Skip to content

Fix main-red: link orphan doc emitted-crate-partition-design.md into the doc graph - #6723

Merged
briansrls merged 1 commit into
mainfrom
fix/doc-orphan-emitted-crate-partition
Jul 16, 2026
Merged

briansrls merged 1 commit into
mainfrom
fix/doc-orphan-emitted-crate-partition

Conversation

@gunbai-bot

@gunbai-bot gunbai-bot Bot commented Jul 16, 2026

Copy link
Copy Markdown
Contributor

What

#6708 added docs/plans/emitted-crate-partition-design.md with no inbound link, orphaning it in the doc-reachability graph. doc_graph_has_no_orphan_docs (dag/test/claim/doc_reachability_witness_test.dag) has been red on main since (run 29471295796, head eba7679), which blocks every open PR via the refs/pull/N/merge ref (head ∪ main).

Fix

Add the doc to the v2-self-hosting tracker's Related list — its topical home (the self-host compiler's emitted crate layout, peer to seed-shrink-census, general-body-producer-design), reachable from the ROADMAP.md root. Authored in the .dag authority (dag/gunbc/plans/v2_self_hosting.dag), generated .md kept in sync.

Placement is provisional — the doc itself says "Implementation owner TBD (operator considering placement)". Re-home freely; the only constraint is reachability.

Verified green-by-execution (release claim_batch)

  • doc_graph_has_no_orphan_docs: PASS (was FAIL on main — the discriminating RED control)
  • doc_graph_has_no_dangling_links: PASS (link targets a real file)
  • run_generated_artifact_drift_gate_body: PASS (.md == emit(.dag), no drift)

🤖 Generated with Claude Code

…the doc graph

#6708 added docs/plans/emitted-crate-partition-design.md with no inbound link,
orphaning it in the doc-reachability graph — doc_graph_has_no_orphan_docs has been
red on main since (run 29471295796), blocking every open PR via the refs/pull/N/merge
ref (head ∪ main).

Fix: add it to the v2-self-hosting tracker's Related list (its topical home — the
self-host compiler's emitted crate layout, peer to seed-shrink-census et al.),
reachable from the ROADMAP.md root. Authored in the .dag authority
(dag/gunbc/plans/v2_self_hosting.dag) and regenerated .md kept in sync.

Placement is provisional — the doc declares "Implementation owner TBD (operator
considering placement)"; re-home freely, the constraint is only that it be reachable.

Verified green-by-execution (release claim_batch, this tree):
- doc_graph_has_no_orphan_docs: PASS (was FAIL on main — the discriminating RED control)
- doc_graph_has_no_dangling_links: PASS (link targets a real file)
- run_generated_artifact_drift_gate_body: PASS (.md == emit(.dag), no drift)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
@briansrls
briansrls merged commit cdbb7db into main Jul 16, 2026
2 checks passed
@briansrls
briansrls deleted the fix/doc-orphan-emitted-crate-partition branch July 16, 2026 05:01
briansrls pushed a commit that referenced this pull request Jul 16, 2026
briansrls added a commit that referenced this pull request Jul 16, 2026
briansrls added a commit that referenced this pull request Jul 16, 2026
The emitted-crate-partition orphan is owned by Wave-2; main #6723
already links it via v2_self_hosting.dag. Do not patch another lane's
file to clear a differential red.

Co-authored-by: Cursor <cursoragent@cursor.com>
briansrls added a commit that referenced this pull request Jul 16, 2026
The emitted-crate-partition orphan is owned by Wave-2; main #6723
already links it via v2_self_hosting.dag. Do not patch another lane's
file to clear a differential red.

Co-authored-by: Cursor <cursoragent@cursor.com>
gunbai-bot Bot pushed a commit that referenced this pull request Jul 16, 2026
Integrate main (#6706 witness-corpus, #6711 deploy, #6723 doc graph)
while retaining the thin emitted pre-push hook and Rust transport from
this branch; witness corpus policy lives in githooks_pre_push_plan.dag.

