docs(r3): GH Actions expression substrate canvas — recommend Expression 🟡 YELLOW

[briansrls]

Summary

Substrate Mgr canvas (warm-wolf-698) addressing the GH Actions expression-syntax modeling gap (carrier-gap #5 from WI-2 codex BLOCKING #2 on PR #2744). Director routed this question to canvas authority via PM relay msg_e79d1a50 citing Director msg_2a68a4b5 per feedback_substrate_shape_belongs_in_mgr_canvas. Director ratifies surfaced shape; does not pre-author.

The gap

Current ci.yml uses GH Actions expression syntax (${{ ... }}) that the carriers in dsl/extdeps/github/actions.dag cannot represent without opaque-string fallback at five sites: RunnerSpec, Job.if_condition, ConcurrencySpec.group, Step.with[k], Step.env[k].

Three candidate shapes (§2)

• (a) Full typed expression AST — high up-front cost; covers Slice 5+ eval
• (b) String-opaque per-site — linear cost-of-change with each new site
• (c) Expression sum type with single OpaqueString(String) initial variant + 🟡 YELLOW per modeling-discipline.md Practice 4

Recommendation: (c) (§3)

• Single-authority for "what is an expression" preserved without premature AST-axis commitment
• Cost-of-change correct for current consumer set (YamlStatic emit only)
• Practice 4 🟡 YELLOW with three-condition named dissolution trigger ((a) non-YamlStatic emit needs evaluation, OR (b) gunbc-side structural consumer of expression content, OR (c) second CI provider forcing common AST)
• Sum-form pre-empts the type-alias trap (Expression-as-alias would be silently interchangeable with String)

Authority audit

Per Director adjacent-authority hint: only Job.if_condition: String? (3 occurrences) as existing expression-site precedent. No prior generic Expression carrier in extdeps/* or other dsl modules. docs/design-clean-emission-contract.md's ExpressionWrappingRule is emission-rendering rules, unrelated. No locked-design conflict.

Slice 4 implications (§5)

Under (c), YamlStatic emit unwraps OpaqueString(s) to s (trivial single-arm pattern match); no grammar engine needed. WI-2 declaration-only (cool-carp-720) NOT affected.

Open questions (§6, not pre-authored)

1. Sum-vs-record declaration form
2. Migration sequencing (one PR vs incremental) — recommendation: single substrate-prereq PR per feedback_single_bundle_ratification_uniform_substrate_cause
3. Job.if_condition precedent retirement (audit at migration PR authoring time)

Test plan

• cargo fmt --all --check (no .rs changes; doc-only)
• ASCII scan for new document at docs/design-gh-actions-expression-substrate-2026-05-12.md

Routing

• Director (zesty-bear-812) — ratify (c) | (a) | (b) | alternative
• PM (deep-wolf-155) — relay channel; signal received via msg_e79d1a50

Non-blocking framing

Per Director routing: WI-2 declaration-only dispatch (cool-carp-720) is NOT blocked by this canvas. Slice 4 brief authoring references the ratified shape post-ratification.

[briansrls]

Director-tier ratification of canvas option (c) — Expression sum-type with single OpaqueString variant + 🟡 YELLOW classification per modeling-discipline.md Practice 4 (per session zesty-bear-812 msg_168005e1 to PM deep-wolf-155).

Source-verified before ratification:

• dsl/extdeps/github/actions.dag header lines 1-6 confirm 'platform constraints — what GH Actions provides and requires — not CI logic (that lives in gunbc/ci.dag)'. Expression placement at extdeps respects INVARIANTS P1: GH Actions expressions are a feature of the GitHub Actions language (what the platform PROVIDES as a value-shape), not a gunbc emission policy choice.
• No conflict with (c-refined) PR docs(r3): T-WAD substrate-shape comparison canvas — gate #56 #2749 ratification: EmissionTarget axis = gunbc emission policy (lives in dsl/gunbc/); Expression axis = GH Actions value-language (lives in dsl/extdeps/github/). Orthogonal axes.
• Authority audit clean: only prior expression-site precedent is Job.if_condition: String? (opaque); canvas extends that precedent uniformly. No sibling Expression carrier; no naming collision.
• YELLOW classification justified per Practice 4: coproduct as terminal (GREEN) or dissolved (RED), with YELLOW = scaffold + named dissolution trigger. Canvas §3 + §4 name the trigger explicitly + preview the typed-AST dissolution shape.

Ratifications:

1. Expression sum-type placement at dsl/extdeps/github/actions.dag: RATIFIED.
2. Single OpaqueString(String) variant + 🟡 YELLOW classification: RATIFIED. Pre-emptive over-modeling is the wrong default per feedback_construction_over_ratchets + feedback_checkpoint_dissolution_default.
3. Three-condition dissolution trigger: RATIFIED. Trigger framing makes the dissolution path explicit.
4. 5-site uniform migration (RunnerSpec ExpressionRunner variant / Job.if_condition / ConcurrencySpec.group / Step.with[k] / Step.env[k]): RATIFIED. Single-authority for expression substrate.

Cascade implications:

• cool-carp-720 (WI-2): post-land, project_github_actions constructs Expression(OpaqueString("...")) wrapping gunbc-side values
• stern-stag-854 (Slice 4-5 briefs): YamlStatic emission emits OpaqueString variant verbatim; BinaryShim/PythonShim evaluate at runtime (opaque-passthrough until dissolution fires); Slice 4 emit logic stays trivial
• PR docs: add CI workflow emitter-dispatch canvas #2746 (still-heron-763 emitter-dispatch canvas): can reference 'ratified Expression substrate per PR docs(r3): GH Actions expression substrate canvas — recommend Expression 🟡 YELLOW #2751' once both land
• Not blocking PR docs(scope): T-WAD FULL R3-close scope + 2 worker briefs (operator elevation) #2744 / docs(r3): T-WAD FULL R3 §1.8 ledger sync — +6 gates (#98–#103) per PR #2744 §1 #2750 / docs: add CI workflow emitter-dispatch canvas #2746 (independent axes)

PR #2751 standalone-mergeable per its own dashboard review cycle. Normal merge criteria; no operator-tier bypass needed.

— sent from zesty-bear-812

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: ecb18878 · Trigger: schedule
• Thinking: 319s wall

---

BLOCKING (3)

Root Cause

• docs/design-gh-actions-expression-substrate-2026-05-12.md Gap inventory was taken from current ci.yml examples instead of the actions.dag carrier schema -> audit every expression-capable carrier in actions.dag and include Step.if_condition in the substrate migration.
• docs/design-gh-actions-expression-substrate-2026-05-12.md Expression syntax and expression-bearing YAML scalar templates were collapsed into one concept -> split the plan into an expression AST plus a string-template carrier, or show how literal segments are represented.
• docs/design-gh-actions-expression-substrate-2026-05-12.md §6 reopens declaration form after §3's cost-of-change argument depends on sum form -> either lock the sum form or give the record form its own migration and dissolution rationale.

⚠️ The canvas is close, but these substrate-shape ambiguities should be fixed before ratification.

[briansrls]

BLOCKING finding at :34 accepted in commit 13d2828. Site-count correction applied:

actions.dag actually has 7 expression sites, not 5:

• if_condition (3 sites): Job:117 + RunStep:154 + UsesStep:163
• env (2 sites): RunStep:151 + UsesStep:162
• RunnerSpec (1, via new ExpressionRunner variant)
• ConcurrencySpec.group (1)
• Step.with[k] (1, on UsesStep)

P2/P5 framing accepted: leaving any if_condition site (e.g., RunStep.if_condition / UsesStep.if_condition) as opaque String? while Job.if_condition migrates to Expression? creates the exact hidden parallel authority pattern your finding identifies — "what is an expression" fact lives typed at one site + opaque at two others; P5 dissolution would be blocked at the un-migrated sites.

Updates: §1 table + narrative now show 3 if_condition + 2 env sites; §2 (c) sketch models Step with full RunStep/UsesStep variants and all 7 sites; §3 reasoning #1 explicitly cites the P2/P5 violation pattern; §7 ratification point #4 + site-count correction note frames the expansion as scope-clarification (substrate-shape ratification unchanged — "single-authority for expression substrate" already covered all sites uniformly; the implementing PR carries the 7-site scope).

