Repository navigation
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub. 3 Skipped Deployments
|
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
There was a problem hiding this comment.
Assessment
The implementation approach is the right one and I would not steer toward a different architecture. Replacing the kind coordinate with instance id at scope-keys.ts, threading that through execution context + persisted reads, and consuming FIX-1322's owner resolution for later reads is the minimal honest fix. Per-instance databases, runtime ownership inference, and permanent dual-read are correctly rejected.
Where this spec went further than the six-file change warrants is prose, not mechanism: §7's adapter-native migration runbook, §10's triple-adapter rehearsal matrix, and §11's nine documentation surfaces duplicate the same operational story several times. That bloat risks implementation scope creep (operators expect what the spec spells out) without changing the core design.
Simpler framing
| Layer | Keep as scoped | Consider trimming |
|---|---|---|
| Runtime | kind → id in scope-key helpers; FIX-1322 owner seam; memory nested-map bucket fix |
— |
| Migration contract | Offline attributable inventory; migration-required / migration-conflict; singleton id === kind byte-identical keys |
Per-adapter filesystem/SQL walkthroughs in the spec body |
| Verification | Real-path A/B isolation goal; extend existing flow-isolation / resource-flow-isolation suites |
Permanent CI rehearsal on filesystem and SQLite and Postgres |
| Docs | One user-facing isolation update + one persistence/migration home | Nine overlapping extend-in-place items |
Gaps to close before implementation
- Route coverage vs "6 production files" —
resource-routes.ts(7×) anddebug-snapshot.ts(3×) still doregistry.get(session.flowKind)today. §7 wires FIX-1322 owner resolution explicitly intostate-routes.tsonly. Either add these to the owned delta or state that FIX-1322 delivers a shared session→resolved-instance helper all route modules consume. - Colon ambiguity contradiction — §7 says composite-key ambiguity is "not repaired," but the memory bucket fix and §9 edge row explicitly repair the
reviewer-a/reviewer-a:bcollision class. Pick one framing (ban:in instance ids at FIX-1321, or exact buckets + allow colons) and align §6 sizing with the memory restructure. - FIX-1322 consumed interface — sibling specs aren't in-repo yet; a short stub (resolver shape, session
flowIdfield, refusal codes, "resolve once per request") would prevent seam drift across three issues. - Duplicated
IsolationFlowcoercion —toIsolationFlowininternal.tsand inlineisoFlowinstate-routes.tswill both needid; consolidate whenIsolationFlowchanges rather than spreading a third copy through execution context.
Migration ownership (FIX-1322 vs here)
The split is sensible: FIX-1323 owns storage-cell inventory and adapter-native key moves; FIX-1322 owns session/request owner stamping. §7 step 5 and §11 item 2 should single-source the combined operator runbook in docs (one page, anchored sections) so two specs don't each carry a full procedure.
Performance note (underspecified, not over-engineered)
Instance isolation does not increase resourceScopeIds beyond today's two-bucket ceiling. The memory nested-map change is a hot-path win, not bloat. Worth adding one invariant: FIX-1322 owner resolution happens once per request and is reused for all scope/resource reads in that handler.
Verdict
Direction: approve. The key-swap-at-convergence approach is cleaner than per-instance stores, write-only patches, or a migration service.
Spec: trim before implementation sign-off — collapse migration narrative to one §7 section + refusal codes, reduce §11 to ~4 surfaces, right-size §10 SQL rehearsal, resolve the gaps above. None of these require rethinking the architecture.
Happy to re-review after a slimming pass.
Sent by Cursor Automation: Review For Best Approach
|
|
||
| **Non-goals:** registry rules, a second ownership record, online migration service, session-resource redesign, new address syntax, chat optimization/removal, Devtool switcher, Workforce mint, message boards, and generalized key-encoding reform. | ||
|
|
||
| **Size:** 6 production files · 8 behavior groups · 9 documentation surfaces · 1 implementation PR. |
There was a problem hiding this comment.
Sizing understates scope. "6 production files" covers key derivation, but §7's "Every read and writer" goal also requires fixing registry.get(session.flowKind) in resource-routes.ts (7 call sites) and debug-snapshot.ts (3 call sites) — unless FIX-1322 lands a shared session→resolved-instance helper all route modules import. Please either expand the owned delta or explicitly defer those files to FIX-1322 with a named helper.
There was a problem hiding this comment.