Co-authored-by: Cursor <cursoragent@cursor.com>
briansrls added a commit that referenced this pull request Jul 16, 2026
* WIP: Machine-shape Phase 1: land std.machine_shape per the operator-agreed en

* WIP: Machine-shape Phase 1: land std.machine_shape per the operator-agreed en

* Drop redundant is_* predicates; scaffold evidence construction wall.

Review feedback on #6687: remove Bool coproduct predicates that duplicate
match discriminators; replace the prose data String with a Disposition
Scaffold bound to quantified_known (dissolves on witness-realization P0).
Witness tests match directly on variants.

Co-authored-by: Cursor <cursoragent@cursor.com>

* Rename Precision<T> to Estimate<T> to avoid approximate_field collision.

std.approximate_field already owns Precision for float significand/exponent
bits; machine-shape exact-vs-interval magnitudes become Estimate<T> with a
matching estimate field on Known.

Co-authored-by: Cursor <cursoragent@cursor.com>

* WIP: Machine-shape Phase 1: land std.machine_shape per the operator-agreed en

* Machine-shape Phase 1: allocation-as-contract reframe (redirect scope)

Rework std.machine_shape to definite grants via shape_from_catalog, drop
Quantified/Evidence; converge PlacementSupplyRow; derive Energy in measure.dag;
update plan doc per operator redirect; witnesses green by execution.

Co-authored-by: Cursor <cursoragent@cursor.com>

* Wire catalog_estimate into GPU catalog rows with green witness

GpuModelCatalogRow.boost_clock becomes Estimate<Hertz>?; RTX 5090 cites
Exact; estimate_floor consumer + witness exercises Exact and Interval.

Co-authored-by: Cursor <cursoragent@cursor.com>

* WIP: Machine-shape Phase 1: land std.machine_shape per the operator-agreed en

* Fix std machine_shape layer inversion; drop catalog_estimate scaffold

std.machine_shape is import-clean: agnostic InterconnectLink, no extdeps
imports. shape_from_catalog moves to extdeps.gpu.machine_shape;
pcie_interconnect_link derives beside PcieLink rows. Drop consumerless
catalog_estimate; revert boost_clock to citation-time Hertz. Plan doc:
layer-clean ruling 10 + census convergence rows.

Co-authored-by: Cursor <cursoragent@cursor.com>

* Address re-review 2: Nanosecond link latency, indexed validation

InterconnectLink.latency replaces ReadLatencyClass (avoids third fork
consumer). Edge validation builds domain/level Map indexes once. Plan
doc: Option-on-shape = structural absence, refuse never default.

Co-authored-by: Cursor <cursoragent@cursor.com>

* Add ExternalAuthority anchor to extdeps.gpu.machine_shape

Fixes extdeps_external_authority_gate_passes CI failure: new extdeps
module must carry extdeps_external_authority_anchor per live corpus policy.

Co-authored-by: Cursor <cursoragent@cursor.com>

* Update realization_measurement_loop plan after PlacementSupplyRow deletion

Replace stale PlacementSupplyRow citations with std.machine_shape grant
convergence (#6687) in embedded plan narrative cells.

Co-authored-by: Cursor <cursoragent@cursor.com>

* Dissolve map-membership predicates into map_contains_key.

Replace hand-rolled domain_id_in_index and owner_has_level_in_index
helpers with the canonical std.algebra surface per review 38676.

Co-authored-by: Cursor <cursoragent@cursor.com>

* Bind emitted-crate-partition doc for doc-graph reachability.

Main introduced docs/plans/emitted-crate-partition-design.md without a
bind: root; doc_graph_has_no_orphan_docs failed post-rebase. Anchor it
beside stage0_crate_layout_emit as the natural workflow-layer consumer.

Co-authored-by: Cursor <cursoragent@cursor.com>

* Fix stage0_crate_layout_emit parse: imports before data.

The doc-graph bind row must follow import declarations; placing it
first broke compile-clean batch 1 on dag_compile_clean_gate_passes.

