jepegit · Sep 26, 2026 · Sep 26, 2026 · Sep 26, 2026
diff --git a/.issueflows/05-epics/epic783_plan.md b/.issueflows/05-epics/epic783_plan.md
@@ -1,136 +1,193 @@
-# Epic #783: cellpy 2.2 (Stage 5) — live refresh first
-
-Anchor: https://github.com/jepegit/cellpy/issues/783
-Status: confirmed
-
-## Goal
-
-Ship Epic L from the Stage 5 tracker: a cell loaded from a growing raw file
-can `update()` so the result matches a full load of the finished file, and a
-batch can refresh or poll on top of that.
-
-Done for this planned slice when #779, #780, #164, #781, and #782 are closed
-and `uv run pytest -m essential` is green on `master`. #778 (the golden
-equality anchor) is already closed.
-
-## Constraints
-
-- Children already exist on GitHub. Specs below record those numbers as
-  `Published:` so `/iflow-epic publish` does not open duplicates.
-- Cellpy-only for L. Do not put `SupportsIncrementalLoad` in cellpycore.
-  Core `update_data` stays the merge primitive (overlap trimmed on
-  `source_datapoint_num`).
-- Design: [live-incremental](https://github.com/cellpy/architecture-plan/blob/main/cellpy2-live-incremental-design.md)
-  and the Stage 5 list on #783. Sequencing from the anchor: L6 first
-  (done), then L1 → L2 → L3 → L4 and L5.
-- v2-migration bugfixes stay `v2.1.x` patches, not this epic.
-- S1–S3 need a cellpy-core PR before a cellpy re-pin. They are not detailed
-  stages here.
-- None of the open L issues are yolo-fit (public protocol or loader/batch
-  behavior). Cycle must stop rather than weaken that.
-
-## Stage 1 — Golden equality anchor
-
-The correctness test lands before any loader protocol. Already closed.
-
-- Goal: incremental `update()` of a split file equals a full load.
-
-### Issue: L6 golden equality test — incremental update() equals full load
-
-- Spec: see #778. Load a truncated file, append the tail, assert `update()`
-  matches a full load of the whole file (summary equality).
-- Goal: the equality test is the anchor for later L work.
-- Model: deep
-- Depends on: none
-- yolo: no — defines the contract for the epic
-- Published: #778
-
-## Stage 2 — Cell update from a marker
-
-Protocol, then the cheap loaders, then `CellpyCell.update()`.
-
-- Goal: `c.update()` appends only new raw data when the loader supports it,
-  and falls back to a full reload when it does not.
-
-### Issue: SupportsIncrementalLoad protocol and marker types
-
-- Spec: see #779. In cellpy (not cellpycore), add a runtime-checkable
-  `SupportsIncrementalLoad` with `load_since(source, marker) -> IncrementalChunk`,
-  plus `LoadMarker` and `IncrementalChunk`. Loaders that cannot do a cheap
-  partial read omit the protocol so `update()` can fall back.
-- Goal: the protocol and types exist and are importable; no loader is
-  required to implement them yet.
-- Model: deep
-- Depends on: #778
-- yolo: no — public protocol
-- Published: #779
-
-### Issue: load_since for cheap-partial loaders
-
-- Spec: see #780. Implement `load_since` for `arbin_res`, `arbin_sql`,
-  `neware_txt`, and `maccor_txt`. Return native-schema rows since the marker
-  (tail overlap allowed) and the new marker. Other loaders stay full-read.
-- Goal: those four loaders return a chunk and a new marker; a loader without
-  the protocol is unchanged.
-- Model: deep
-- Depends on: #779
-- yolo: no — four loader behaviors
-- Published: #780
-
-### Issue: CellpyCell.update loads only new data
-
-- Spec: see #164. After `cellpy.get`, `c.update()` finds the last loaded
-  point, reads from there (protocol or full reload), and refreshes summaries
-  from the first new step (or the last incomplete old step). Several raw
-  files on one cell must merge, not clobber.
-- Goal: `c.update()` on a grown raw file matches a full reload of that file
-  for the #778 equality case.
-- Model: deep
-- Depends on: #780
-- yolo: no — public `CellpyCell` behavior
-- Published: #164
-
-## Stage 3 — Poll and batch refresh
-
-Both sit on `c.update()`. L5 does not need the L4 poll helper.
-
-- Goal: one cell can be polled, and a batch can refresh or poll without a
-  new core API.
-
-### Issue: live.py poll loop and retire processor.py
-
-- Spec: see #781. Replace the `utils/live.py` stub with
-  `poll(cell_or_path, interval, on_update=, stop_when_complete=True)` calling
-  `c.update()`. Delete `utils/processor.py` and fold its thread-pool
-  `cellpy.get` fan-out into `batch.runner`'s executor. If
-  `accessor_label.lstrip` is still in `batch_core`, switch it to
-  `removeprefix` (the #783 note says this may already be gone).
-- Goal: `poll` calls `update` on an interval; `processor.py` is gone; no
-  `lstrip` prefix bug remains.