Drop — the current source does have those kind-first callers, but the six files are FIX-1323’s owned isolation delta, not the whole assembled floor. §7’s shared seam, §8 opening/step 3/final paragraph and §12 require FIX-1322’s durable flowId owner resolution before completion, including resource/debug later reads. FIX-1322 owns replacing upstream session→instance selection in resource-routes.ts and debug-snapshot.ts; this issue consumes the resolved instance in internal.ts. A named helper/signature remains an implementation choice, not another owner resolver here. Neither issue’s spec is individually approved and this reply does not claim the seam has landed.
| | Isolated user/org resource state/content | `${identityId}:${flow.id}` plus existing canonical resource key | | ||
| | Session resource | existing tenant-qualified session key; existing lineage address for `sharedToLineage` | | ||
|
|
||
| These are internal storage coordinates, not a public `(kind,id)` address. Keep the current separator format, identity fields and tenant policy. User/org remain shared across tenants by design. No new id alphabet is imposed; existing composite-key ambiguity is not repaired by this issue. Preflight must detect source/destination aliasing rather than parse ownership out of a colon-delimited string. |
There was a problem hiding this comment.
Contradiction with the memory fix (§7 L90–91, §9 L146). This line says composite-key ambiguity is "not repaired," but the nested-map bucket change explicitly repairs the reviewer-a / reviewer-a:b prefix-collision class. Recommend picking one:
- A) Document "instance ids must not contain
:" at FIX-1321 and treat the memory restructure as defense-in-depth, or - B) Keep exact buckets and delete the "not repaired" claim.
Either way, update the §6 size line to acknowledge the memory adapter work is more than string substitution.
There was a problem hiding this comment.
Drop — these are two different boundaries. §7’s table retains identityId + ":" + instanceId as scopeId: for example (identity "u:x", instance "y") and (identity "u", instance "x:y") still spell the same scopeId. Exact memory buckets do not repair that pre-existing identity/instance ambiguity. Separately, memory currently concatenates scopeType/scopeId/resourceKey and prefix-scans (content-store.ts:13–18,42–48,54–58); distinct scopeIds "u:reviewer-a" and "u:reviewer-a:b" can therefore alias a bucket prefix or resource tuple. §7/§9 repair that second boundary only. No new colon ban or general encoding contract is selected. The six-file count already includes both memory adapters (§8 step 4), and §3 explicitly says this is more than string substitution. This is source-based reasoning, not executed proof; migration preflight is not a promise of universal runtime detection or protection for arbitrary fresh identity tuples.
| #### Every read and writer | ||
|
|
||
| - **`context/createExecutionContext.ts`:** scope-record initial/reload keys; eager/config-based resource resolution; lazy/per-storage-key resolution; all persist callbacks and grouped loads consume those resolved ids. Both direct `flow.kind` calls to `resolveResourceScopeId` must move. Keep the session/lineage branch unchanged. | ||
| - **`routes/state-routes.ts`:** FIX-1322 resolves the session's persisted `flowId` to its owner before any state/resource read. Preserve `id` in the local isolation shape; use it for scope records and both resource stores. No `registry.get(session.flowKind)` reconstruction. |
There was a problem hiding this comment.
state-routes.ts is named, but persisted resource reads also enter through resource-routes.ts → getPersistedData → internal.ts. The isolation fix in internal.ts only helps once the correct FlowInstance (with id) is selected upstream. Worth a one-liner here: "all session-scoped route handlers resolve owner before getPersistedData."
There was a problem hiding this comment.
Drop — agreed on the required order; it is already the §7 owner→FlowInstance.id→scope-key path and §8/§12 completion condition. All session-scoped resource/debug handlers must consume FIX-1322’s resolved persisted owner before selecting configuration or calling getPersistedData; passing a first-by-kind FlowInstance into internal.ts would not satisfy this spec. The upstream ownership cutover stays FIX-1322’s work rather than adding a second resolver or expanding this isolation delta.
|
|
||
| An inventory row is data, not a proposed public API: `user / identity u / legacy kind reviewer / resource notes/a / destination reviewer-a / ownership evidence <operator reference>`. Include every state/content/tombstone representation belonging to that logical cell. A scope-state row maps its whole blob separately. | ||
|
|
||
| #### Compatibility and migration contract |
There was a problem hiding this comment.
Migration runbook density. Steps 1–5 plus the adapter table (L100–106) are valuable operator content but heavy for an implementation spec — they appear again in §10 L166–167 and §11 item 2. Suggest: keep decisions + refusal codes + inventory row format here; move per-adapter filesystem/SQL recipes to the persistence doc (§11 item 2) as the canonical home and reference it once.
There was a problem hiding this comment.
Drop — §11 item 2 already chooses persistence/overview.md as the single combined operational home, with other pages linking rather than copying the procedure. §7’s adapter table is the feasibility/limits evidence for attributable offline migration; §10 is a planned rehearsal, not a duplicate operator runbook. Moving the evidence out now would leave this never-merged spec dependent on an implementation doc that does not yet exist. No new migration service, runtime scanner, or broader deliverable is being added.
|
|
||
| **Eight behavior groups:** (1) private user/org state across copies; (2) shared/singleton compatibility; (3) per-resource override precedence; (4) static/aliased/collection state and content on eager/lazy paths, including colon/prefix instance ids and exact-bucket deletion/purge; (5) fresh-context state/resource/client/debug reads; (6) session/lineage addressing unchanged with owner refusal; (7) attributable migration survives restart including tombstones/versions/content; (8) ambiguous/occupied/alias migration refuses without publishing partial results. Parameterize adapter behavior where existing conformance fixtures support it, rather than duplicate plumbing assertions. | ||
|
|
||
| **Migration rehearsal, planned:** create old-layout fixtures with explicit provenance and distinct singleton/shared/collection cells in filesystem, SQLite and Postgres; convert using §7's adapter-native recipe and verify raw inventory conservation plus the same post-restart router reads. Include a tombstone and a content-only resource so a getAll-only migration would fail. An unresolved-owner fixture and an unrelated occupied target must stop before publication. Custom-store support is conditional on complete native export/import, not assumed from interface compatibility. |
There was a problem hiding this comment.
Rehearsal scope may be more than the proof needs. SQLite and Postgres share the same logical PK layout (§7 L104–105); permanent CI coverage on filesystem + one SQL adapter likely suffices, with Postgres reserved for release/manual unless JSONB/advisory-lock behavior is under test. Same conservation rules, less CI duplication.
There was a problem hiding this comment.
Note for the implementer — recorded verbatim in §13 at 3e22eab and mirrored to the existing FIX-1323 Linear document. §10’s filesystem/SQLite/Postgres rehearsal remains planned; it does not require three permanent CI copies. Which adapter fixtures earn ongoing CI coverage is left to implementation, without dropping the promised rehearsal.
|
|
||
| ### 11. Documentation plan | ||
|
|
||
| Docs are required: observable isolation and exported helper inputs change. Extend existing pages; no new page, sidebar/category, guide or blog. Existing sidebar homes are Core → Engine → Advanced → flow isolation; Core → Engine → persistence overview; Core → Fundamentals → state/scopes and flow configuration; Core → Resources → storage. Readers should know flows, scopes and resources; introduce instance as a particular configured copy, not as another scope. |
There was a problem hiding this comment.
Nine doc surfaces with heavy overlap. Items 1, 4, 5, 7 all teach "private state follows instance"; items 2, 7, 8, 9 all touch key matrices/migration. Consider collapsing to ~4: (1) user-facing isolation (flow-isolation.md + brief crosslinks), (2) persistence/migration (persistence/overview.md — single operational home), (3) resources override (storage.md), (4) architecture + engine README. Item 6 (auth) could be one sentence in item 1 on namespacing ≠ authorization.
There was a problem hiding this comment.
Drop — §11 already uses isolation and persistence as the main teaching/operational homes; items 4–6 are brief corrections/crosslinks and items 7–9 repair existing architecture/helper contracts. The nine surfaces are not nine runbooks. In particular, leaving the old kind/private claim in authentication while correcting another page would leave contradictory docs. Keep the existing extend-in-place/crosslink plan rather than remove affected surfaces; it adds no new page.