Co-authored-by: Cursor <cursoragent@cursor.com>

* Revert stage0_crate_layout_emit.dag doc-bind hack.

The emitted-crate-partition orphan is owned by Wave-2; main #6723
already links it via v2_self_hosting.dag. Do not patch another lane's
file to clear a differential red.

Co-authored-by: Cursor <cursoragent@cursor.com>

* WIP: Machine-shape Phase 1: land std.machine_shape per the operator-agreed en

* Merge main and regen live-deploy apply golden.

CI failed live_deploy_emit_holds: branch lacked #6719 committed-golden
witness; after merging main the apply script drifted from emit (ROADMAP
projection). main_wet sync restores byte-identical golden.

Co-authored-by: Cursor <cursoragent@cursor.com>

* WIP: Machine-shape Phase 1: land std.machine_shape per the operator-agreed en

---------

Co-authored-by: Brian Searls <briansrls@gunb.ai>
Co-authored-by: Cursor <cursoragent@cursor.com>
briansrls added a commit that referenced this pull request Jul 16, 2026
…cAccount pricing (#6741)

* WIP: Machine-shape Phase 1: land std.machine_shape per the operator-agreed en

* WIP: Machine-shape Phase 1: land std.machine_shape per the operator-agreed en

* Drop redundant is_* predicates; scaffold evidence construction wall.

Review feedback on #6687: remove Bool coproduct predicates that duplicate
match discriminators; replace the prose data String with a Disposition
Scaffold bound to quantified_known (dissolves on witness-realization P0).
Witness tests match directly on variants.

Co-authored-by: Cursor <cursoragent@cursor.com>

* Rename Precision<T> to Estimate<T> to avoid approximate_field collision.

std.approximate_field already owns Precision for float significand/exponent
bits; machine-shape exact-vs-interval magnitudes become Estimate<T> with a
matching estimate field on Known.

Co-authored-by: Cursor <cursoragent@cursor.com>

* WIP: Machine-shape Phase 1: land std.machine_shape per the operator-agreed en

* Machine-shape Phase 1: allocation-as-contract reframe (redirect scope)

Rework std.machine_shape to definite grants via shape_from_catalog, drop
Quantified/Evidence; converge PlacementSupplyRow; derive Energy in measure.dag;
update plan doc per operator redirect; witnesses green by execution.

Co-authored-by: Cursor <cursoragent@cursor.com>

* Wire catalog_estimate into GPU catalog rows with green witness

GpuModelCatalogRow.boost_clock becomes Estimate<Hertz>?; RTX 5090 cites
Exact; estimate_floor consumer + witness exercises Exact and Interval.

Co-authored-by: Cursor <cursoragent@cursor.com>

* WIP: Machine-shape Phase 1: land std.machine_shape per the operator-agreed en

* Fix std machine_shape layer inversion; drop catalog_estimate scaffold

std.machine_shape is import-clean: agnostic InterconnectLink, no extdeps
imports. shape_from_catalog moves to extdeps.gpu.machine_shape;
pcie_interconnect_link derives beside PcieLink rows. Drop consumerless
catalog_estimate; revert boost_clock to citation-time Hertz. Plan doc:
layer-clean ruling 10 + census convergence rows.

Co-authored-by: Cursor <cursoragent@cursor.com>

* Address re-review 2: Nanosecond link latency, indexed validation

InterconnectLink.latency replaces ReadLatencyClass (avoids third fork
consumer). Edge validation builds domain/level Map indexes once. Plan
doc: Option-on-shape = structural absence, refuse never default.

Co-authored-by: Cursor <cursoragent@cursor.com>

* Add ExternalAuthority anchor to extdeps.gpu.machine_shape

Fixes extdeps_external_authority_gate_passes CI failure: new extdeps
module must carry extdeps_external_authority_anchor per live corpus policy.

Co-authored-by: Cursor <cursoragent@cursor.com>

* Update realization_measurement_loop plan after PlacementSupplyRow deletion

Replace stale PlacementSupplyRow citations with std.machine_shape grant
convergence (#6687) in embedded plan narrative cells.