-- Model: deep
-- Depends on: #164
-- yolo: no — deletes a module and changes the live loop
-- Published: #781
-
-### Issue: Batch live refresh
-
-- Spec: see #782. `b.update(live=True)` calls `c.update()` for journal
-  cells. `b.poll(interval=, until=)` repeats that and re-runs collectors /
-  report each tick. No new core types.
-- Goal: a batch refresh and a poll both pick up appended raw data through
-  `c.update()`.
-- Model: deep
-- Depends on: #164
-- yolo: no — public batch API
-- Published: #782
-
-## Later (unstaged)
-
-Not sequenced into issues here. The anchor already tracks them.
-
-- Epic S — #313, #312, #359 are core-first (core PR, PyPI, cellpy re-pin).
-  #888 (RPT filters) is cellpy-only.
-- Epic I — #270, #338, #306, #761, #827.
-- Epic R — #687, then #691.
-- Deferred to 2.3: versioned headers, #73 GITT/PITT, #889 Fredrik ICA.
-  Design-only after 2.2: #206.
+# Epic #783: cellpy 2.2 (Stage 5) — live refresh first
+
+Anchor: https://github.com/jepegit/cellpy/issues/783
+Status: confirmed
+
+## Goal
+
+Ship Epic L from the Stage 5 tracker: a cell loaded from a growing raw file
+can `update()` so the result matches a full load of the finished file, and a
+batch can refresh or poll on top of that.
+
+Done for this planned slice when #779, #780, #164, #781, and #782 are closed
+and `uv run pytest -m essential` is green on `master`. #778 (the golden
+equality anchor) is already closed. **Epic L shipped 2026-09-26** (#1016,
+#1100–#1104).
+
+Revision 2026-09-26: Stage 4 added — the **Epic M read path** (#784, BatBase
+metadata) is pulled forward because the maintainer wants to interact with
+BatBase as soon as possible. Push/write-back stays in 2.3 (Stage 6 plan in
+`cellpy-design-and-development/roadmap/stages/stage6-github-issues.md`).
+
+## Constraints
+
+- Children already exist on GitHub. Specs below record those numbers as
+  `Published:` so `/iflow-epic publish` does not open duplicates.
+- Cellpy-only for L. Do not put `SupportsIncrementalLoad` in cellpycore.
+  Core `update_data` stays the merge primitive (overlap trimmed on
+  `source_datapoint_num`).
+- Design: [live-incremental](https://github.com/cellpy/architecture-plan/blob/main/cellpy2-live-incremental-design.md)
+  and the Stage 5 list on #783. Sequencing from the anchor: L6 first
+  (done), then L1 → L2 → L3 → L4 and L5.
+- v2-migration bugfixes stay `v2.1.x` patches, not this epic.
+- S1–S3 need a cellpy-core PR before a cellpy re-pin. They are not detailed
+  stages here.
+- None of the open L issues are yolo-fit (public protocol or loader/batch
+  behavior). Cycle must stop rather than weaken that.
+
+## Stage 1 — Golden equality anchor
+
+The correctness test lands before any loader protocol. Already closed.
+
+- Goal: incremental `update()` of a split file equals a full load.
+
+### Issue: L6 golden equality test — incremental update() equals full load
+
+- Spec: see #778. Load a truncated file, append the tail, assert `update()`
+  matches a full load of the whole file (summary equality).
+- Goal: the equality test is the anchor for later L work.
+- Model: deep
+- Depends on: none
+- yolo: no — defines the contract for the epic
+- Published: #778
+
+## Stage 2 — Cell update from a marker
+
+Protocol, then the cheap loaders, then `CellpyCell.update()`.
+
+- Goal: `c.update()` appends only new raw data when the loader supports it,
+  and falls back to a full reload when it does not.
+
+### Issue: SupportsIncrementalLoad protocol and marker types
+
+- Spec: see #779. In cellpy (not cellpycore), add a runtime-checkable
+  `SupportsIncrementalLoad` with `load_since(source, marker) -> IncrementalChunk`,
+  plus `LoadMarker` and `IncrementalChunk`. Loaders that cannot do a cheap
+  partial read omit the protocol so `update()` can fall back.
+- Goal: the protocol and types exist and are importable; no loader is
+  required to implement them yet.
+- Model: deep
+- Depends on: #778
+- yolo: no — public protocol
+- Published: #779
+
+### Issue: load_since for cheap-partial loaders
+
+- Spec: see #780. Implement `load_since` for `arbin_res`, `arbin_sql`,
+  `neware_txt`, and `maccor_txt`. Return native-schema rows since the marker
+  (tail overlap allowed) and the new marker. Other loaders stay full-read.
+- Goal: those four loaders return a chunk and a new marker; a loader without
+  the protocol is unchanged.
+- Model: deep
+- Depends on: #779
+- yolo: no — four loader behaviors
+- Published: #780
+
+### Issue: CellpyCell.update loads only new data
+
+- Spec: see #164. After `cellpy.get`, `c.update()` finds the last loaded
+  point, reads from there (protocol or full reload), and refreshes summaries
+  from the first new step (or the last incomplete old step). Several raw
+  files on one cell must merge, not clobber.
+- Goal: `c.update()` on a grown raw file matches a full reload of that file
+  for the #778 equality case.
+- Model: deep
+- Depends on: #780
+- yolo: no — public `CellpyCell` behavior
+- Published: #164
+
+## Stage 3 — Poll and batch refresh
+
+Both sit on `c.update()`. L5 does not need the L4 poll helper.
+
+- Goal: one cell can be polled, and a batch can refresh or poll without a
+  new core API.
+
+### Issue: live.py poll loop and retire processor.py
+
+- Spec: see #781. Replace the `utils/live.py` stub with
+  `poll(cell_or_path, interval, on_update=, stop_when_complete=True)` calling
+  `c.update()`. Delete `utils/processor.py` and fold its thread-pool
+  `cellpy.get` fan-out into `batch.runner`'s executor. If
+  `accessor_label.lstrip` is still in `batch_core`, switch it to
+  `removeprefix` (the #783 note says this may already be gone).
+- Goal: `poll` calls `update` on an interval; `processor.py` is gone; no
+  `lstrip` prefix bug remains.
+- Model: deep
+- Depends on: #164
+- yolo: no — deletes a module and changes the live loop
+- Published: #781
+
+### Issue: Batch live refresh
+
+- Spec: see #782. `b.update(live=True)` calls `c.update()` for journal
+  cells. `b.poll(interval=, until=)` repeats that and re-runs collectors /
+  report each tick. No new core types.
+- Goal: a batch refresh and a poll both pick up appended raw data through
+  `c.update()`.
+- Model: deep
+- Depends on: #164
+- yolo: no — public batch API
+- Published: #782
+
+## Stage 4 — BatBase read path (Epic M, cross-repo)
+
+Prerequisites exist: BatBase OAuth2 client-credentials + self-service API
+clients (ife-bat/batbase#390, #391 closed); `cellpy-connectors` shared base
+(cellpy/cellpy-connectors#3, #4 closed).
+
+- Goal: `cellpy.get(...)` / `batch.load(...)` can take cell/test metadata from
+  BatBase through the `MetaResolver` JOURNAL/DB layer, read-only; an
+  unreachable source degrades to an empty layer.
+
+### Issue: M0 BatBaseClient and CLI passthrough
+
+- Spec: see cellpy/cellpy-connectors#1. Client-credentials token fetch with
+  in-memory expiry cache, one re-auth on 401, keyring/env credentials,
+  `cellpy connectors configure batbase`, `scope="read"` default,
+  `BatBaseAuthError`. Add `get(path, **params)` and
+  `cellpy connectors batbase get <endpoint>` so the API can be explored at once.
+- Goal: a configured user can call any BatBase read endpoint from Python or
+  the CLI without handling tokens.
+- Model: fast
+- Depends on: none
+- yolo: yes — well specified, on the shipped connector base, no cellpy change
+- Published: cellpy/cellpy-connectors#1 (external repo)
+- Status: shipped 2026-09-26 via cellpy/cellpy-connectors#8 (squash-merged).
+
+### Issue: M1 MetadataSource protocol and resolver hook
+
+- Spec: see #784 (read scope only). `MetadataSource` Protocol
+  (`fetch(key) -> MetaRecord | None`), `cellpy.metadata_sources` entry-point
+  registry, `MetaResolver` JOURNAL/DB-layer precedence with provenance naming
+  the source, null-object on failure, `CellMeta.uuid` + `external_id` /
+  `source_uri` back-link. No push.
+- Goal: cellpy has a source-agnostic slot for external metadata; the BatBase
+  adapter plugs in without importing a cellpy base class.
+- Model: deep
+- Depends on: none (M2 needs it)
+- yolo: no — public Protocol and resolver precedence
+- Published: #784
+
+### Issue: M2 BatBase MetadataSource adapter
+
+- Spec: see cellpy/cellpy-connectors#2. Map `/api/test-cellpy-tag/` (+ batch
+  and cell rows) to `MetaRecord`; decide the query key (cellpy tag vs cell
+  name); offline ⇒ empty layer; registered via the entry point.
+- Goal: a BatBase-tagged cell loads with BatBase mass / area / nominal
+  capacity / project without journal edits.
+- Model: deep
+- Depends on: M0, M1
+- yolo: no — mapping decisions against an unsettled API
+- Published: cellpy/cellpy-connectors#2 (external repo)
+
+## Later (unstaged)
+
+Not sequenced into issues here. The anchor already tracks them.
+
+- Epic S — #313, #312, #359 are core-first (core PR, PyPI, cellpy re-pin).
+  #888 (RPT filters) is cellpy-only.
+- Epic I — #270, #338, #306, #761, #827.
+- Epic R — #687, then #691.
+- Deferred to 2.3 (Stage 6 draft): versioned headers, #73 GITT/PITT, #889
+  Fredrik ICA, #206 phase 1, loader-shell retirement, Epic M push (M3).