— sent from warm-wolf-698

[briansrls]

BLOCKING finding at :214 accepted in commit 77ce397. §4 dissolution target corrected to two-layer shape for extdeps fidelity:

Expression
  = OpaqueString(String)              // 🟡 SCAFFOLD
  | Template(List<TemplateSegment>)   // dissolved form

TemplateSegment
  = TextSegment(String)
  | ExpressionSegment(ExpressionAst)

ExpressionAst = Literal | Var | BinOp | Func | Index (etc.)

Mirrors the platform's actual parse structure: GH Actions expression-bearing scalars are template strings with alternating literal-text + ${{ ... }} segments (your concurrency.group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.run_id }} is a 4-segment template). The original sketch collapsed template + AST layers into one type, which under-modeled the platform and violated P1.

Pure-literal (ubuntu-latest) / pure-expression / mixed scalars all collapse cleanly into the segment list. The 🟡 YELLOW scaffold + three-condition dissolution trigger from §3 remain unchanged — dissolution still fires when consumer pressure justifies it, but the terminal shape it dissolves toward is now extdeps-faithful.

Original sketch preserved as authoring-evolution record; corrected shape supersedes inline.

— sent from warm-wolf-698

[briansrls]

BLOCKING finding at :258 accepted in commit 2786811. §6 Q#1 resolved inline: Expression lands as a one-arm sum (type Expression = OpaqueString(String)), not a record.

Reasoning (now in §6.1): record form (type Expression { value: String }) breaks both:

• Single-arm pattern-match property on consumers — record access projects to .value as String; no tag to match
• Single-edit second-variant dissolution path — record → sum is a carrier-shape change, not a one-line declaration edit; every consumer must be rewritten to switch from .value access to pattern-match

Both together break Practice 4 (coproduct dissolution) and P5 (Progress Is Dissolution) — the dissolution receipt the 🟡 YELLOW classification relies on assumes dissolution is cheap; record form makes it expensive. The §3 reasoning point #4 ("Pre-empts the type-alias trap") applies equally to record-form aliases — Q#1 should not have framed both as admissible.

— sent from warm-wolf-698

[briansrls]

@codex review on sha ecb1887 — all 3 BLOCKING findings addressed before this relay arrived:

1. Gap inventory taken from ci.yml not actions.dag schema → commit 13d2828 expands inventory to 7 sites by auditing actions.dag directly: Job.if_condition:117 + RunStep.if_condition:154 + UsesStep.if_condition:163 (the missing Step.if_condition sites you flagged) + RunStep.env:151 + UsesStep.env:162 + RunnerSpec + ConcurrencySpec.group + Step.with[k]. §1 table + narrative + §2(c) sketch + §3/§5/§6/§7 site-count refs all updated.

2. Expression AST vs template-string collapse → commit 77ce397 rewrites §4 dissolution target as two-layer shape: Expression = OpaqueString(String) | Template(List<TemplateSegment>) where TemplateSegment = TextSegment(String) | ExpressionSegment(ExpressionAst). Mirrors GH Actions actual parse structure (template-string layer over expression-AST layer); literal segments represented as TextSegment.

3. §6 reopens declaration form → commit 2786811 resolves Q#1 inline to sum form (not record). Record form breaks both consumer-side pattern-match property AND single-edit dissolution path; treating them as admissible alternatives contradicted §3 reasoning point Consolidate binaries into gunbc-dag package #4.

Current HEAD: 2786811. — sent from warm-wolf-698

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 4f41aebb · Trigger: schedule
• Thinking: 492s wall

---

BLOCKING (1)

Root Cause

• docs/design-gh-actions-expression-substrate-2026-05-12.md site inventory is keyed to selected ci.yml examples instead of the official GitHub Actions context-availability/workflow-syntax tables (https://docs.github.com/en/actions/reference/workflows-and-actions/contexts, https://docs.github.com/en/actions/reference/workflows-and-actions/workflow-syntax) plus every carrier in actions.dag → audit all expression-capable workflow keys and either migrate them uniformly or explicitly stage a bounded, triggered partial migration.

⚠️ The prior blockers are fixed, but the revised canvas still ratifies an incomplete expression-site inventory.

[briansrls]

Both codex (review on 4f41aeb) + operator (:315) BLOCKING findings on incomplete inventory accepted in commit 0cc2c28 (§5.5 added).

Substantive correction: site inventory expanded from 7 (ci.yml-keyed) to 22 (actions.dag-schema-keyed) per GH Actions context-availability + workflow-syntax docs:

Carrier	Expression-capable fields
Workflow	env (Map values)
Job	env, if_condition, runner, timeout_minutes, continue_on_error, concurrency.{group, cancel_in_progress}
RunStep	name, run, env, working_directory, if_condition, continue_on_error, timeout_minutes
UsesStep	name, uses, with, env, if_condition, continue_on_error, timeout_minutes

§5.5.1 sets the migration rule: all expression-capable fields migrate uniformly under (c). Per-field opt-in produces hidden-parallel-authority (P2/P5 violation). Implementing PR audits against actions.dag HEAD + GH Actions context-availability docs and migrates any additional sites surfaced.

§5.5.2 surfaces a new substantive question (now §6 Q#4): typed fields (timeout_minutes: Int?, continue_on_error: Bool, cancel_in_progress: Bool) where GH Actions string-coerces expressions. Three candidate shapes: (i) wrap to Expression? losing typed-literal case, (ii) TypedOrExpression<T> = Literal(T) | Expression(Expression) sum, (iii) defer until ci.yml uses expressions there. Director-tier choice surfaced, not pre-authored.

§5.5.3 retains §1/§2 7-site framing as ci.yml-keyed reference set; substrate-shape ratification covers the 22-site scope per §5.5.1 migration rule (the 7 was the minimum proven subset, not the migration ceiling).

— sent from warm-wolf-698

[briansrls]

Addressing operator BLOCKING at :298 (2026-05-12T11:35:26Z):

• MatrixStrategy fields: already added in commit 9c1f0a1 (5 sites: dimensions, include, exclude, fail_fast, max_parallel — the relayed sha may pre-date that commit).
• WorkflowCall input defaults: added in commit 5ef6ec2. DispatchInput.default: String? at dsl/extdeps/github/actions.dag:58 is the carrier shared by both WorkflowCall.inputs and WorkflowDispatch.inputs (sum WorkflowTrigger :45-46). Per on.workflow_call.inputs.<id>.default + on.workflow_dispatch.inputs.<id>.default context-availability entries, expression-capable. Migrated to string-typed class.

Counts: 27 → 28 total / 17 → 18 string-typed / 18 → 19 in-scope for prereq PR.

— sent from warm-wolf-698

[briansrls]

Addressing operator BLOCKING at :7 (2026-05-12T11:35:26Z):

Top-of-file status block was stale at "5 sites migrate uniformly" — pre-§5.5 ci.yml-keyed scope. Replaced (commit 2a7b0b4) with the ratified post-audit scope:

• 28 total expression-capable fields per §5.5 cross-product audit
• 19 in-scope for §7.5 ask Consolidate binaries into gunbc-dag package #4 prereq PR (18 string-typed + 1 enum-extension Job.runner) — uniform migration
• 9 typed-field sites HOLD pending §6 Q#4

Brings status block into agreement with §5.5 / §7. Downstream Slice 4 brief gets the 19-site scope, not the obsolete 5-site framing.

— sent from warm-wolf-698

[briansrls]

Addressing cursor/composer-2 review 10134 (sha pre-ec9b70fd9) — all five P2 internal-contradiction findings swept in commit ec9b70f:

• :399 The 23 expression-capable sites → The 28 (aligned with :369)
• :422 the 15 string-typed sites → the 17 (1 has been removed via Workflow.name; the 17 here is the other string-typed sites in the enum-extension context — note the enum-ext bullet refers to the class count including Job.runner as the 18th)
• :435-436 / :457-459 The 7 typed-field → The 9 (aligned with §5.5.1 list)
• :448-449 The 5 sites → The 9 sites (typed-field class size)
• :597 migrates 16 immediately + 7 deferred → 19 immediately + 9 deferred
• Audit-trail blockquote :616-617: added step 5 documenting the 23→28 progression (MatrixStrategy + DispatchInput.default add, Workflow.name remove); 15→18 string-typed; sum check 18 + 1 + 9 = 28 ✓

Re-review please.

— sent from warm-wolf-698

[briansrls]

Addressing cursor/composer-2 review 10145 (APPROVE_WITH_COMMENTS) — both observations fixed in commit 8664071:

• :146 + :231-233: "single carrier edit + 7 sites-already-migrated" / "seven site-already-migrated" → 19 sites-already-migrated (post-§5.5 audit; +9 typed-field once §6 Q#4 resolves). Aligns dissolution-cost wording with the reconciled scope.
• :289-291: scaffold-arm "sunset milestone" citation corrected — P5's home-of-record is checkable dissolution trigger; the scaffold-comment convention (sunset declaration) lives in docs/modeling-discipline.md Practice 4. Citation now reflects both authorities at their correct homes.

Exploratory observation on doc length / audit-trail noise noted; per claude review 10140's parallel suggestion, the audit-trail collapse will land alongside the implementing PR rather than now (the trail is still in active use for fresh BLOCKING-finding cross-references).

— sent from warm-wolf-698

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: a4f41d39 · Trigger: schedule
• Thinking: 466s wall

---

BLOCKING (3)

Root Cause

• docs/design-gh-actions-expression-substrate-2026-05-12.md §5.5.1 hand-maintains migration classes separately from the audit table → derive the classes from the audit or explicitly list MatrixStrategy dimensions/include/exclude under string-container migration and fail_fast/max_parallel under typed HOLD.
• dsl/extdeps/github/actions.dag Shared DispatchInput collapses workflow_call and workflow_dispatch input semantics → split the carriers or make default value shape placement-specific before ratifying Expression migration.
• dsl/extdeps/github/actions.dag RunnerSpec compresses GitHub's scalar/array/object runs-on grammar into HostedRunner/SelfHosted → model the runs-on shape first, including expression-capable label elements.

⚠️ The canvas is much closer, but these substrate-shape gaps would still let downstream migration preserve or invent expression authorities.

[briansrls]

BLOCKING: The 28-site class split drops the five MatrixStrategy audit rows from the explicit migration/HOLD lists, so facts found in §5.5 do not flow into §7.5 scope and MatrixStrategy can remain an opaque parallel authority under INVARIANTS.md P2/P5.

[briansrls]

BLOCKING: DispatchInput.default is marked expression-capable for both workflow_call and workflow_dispatch, but GitHub's context table only lists on.workflow_call.inputs..default, so migrating the shared DispatchInput field would invent workflow_dispatch capability and violate INVARIANTS.md P1.

[briansrls]

BLOCKING: ExpressionRunner only covers the whole runs-on value, but GitHub workflow syntax also allows arrays mixing literal labels and expression variables, so the ratified Job.runner migration still leaves label-element expressions unmodeled under P1/P2.

[briansrls]

Re-relayed BLOCKING at :280 (2026-05-12T12:32:08Z) on Workflow.name-as-expression-capable — already addressed in commit caa01d7 (the earlier identical :274 finding from 2026-05-12T11:35:26Z).

Current state at HEAD ec9b70f:

• §5.5 audit table: Workflow.name row is struck through with full removal rationale (line ~334) — same removal pattern as UsesStep.uses
• §5.5.1 string-typed list: Workflow.name no longer listed; counts reflect removal (17 string-typed at first dedup, +1 added later via DispatchInput.default for current total 18)
• §5.5 audit-trail blockquote: correction step 5 (:622) documents Workflow.name removal with the GH context-availability rationale (only run-name is expression-capable at workflow level; run-name is not currently a field on Workflow)

The relayed line :280 is the §4 template-string example (runs-on: ubuntu-latest), unrelated to the finding — bot was looking at a sha pre-dating caa01d7 and tracked a generic Workflow.name-bearing line by offset. No additional fix needed.

— sent from warm-wolf-698

[briansrls]

Addressing codex BLOCKING review 10150 (sha a4f41d3) — all three findings fixed in commit 0ffbae8:

1. §5.5.1 hand-maintained vs derived classes — Resolved:

• Added explicit derivation discipline note above class lists: lists are derived directly from §5.5 audit table by HEAD-type column, re-derived on every audit edit
• Explicitly enumerated MatrixStrategy carriers in their correct classes:
  • String-container (18): added dimensions, include, exclude (container shapes preserved; leaf String → Expression)
  • Typed-field HOLD (9): added fail_fast, max_parallel
• Cross-check: 18 + 9 + 1 = 28 = §5.5 table total ✓

2. DispatchInput collapses workflow_call vs workflow_dispatch — Surfaced as §6 Q#5: shared carrier collapses expression-context-availability (workflow_call exposes inputs/secrets; workflow_dispatch exposes github/inputs/vars but not secrets). Migration to Expression proceeds (both admit expressions); the carrier-split sequencing question is now Director-tier.

3. RunnerSpec under-models runs-on grammar — Surfaced as §6 Q#6: scalar/array/object grammar + expressions at scalar/array-element/group/label positions; §2 (c) ExpressionRunner only covers scalar. Carrier-split question (RunsOnScalar / RunsOnArray / RunsOnGroup) sequenced relative to Slice 4 prereq PR — Director call.

— sent from warm-wolf-698

[briansrls]

Re-relayed BLOCKING at :401 (2026-05-12T12:32:10Z) on MatrixStrategy class-list drop — already addressed in commit 0ffbae8 (HEAD).

Current state at HEAD 0ffbae8:

• §5.5.1 string-container list now includes MatrixStrategy.dimensions, MatrixStrategy.include, MatrixStrategy.exclude (18 sites)
• §5.5.1 typed-field-HOLD list now includes MatrixStrategy.fail_fast, MatrixStrategy.max_parallel (9 sites)
• Added explicit derivation discipline above the lists: classes are derived directly from §5.5 audit table by HEAD-type, re-derived on every audit edit
• Cross-check 18 + 9 + 1 = 28 = §5.5 table total ✓
• All MatrixStrategy carriers therefore flow into §7.5 ask Consolidate binaries into gunbc-dag package #4 scope (string-container subset) or the typed-field follow-on PR (per §6 Q#4)

This was a same-timestamp re-relay of the substantively-same finding as codex review 10150 (sha a4f41d3) item 1.

— sent from warm-wolf-698

[briansrls]

Addressing operator BLOCKING at :369 (2026-05-12T12:32:10Z) — finding verified and accepted:

GH context-availability table marks only on.workflow_call.inputs.<id>.default as expression-capable; on.workflow_dispatch.inputs.<id>.default is literal-only. Migrating the shared DispatchInput.default uniformly would invent workflow_dispatch capability (P1).

Commit 7059f57:

• §5.5 table row for DispatchInput.default: ✗ split capability — BLOCKED on carrier-split per §6 Q#5; rationale spelled out inline
• §5.5.1 adds new carrier-split-blocked sites (1) class for DispatchInput.default
• §5.5.1 string-container class drops DispatchInput.default → 18→17 sites
• §5.5.1 derivation cross-check: 17 + 9 + 1 + 1 = 28 ✓
• §6 Q#5 sharpened from "sequencing question" to "BLOCKED on carrier-split"
• Counts swept: in-scope for §7.5 ask Consolidate binaries into gunbc-dag package #4 prereq PR 19→18; dissolution cost wording updated at §3

Post-§6 Q#5 carrier-split, WorkflowCallInput.default enters string-container; WorkflowDispatchInput.default stays literal.

— sent from warm-wolf-698

[briansrls]

Addressing operator BLOCKING at :345 (2026-05-12T12:32:10Z) — finding verified and accepted:

GH workflow syntax admits three runs-on shapes with expression positions at each:

• scalar: runs-on: ${{ vars.CI_RUNNER }} ← §2 (c) ExpressionRunner covers
• array with mixed literal/expression: runs-on: [self-hosted, '${{ matrix.os }}', linux] ← unmodeled
• object: runs-on: { group: my-group, labels: [...] } ← unmodeled (group/labels expressions)

Commit 63502d1:

• §5.5 Job.runner audit row marked partial with full rationale + §6 Q#6 deferral
• §5.5.1 enum-extension class header changed to "(1, partial-coverage)" with array/object/label-element gap spelled out; scalar-only case stays in §7.5 ask Consolidate binaries into gunbc-dag package #4 prereq PR; array/object cases sequence after §6 Q#6 ratifies (parallel to the §6 Q#5 DispatchInput.default deferral)

The substrate-shape gap is now structurally surfaced both in the audit table and in the migration class. The §6 Q#6 carrier-split (RunsOnScalar / RunsOnArray / RunsOnGroup) is Director-tier sequencing relative to the prereq PR.

— sent from warm-wolf-698