Co-authored-by: Cursor <cursoragent@cursor.com>

* Dissolve map-membership predicates into map_contains_key.

Replace hand-rolled domain_id_in_index and owner_has_level_in_index
helpers with the canonical std.algebra surface per review 38676.

Co-authored-by: Cursor <cursoragent@cursor.com>

* Bind emitted-crate-partition doc for doc-graph reachability.

Main introduced docs/plans/emitted-crate-partition-design.md without a
bind: root; doc_graph_has_no_orphan_docs failed post-rebase. Anchor it
beside stage0_crate_layout_emit as the natural workflow-layer consumer.

Co-authored-by: Cursor <cursoragent@cursor.com>

* Fix stage0_crate_layout_emit parse: imports before data.

The doc-graph bind row must follow import declarations; placing it
first broke compile-clean batch 1 on dag_compile_clean_gate_passes.

Co-authored-by: Cursor <cursoragent@cursor.com>

* Revert stage0_crate_layout_emit.dag doc-bind hack.

The emitted-crate-partition orphan is owned by Wave-2; main #6723
already links it via v2_self_hosting.dag. Do not patch another lane's
file to clear a differential red.

Co-authored-by: Cursor <cursoragent@cursor.com>

* WIP: Machine-shape Phase 3 - locality: affinity as derived topology + Traffic

* WIP: Machine-shape Phase 3 - locality: affinity as derived topology + Traffic

* Give the unreachable-but-honest footprint-absent branch its own refusal cause

Review nit (review 38844): derive_affinity's Absent-footprint arm (structurally
unreachable once >=2 consumers implies a collected OperandFlow) was mislabeled
AffinityInsufficientConsumers. Split it into AffinityFootprintUnresolved so a
typed refusal never claims a cause it didn't observe.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

* WIP: Machine-shape Phase 3 - locality: affinity as derived topology + Traffic

* derive_affinity: refuse on operand mismatch, witness the unresolved-flow refusal path

Review findings (review 38850, REQUEST_CHANGES):
- affinity_collect_step discarded OperandFlow.operand and never checked it
  against derive_affinity's operand parameter -- a caller passing an
  unfiltered flows list could get a claim built from a different operand's
  rows (fail-open fabrication, DESIGN §5). Fixed by threading the expected
  operand into the fold and refusing typed (AffinityOperandMismatch) on
  disagreement.
- the OperandFlowRefused -> AffinityFlowRefused propagation path (module
  note's own promise) had no witness. Added
  witness_derive_affinity_refuses_unresolved_operand_edge and
  witness_derive_affinity_refuses_operand_mismatch.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

* locality_affinity: refuse typed on footprint disagreement across rows

affinity_collect_step silently kept the last OperandFlow row's footprint
for a shared operand, so inconsistent footprints across consumers would
fabricate an AffinityClaim/TrafficAccountRow instead of refusing per
DESIGN §5 (fail-closed). Add AffinityFootprintMismatch and a witness
that proves the refusal fires (review 38864).

* locality_affinity: dedupe consumers by content hash, not flow-row count

affinity_collect_step appended one consumer entry per matching
OperandFlow row, so a dependent reached by two edges for the same
operand double-counted toward AffinityInsufficientConsumers's >=2
threshold, and a genuinely single-consumer operand (two edges, one
dependent) could wrongly pass as having two consumers. Dedupe on
append (review 38995).

* WIP: Machine-shape Phase 3 - locality: affinity as derived topology + Traffic

* locality_affinity: route list membership through canonical any/hashes_contain

list_contains_hash and list_contains_int were parallel hand-rolled
membership predicates duplicating hashes_contain (v2.std.materialize)
and any (v2.std.algebra) respectively (DESIGN §2/§3 single-authority;
review 39006). list_add_distinct_hash now calls hashes_contain
directly; the batch dedup fold in distinct_batches_for_consumers now
calls any inline. No behavior change, verified green via
gunbc run --claim-run.

---------

Co-authored-by: Brian Searls <briansrls@gunb.ai>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant