk — deliverables present, exit
-criteria met by inspectable evidence, stop conditions cleared — not from what the task assumes. If you
+Read `docs/PROJECT_STATUS.md` and `docs/ROADMAP.md`. Determine the current position from what exists on
+disk — deliverables present, evidence inspectable — not from what the task assumes. If you
 cannot establish it, say so and treat timing as unresolved.
 
 ## 2. Scope

.claude/commands/review-migration.md

@@ -44,5 +44,5 @@ Use the review block at the end of the `migration-and-schema-evolution` skill. C
 `SAFE`, `SAFE WITH SEQUENCING`, or `UNSAFE` — and an explicit *Untested claims* line.
 
 The migration file existing is not evidence the migration is safe. Until it has been exercised against
-production-like data with a documented rollback (`docs/ROADMAP.md` Slice 6), migration safety is
+production-like data with a documented rollback (`AI_CONTRACT.md` §10), migration safety is
 `implemented`, never `demonstrated`.

.claude/skills/evidence-status-discipline/SKILL.md

@@ -34,7 +34,7 @@ These are the specific overclaims this architecture invites. Use the right-hand
 | exactly-once execution | at-least-once delivery with idempotent effect handling; at most one accepted outcome per intended attempt | The worker protocol is at-least-once (ADR-002). Uniqueness constraints deduplicate *effects*; they do not make *execution* exactly-once. |
 | deterministic execution | deterministic state reconstruction from the ordered event stream | Invariant 6 promises replay reproduces logical state. Infrastructure timing may differ between executions. |
 | guaranteed no data loss | no accepted observation lost across the tested process-termination boundaries | A guarantee is universal; a test covers the boundaries it injected. Name them. |
-| production-ready | deployed and operated under the documented runbook, with the recovery evidence in `<artifact>` | `docs/ROADMAP.md` Slice 6: a cloud URL is not production proof. |
+| production-ready | deployed and operated under the documented runbook, with the recovery evidence in `<artifact>` | `docs/ROADMAP.md` open operational gaps: a cloud URL is not production proof. |
 | demonstrates / proves | implements, when there is no artifact | Reserve for `demonstrated`. |
 | fault-tolerant | fault-aware: every failure produces a durable outcome and a classified failure code | Tolerance implies continued correct service; awareness is what the design provides. |
 | calibrated uncertainty | model uncertainty from `<policy>`, uncalibrated | `AI_CONTRACT.md` §11 forbids calibrated-uncertainty claims without a defined calibration procedure and evaluation. |

.claude/skills/her-source-discipline/SKILL.md

@@ -6,7 +6,7 @@ paths: scripts/fetch_her.py, scripts/inspect_her.py, src/labbridge/infrastructur
 
 # HER source discipline
 
-Authority: `AI_CONTRACT.md` invariant 11 and §7; `docs/DATA_STRATEGY.md` §2; `docs/ROADMAP.md` Gate 0.
+Authority: `AI_CONTRACT.md` invariant 11 and §7; `docs/DATA_STRATEGY.md` §2, including the §2.3 source, licence, and schema gate.
 
 Pinned sources: Zenodo DOI `10.5281/zenodo.20439519` (dataset, authoritative for archive contents) and
 arXiv `2606.00779` (preprint, authoritative for method and interpretation).

.claude/skills/migration-and-schema-evolution/SKILL.md

@@ -7,7 +7,7 @@ paths: alembic/**, migrations/**, src/labbridge/infrastructure/postgres/**, src/
 # Migration and schema evolution
 
 Authority: `AI_CONTRACT.md` §6 and §9; `docs/SPEC.md` §4.1, §5; `docs/FAILURE_MATRIX.md` F-019, F-039,
-F-042; `docs/ROADMAP.md` Slice 3 and Slice 6.
+F-042; `AI_CONTRACT.md` §10 and `docs/ROADMAP.md`.
 
 Two distinct kinds of evolution live here. Do not conflate them.
 
@@ -58,7 +58,7 @@ migration upgrade tests and downgrade tests where safe and supported.
 
 ### Deployment
 
-`docs/ROADMAP.md` Slice 6 requires a migration exercised against production-like data and a documented
+`AI_CONTRACT.md` §10 requires a migration exercised against production-like data and a documented
 application rollback procedure, plus the interrupted-migration recovery path. Until that has been run
 and recorded, migration safety is `implemented`, never `demonstrated`.
 

.claude/tools/gates.py

@@ -216,43 +216,41 @@ def collect(root: Path) -> list[Gate]:
             "pytest-integration",
             "pytest -q -m integration",
             LIVE if has_integration and _has("pytest") else SCAFFOLDED,
-            "requires PostgreSQL and MinIO running; ROADMAP Slice 1 brings up the stack"
+            "requires `docker compose --profile infrastructure up -d`"
             if has_integration
-            else "no test carries @pytest.mark.integration yet (ROADMAP Slice 1)",
+            else "no test carries @pytest.mark.integration yet",
         ),
         Gate(
             "pytest-data",
             "pytest -q -m data",
             LIVE if has_data and _has("pytest") else SCAFFOLDED,
-            "requires the fetched HER archive on disk (ROADMAP Gate 0)"
+            "requires the fetched HER archive on disk (`labbridge fetch-her`)"
             if has_data
-            else "no test carries @pytest.mark.data yet (ROADMAP Gate 0)",
+            else "no test carries @pytest.mark.data yet",
         ),
         Gate(
             "migrations",
             "pytest -q -m integration -k migration",
             LIVE if migrations and has_migration_test else SCAFFOLDED,
             "migration directory and matching integration test present"
             if migrations and has_migration_test
-            else "no alembic/ or migrations/ directory yet (ROADMAP Slice 1)"
+            else "no alembic/ or migrations/ directory yet"
             if not migrations
-            else "no integration test matches -k migration (ROADMAP Slice 1)",
+            else "no integration test matches -k migration",
         ),
         Gate(
             "artifacts",
             "PYTHONPATH=src python -m labbridge.cli validate-artifacts",
             LIVE if artifacts_cmd else SCAFFOLDED,
             "command responds to --help; verifies the committed artifacts/ tree"
             if artifacts_cmd
-            else "`validate-artifacts` is not implemented yet (ROADMAP Slice 3)",
+            else "`validate-artifacts` is not implemented yet",
         ),
         Gate(
             "compose",
-            "docker compose up --build",
+            "docker compose --profile demo up --build",
             LIVE if compose and _has("docker") else SCAFFOLDED,
-            "compose file present"
-            if compose and _has("docker")
-            else "no compose file yet (ROADMAP Slice 1)",
+            "compose file present" if compose and _has("docker") else "no compose file yet",
         ),
         # --- release-level gates ------------------------------------------------------------------
         Gate(
@@ -263,15 +261,15 @@ def collect(root: Path) -> list[Gate]:
             LIVE if has_replay and _has("pytest") else SCAFFOLDED,
             "proves PO-01"
             if has_replay
-            else "no integration test named test_replay_determinism* yet (ROADMAP Slice 2)",
+            else "no integration test named test_replay_determinism* yet",
         ),
         Gate(
             "fault-campaign",
             "pytest -q -m slow -k fault_campaign",
             LIVE if has_fault_campaign and _has("pytest") else SCAFFOLDED,
             "process-boundary checkpoint proof; release command runs at least 100 seeded campaigns"
             if has_fault_campaign
-            else "no slow test named fault_campaign yet (ROADMAP Phase 7)",
+            else "no slow test named fault_campaign yet",
         ),
         Gate(
             "backup-restore",

AGENTS.md

@@ -9,7 +9,8 @@ invariants, architectural boundaries, proof requirements, and document precedenc
 relevant sections of:
 
 - [`docs/SPEC.md`](docs/SPEC.md) for required behaviour;
-- [`docs/ROADMAP.md`](docs/ROADMAP.md) for the active delivery slice;
+- [`docs/PROJECT_STATUS.md`](docs/PROJECT_STATUS.md) for the current status of every capability;
+- [`docs/ROADMAP.md`](docs/ROADMAP.md) for open gaps and deferred tracks;
 - [`docs/DATA_STRATEGY.md`](docs/DATA_STRATEGY.md) for source, metadata, and lineage rules;
 - [`docs/FAILURE_MATRIX.md`](docs/FAILURE_MATRIX.md) for required failure semantics;
 - [`docs/ARCHITECTURE_DECISIONS.md`](docs/ARCHITECTURE_DECISIONS.md) for accepted decisions;
@@ -21,7 +22,7 @@ Report contradictions instead of choosing the most convenient interpretation.
 ## Working rules
 
 - Inspect the existing implementation, migrations, tests, fixtures, and source data before editing.
-- Keep each change to the smallest coherent unit that satisfies a current roadmap exit criterion.
+- Keep each change to the smallest coherent unit that satisfies a falsifiable acceptance criterion.
 - Add or update tests at the layer required by the claim; do not weaken an invariant to make a test
   pass.
 - Never infer source columns, units, electrochemical conventions, or dataset semantics from memory.

AI_CONTRACT.md

@@ -12,7 +12,8 @@ The companion documents define:
 
 - `docs/SPEC.md`: what the system must do;
 - `docs/DATA_STRATEGY.md`: data sources and scientific boundaries;
-- `docs/ROADMAP.md`: dependency order and exit criteria;
+- `docs/PROJECT_STATUS.md`: the current status of every capability and the evidence behind it;
+- `docs/ROADMAP.md`: what remains open and what is deferred;
 - `docs/SIMULATOR_MODEL.md`: the biosensor simulator's scientific contract;
 - `docs/FAILURE_MATRIX.md`: failure scenarios the runtime must handle;
 - `docs/ARCHITECTURE_DECISIONS.md`: accepted architectural decisions.
@@ -38,7 +39,7 @@ An implementation agent MUST report a contradiction rather than silently choosin
 - Write code, comments, identifiers, schemas, commit messages, and technical documentation in English.
 - Be short and direct. Do not add promotional narration.
 - Ask a question only when an ambiguity materially changes scientific validity, architecture, data integrity, or public claims.
-- When a safe, simpler interpretation exists within the current roadmap slice, use it and state the assumption instead of blocking progress.
+- When a safe, simpler interpretation exists within the current task's scope, use it and state the assumption instead of blocking progress.
 - Present alternatives when they carry meaningfully different trade-offs.
 
 ### Behavioural guidelines
@@ -99,10 +100,10 @@ diary.
 
 Turn failure-prone electrochemical measurement records into validated, provenance-tracked scientific datasets, and execute experimental campaigns through a durable, auditable, resumable runtime that treats failures and recovery as explicit domain outcomes.
 
-LabBridge is demonstrated through two separate environments behind a shared runtime interface:
+LabBridge defines two separate environments behind a shared runtime interface:
 
-- an observed Au–Ir–Rh HER dataset executed in replay mode;
-- a synthetic, electrochemistry-informed biosensor environment executed in simulation mode.
+- an observed Au–Ir–Rh HER dataset executed in replay mode, which is implemented;
+- a synthetic, electrochemistry-informed biosensor environment executed in simulation mode, which is `deferred` and has no adapter.
 
 The two environments test the same infrastructure abstractions. They are not fidelities of one shared scientific candidate space.
 
@@ -319,7 +320,7 @@ The V1 stack is intentionally small:
 - pytest and pytest-asyncio;
 - mypy or pyright in strict mode;
 - ruff for formatting and linting;
-- one managed cloud deployment during portfolio hardening.
+- one managed cloud deployment.
 
 No heavyweight ML framework or GPU path belongs in V1.
 
@@ -512,7 +513,7 @@ The V1 release MUST satisfy the proof obligations in `docs/SPEC.md` and the mand
 
 A change is complete only when:
 
-1. it is in scope for the current roadmap slice;
+1. it is in scope for the requested task and does not silently start a deferred track;
 2. its typed interface exists;
 3. relevant failure modes are explicit;
 4. relevant unit and integration tests pass;
@@ -589,10 +590,10 @@ Implementation agents MUST NOT:
 
 ## 12. Agent execution protocol
 
-Before coding a roadmap slice, an implementation agent MUST:
+Before coding a task, an implementation agent MUST:
 
 1. read this contract and the relevant specification sections;
-2. confirm that the task belongs to the current roadmap slice;
+2. confirm that the task is in scope and is not a deferred track in `docs/ROADMAP.md`;
 3. identify the invariants touched;
 4. state the intended transaction, process, and failure boundaries;
 5. list the tests and artifacts that will prove the exit criterion;