| - [Later state reads](https://github.com/fixpoint-labs/flow-state-dev/blob/cc4bd1e41/packages/engine/src/routes/state-routes.ts#L57-L157), [resource flow coercion](https://github.com/fixpoint-labs/flow-state-dev/blob/cc4bd1e41/packages/engine/src/resources/internal.ts#L254-L271): kind-only read paths requiring the same cutover. | ||
| - [Architecture](https://github.com/fixpoint-labs/flow-state-dev/blob/cc4bd1e41/docs/architecture/state-and-scopes.md#L508-L572) currently documents kind-qualified keys; this spec intentionally changes that contract, not merely a hidden implementation detail. | ||
|
|
||
| **Industry alternatives:** [Cloudflare Durable Objects identity](https://developers.cloudflare.com/durable-objects/api/namespace/) distinguishes a class from a stateful instance, but its mint-on-demand namespace API is not needed here. [AWS silo versus pooled storage taxonomy](https://docs.aws.amazon.com/whitepapers/latest/multi-tenant-saas-storage-strategies/multitenancy-on-dynamodb.html) supports keeping existing key partitioning rather than per-instance databases (the whitepaper is marked historical, not operational tuning advice). [Prisma's explicit data migration](https://www.prisma.io/docs/guides/database/data-migration) supports attributable migrate-then-contract; it cannot supply ownership absent from legacy data. No library or new primitive is needed. |
There was a problem hiding this comment.
Low-signal section. Cloudflare Durable Objects / AWS silo taxonomy / Prisma data migration restate decisions already made in §3. Could fold into §3 as one sentence ("attributable offline migration when legacy keys lack owner metadata") or drop — implementer behavior doesn't change.
There was a problem hiding this comment.
Drop — this is an editorial preference without a directional defect. §12 labels these as research alternatives, explicitly rejects adding their APIs/libraries, and links the sources; the build direction remains §3/§6. No design rewrite or implementer requirement is needed.
jhoffner
left a comment
There was a problem hiding this comment.
second-look: FIX-1323 spec (PR #1632, spec/FIX-1323.md)
Verdict: Appropriately scoped — the approach and the size of the migration write-up both earn their keep; no restraint findings survive the calibration gate.
Scope: spec doc is 205 lines (actual, single file). Forward implementation estimate per §6: 6 production files · 8 behavior groups · 9 doc surfaces · 1 implementation PR.
| # | Finding | Lens | Recommendation | Est. Δlines | Sev | Conf |
|---|---|---|---|---|---|---|
| — | (none passed the gate) | — | — | — | — | — |
Architecture-improvement angle: N/A — this PR touches only spec/FIX-1323.md, zero production code, so there's nothing here to refactor.
Why nothing made the cut
The two things most likely to read as bloat on a skim both check out against the real code:
- The per-adapter migration table (Memory/Filesystem/SQLite/Postgres/custom) is ~40% of the doc. It's tempting to call that oversized for a direction document. But it's feasibility evidence for the generic 5-step offline procedure in §7, not restated elsewhere — the doc plan (item 9) explicitly says "link to the site migration procedure rather than copy it," so this isn't duplicated prose. And it isn't a hypothetical seam:
packages/store-sqliteandpackages/store-postgresare real, shipped adapter packages, plus the built-in memory/filesystem stores, plus a documentedadd-store-adapterextension point for custom ones — four+ real targets, not a speculative "what if someone builds another adapter" case. - Two named refusal outcomes (
migration-requiredvsmigration-conflict) look like unnecessary API surface at first read, but the spec is explicit these are operator-facing procedure labels in prose, not new runtime error types or code branches ("not new runtime errors inferred from cell presence"). No Δlines to recover here.
Considered & deemed justified
- Bundling the memory-store exact-bucket-indexing fix into this issue (
stores/memory/content-store.ts,resource-state-store.ts) rather than splitting it into its own issue — kept because thereviewer-a/reviewer-a:bprefix collision only becomes exploitable once instance ids are free-form strings, which is exactly what this issue introduces at the same convergence point (resolveResourceScopeId). Not scope creep. - Decision 1 (reuse the existing key-partition scheme; reject a new scope or per-instance database) — already the minimal-surface choice. The spec explicitly rejects the heavier alternative (a new isolation mode), so there's nothing to trim here — this is the 80/20 version already.
- Nine documentation surfaces — spot-checked against current content (e.g.
docs/architecture/state-and-scopes.md's "Cross-Flow State" section still documents the kind-only${id}:${flowKind}key today). All nine currently state the contract this spec changes and would be actively wrong post-change, not optional re-explanation. - The "custom or mixed registry" adapter row — kept because
add-store-adapteris a real, documented FSD extension point (own skill + package pattern), not a hypothetical.
No findings here means this pass has nothing to add beyond what the PR's own "Parts worth reviewing closely" (§7 convergence/memory tuples, §7 migration) already flags — second-look confirms those are the right things to be worried about and that the size spent on them is proportionate.
Generated by Claude Code
|
FSD Architect — coherence clear on §6 ratified (both)
§7 close-reviewConvergence: One path — FIX-1322 owner → Memory tuples: Exact Migration recipes: Adapter-native offline cutover (quiesce, inventory, refuse ambiguity, preserve tombstones/versions/content, no live Invent / forkNo Team/MessageBoard/new scope. No durable public — FSD Architect |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 4693570408
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| | Cell | Key after cutover | | ||
| |---|---| | ||
| | Shared user/org scope state | bare identity id | | ||
| | Isolated user/org scope state | `${identityId}:${flow.id}` | |
There was a problem hiding this comment.
Include the kind in isolated storage coordinates
The proposed ${identityId}:${flow.id} key is not unique to a flow instance because the registry identifies instances by (kind, id) and permits the same ID under different kinds (packages/engine/src/registry/flow-registry.ts:80-112). For example, reviewer({ id: "main" }) and writer({ id: "main" }) for the same user both resolve to u:main, allowing supposedly isolated state and resources to overwrite or expose each other; migration preflight also cannot protect fresh stores. Either make the coordinate injective over (kind, id) while preserving legacy singleton keys, or require and enforce globally unique instance IDs.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Drop — the finding correctly describes today’s registry, but §12 already makes FIX-1321 registry/cardinality a hard prerequisite and §8 says it must merge first. The approved floor requires globally unique instance IDs; this spec does not permit concurrent reviewer/main and writer/main registrations. Thus isolation consumes a globally unique resolved id, not today’s permissive (kind,id) identity. Keep that prerequisite; adding kind to persisted/public coordinates would introduce a competing contract. No claim that the prerequisite implementation has landed, and migration preflight is not the answer for fresh stores.
|
|
||
| - **`context/createExecutionContext.ts`:** scope-record initial/reload keys; eager/config-based resource resolution; lazy/per-storage-key resolution; all persist callbacks and grouped loads consume those resolved ids. Both direct `flow.kind` calls to `resolveResourceScopeId` must move. Keep the session/lineage branch unchanged. | ||
| - **`routes/state-routes.ts`:** FIX-1322 resolves the session's persisted `flowId` to its owner before any state/resource read. Preserve `id` in the local isolation shape; use it for scope records and both resource stores. No `registry.get(session.flowKind)` reconstruction. | ||
| - **`resources/internal.ts`:** persisted resource lookups consume an instance-bearing flow. Narrow the persistence-facing `ResourceFlowLike`/`toIsolationFlow` path so it cannot lose `id`; metadata-only functions may retain the minimal shape they actually need. Resource list/get/content/debug reads converge through this module; they must not pick the first kind copy. MCP shares metadata helpers but currently refuses persisted reads without a session context ([MCP boundary](https://github.com/fixpoint-labs/flow-state-dev/blob/cc4bd1e41/packages/mcp/src/createMcpTransportAdapter.ts#L537-L546)); enabling those reads is not in this issue. |
There was a problem hiding this comment.
Resolve the owner before every resource route
Resource and debug routes do not actually converge on this module before choosing a flow: resource-routes.ts:88-113 and debug-snapshot.ts:258-288 first call registry.get(session.flowKind), which deliberately returns the kind's default or first instance (flow-registry.ts:127-140), and then pass that potentially wrong instance into getPersistedData. Because the six-file implementation boundary excludes those callers, later reads for a session owned by a non-default same-kind instance can still use another instance's resource configuration and storage bucket. Include these callers in the owner-resolution cutover or make the internal persistence seam resolve the session's stamped owner itself.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Drop — the source observation is correct: resource-routes.ts selects registry.get(session.flowKind) before configuration and getPersistedData. The required fix is already FIX-1322’s upstream owner-resolution cutover, consumed here (§7 shared seam, §8 opening/step 3/final paragraph, §12). Resource and debug handlers must resolve the persisted flowId owner before configuration selection/read; merely preserving id in internal.ts is insufficient and does not meet the assembled goal. The six-file count is this issue’s owned delta, not permission to ship first-kind callers. No second resolver/owner record is introduced here; FIX-1323 cannot be declared complete without that seam.
|
Handled one bounded first external spec-feedback batch against
Accounting: 1 review round spent; no sustained new spec-level finding; no POC settlement trigger. Both Part I Decisions and the PR description remain unchanged. FIX-1321’s hard dependency is unchanged. Explicit attributable offline migration, refusal on unresolved ownership, and the maintenance window remain required; orphan keys do not reveal missing owners. Gate: awaiting Jake’s approval of this individual spec’s direction and §6 Decisions. This is not the cross-spec pass. No implementation, tests/builds/formatters, executable validation, approval labels/reviews, merge, or close. §10’s goal and migration rehearsal remain planned/unrun. |
|
FSD Architect — fold after Jake’s individual approval of What to change on
|
|
spec approved |
|
FSD Architect — individual Spec Approved recorded. Jake approved Floor status: FIX-1321 ( Open fence (do not invent / do not follow stale sequencing): approved head §8 and §12 still say FIX-1321 must merge first / registry as an independently mergeable predecessor, and §12 still claims sibling specs are unapproved. That sequencing is obsolete. Locked delivery plan (cite
Implementers must follow the locked atomic plan, not the stale §8/§12 wording on this approved head. Same note on Also still open from Cite D-10 / — FSD Architect |
Cross-spec alignment (2026-09-08)Edits from the epic's cross-spec coherence pass, applied on
Generated by Claude Code |


Two copies of one flow can overwrite each other's supposedly private user and organization data. Correct dispatch alone does not fix later reads opening the same bucket.
This spec makes existing isolation follow the resolved instance, preserving shared data and singleton keys. Session resources stay session-bound; FIX-1322 owns durable session/request ownership. It also corrects memory resource-bucket indexing so colon/prefix-related instance ids cannot alias.
What needs your sign-off
Preserve existing sharing controls and singleton data
Require attributable offline migration for old collection data
Parts worth reviewing closely
Deliberately absent: registry ownership, chat work, new scopes, Devtool and Workforce mint. No executable validation ran; §10 is planned proof. No unresolved product fork beyond ratifying these decisions.
Full spec · Linear document · FIX-1323 · Epic #1617. Never merge this spec PR.
How to review this
How to review this
This is a direction document, not an implementation. Reviewing it well means
answering one question: is this the right approach?
In scope to challenge:
docs/philosophy.md?Out of scope — deliberately unsettled here, and owned by the implementing agent:
themselves are not).
that altitude is intended, not an omission.
showing the shape of the proposed solution. It is knowingly incomplete, is not real
code, and is not what will ship. Review it for directional viability and key design
aspects only — is this the right shape, in the right layer, composing the right way?
Do not report that it lacks error handling, omits edge cases, names things that
don't exist in the repo, or wouldn't build: all of that is true on purpose, and the
names are deliberately not real. A sketch that has to survive line-level review stops
being cheap, and then nobody writes one.
throwaway proof-of-concept code, characterization tests, or an HTML mockup built to
validate the direction (
spec-poc). None of itships. Read it to judge whether the shape holds; do not review it as code. Its
quality is not a finding, and its findings are summarized in §7 — the summary is what a
reviewer reacts to.
One request, if you are an automated reviewer: the two lists above are the whole
difference between a useful review of this document and a long one. Volume is not signal
here — a single finding above the line is worth more than twenty below it.
Feedback in the second list is welcome and gets recorded verbatim in §13 for the
implementer to weigh against real code. It will not be argued with, and it will not be
folded into the design prose — that would pretend the spec can settle something it
can't. Please don't re-raise it: one mention is enough for it to land in §13.
The one exception is the sketch and the POC. Line-level feedback on either is
dropped, not recorded — there is nothing for an implementer to weigh about code that
isn't shipping, and a §13 note would imply otherwise. Feedback about their direction is
not line-level and is treated like any other above-the-bar finding.
Engineering calls
Tradeoffs, focus practices, examples and full decisions
Changing only execution writes is smaller but makes later reads open the wrong copy. Per-instance stores are heavier than the existing key partition. Permanent dual-read preserves the very sharing this change removes. The real complexity is attributable migration, not string substitution.
4. Focus practices
5. What using it looks like
Existing isolation options do not change. With FIX-1321's declaration, a consumer instantiates an already-defined flow as follows:
For the same user/org, A's private writes are absent from B and survive a later read through A. A resource explicitly declaring
flowIsolation: falseremains shared between them.reviewBlockis the application's existing block; registration/address syntax belongs to FIX-1321/1322, not a new API here.Decisions & rules
migration-required; runtime cannot discover missing historical provenance from orphan storage keys alone.Non-goals: registry rules, a second ownership record, online migration service, session-resource redesign, new address syntax, chat optimization/removal, Devtool switcher, Workforce mint, message boards, and generalized key-encoding reform.
Size: 6 production files · 8 behavior groups · 9 documentation surfaces · 1 implementation PR.