diff --git a/docs/backlog/P3/B-0030-lint-with-exclusions-tool-typescript-otto-346-fourth-violation-with-real-cost.md b/docs/backlog/P3/B-0030-lint-with-exclusions-tool-typescript-otto-346-fourth-violation-with-real-cost.md
new file mode 100644
index 00000000..6f01f0c7
--- /dev/null
+++ b/docs/backlog/P3/B-0030-lint-with-exclusions-tool-typescript-otto-346-fourth-violation-with-real-cost.md
@@ -0,0 +1,116 @@
+---
+id: B-0030
+priority: P3
+status: open
+title: Extract `tools/hygiene/lint-md-with-exclusions.ts` (TypeScript) — markdownlint-with-repo-aware-exclusions tool; Otto-346 violation #4 this session, this one with real cost (~60s instead of ~3s)
+tier: hygiene-tooling
+effort: S
+ask: Aaron 2026-04-26 — *"this is like the python smell but with python and this one had a real cost it forgot to ignore upstram so it took like a minute to run instead of a few seconds, if it was cononalized in code like in ../scratch it would never forget to exclude directoris like our references"*. The bash pipeline `markdownlint-cli2 "**/*.md" | grep -E 'MD[0-9]{3}'` I composed inline lacked proper repo-aware exclusions for vendored / mirrored directories, ran ~60s instead of expected ~3s. Same Otto-346 pattern (recurring inline composition = missing substrate primitive). Per B-0015 P2 priority bump: target is TypeScript via Bun, not bash and not Python.
+created: 2026-04-26
+last_updated: 2026-04-26
+composes_with: [feedback_otto_346_dependency_symbiosis_is_human_anchoring_via_upstream_contribution_good_citizenship_dont_blaze_past_2026_04_26.md, B-0015, B-0027, B-0028, B-0031, tools/hygiene/fix-markdown-md032-md026.py]
+tags: [otto-346, recurring-pattern, missing-primitive, tooling-extraction, markdownlint, repo-aware-exclusions, real-cost, typescript, ts-migration]
+---
+
+# B-0030 — extract markdownlint-with-repo-aware-exclusions tool
+
+## Origin — Aaron 2026-04-26 catch with cost-evidence
+
+Aaron 2026-04-26 caught the pattern AND named the cost:
+
+> *"this is like the python smell but with python and this one had a real cost it forgot to ignore upstram so it took like a minute to run instead of a few seconds, if it was cononalized in code like in ../scratch it would never forget to exclude directoris like our references (not upstream that's proabalby a bad name i randomly chose, we should rectify to avoid wars/confusion becasue im using upstream incorrectly)"*
+
+This is **Otto-346 violation #4** this session:
+
+1. PR #541 — sort-tick-history-canonical.py (Python tool extracted)
+2. PR #542 — fix-markdown-md032-md026.py (Python tool extracted)
+3. B-0028 — gh-pr-state-summary tool (TS target; awaiting first-migration unblock)
+4. **B-0030 (this row)** — lint-with-exclusions tool (TS target)
+
+The differentiating factor: **this one had measurable cost**. Slow run (~60s) when properly-bounded would be ~3s. That's a 20x cost penalty per invocation, multiplied by every time I run the inline pipeline.
+
+## What the tool would do
+
+**Problem class**: ad-hoc invocations of markdownlint (or other lint tools) on `**/*.md` patterns lack ergonomic defaults for repo-specific exclusions. Each inline use forgets:
+
+- `references/` directory (vendored / mirrored upstream code we don't own)
+- `tools/lean4/.lake/packages/` (Lean dependencies)
+- Other generated / vendored / archive directories
+
+**Tool behavior** (proposed):
+
+- `tools/hygiene/lint-md-with-exclusions.ts [paths...]` — wrap markdownlint-cli2 with repo-aware default exclusions
+- `--strict` — fail on any violation (default)
+- `--summary` — print only error counts per file, not full output
+- `--target <pattern>` — override default scope to specific paths
+- TypeScript via Bun; reads exclusion config from `.markdownlint-cli2.jsonc` and applies before invocation
+
+**Cost reduction**: from ~60s with missed exclusions → ~3s with canonical exclusions. 20x speedup is real productivity.
+
+## Composition with sibling tools
+
+- `tools/hygiene/fix-markdown-md032-md026.py` (PR #542) — sibling: applies fix; this one detects with proper exclusions
+- B-0028 (`gh-pr-state-summary.ts`) — sibling extraction from same Otto-346 pattern; both target TS
+- `tools/hygiene/check-tick-history-order.sh` + `check-no-conflict-markers.sh` — sibling architectural shape (shell now; eventual TS rewrite per B-0015)
+
+The cumulative `tools/hygiene/` post-install batch awaiting TS migration:
+
+- B-0027 (markdown-table-cell-count fix tool — owed-build, TS target)
+- B-0028 (gh-pr-state-summary — owed-build, TS target)
+- B-0030 (this row — lint-with-exclusions — owed-build, TS target)
+- + eventual rewrites of #541, #542
+
+## Why TypeScript
+
+Per Aaron's prior priority bump on B-0015 (P3 → P2):
+
+> *"we need to move the typescript migration of our scripts to higher priority so you will stop trying to write python and shell code lol"*
+
+POST-install scripts target TypeScript via Bun. This is post-install (developer + CI machines have Bun).
+
+## Effort sizing
+
+- **Build the tool**: S (under a day). Wrap `markdownlint-cli2` with config-aware exclusion defaults.
+- **Read existing `.markdownlint-cli2.jsonc`** for current exclusion patterns; compose with directory-aware logic.
+- **Verify cost reduction**: measure before/after run-time on full repo.
+
+## Composes with
+
+- **B-0015** (TS-migration P2 priority — first-migration unblock applies)
+- **B-0027** (markdown-table-cell-count tool — sibling extraction)
+- **B-0028** (gh-pr-state-summary — sibling extraction)
+- **B-0031** (references/ directory rename — paired concern from same Aaron observation)
+- **Otto-346** (recurring-pattern absorption; this is the FOURTH instance this session)
+- **Otto-341** (mechanism over discipline; tools absorb the pattern)
+- **`tools/hygiene/fix-markdown-md032-md026.py`** (PR #542) — sibling Python tool
+
+## Meta-observation captured for substrate
+
+**Otto-346 violation #4 this session — the cumulative count IS the signal**:
+
+| # | Pattern | Outcome |
+|---|---|---|
+| 1 | Inline Python sort | PR #541 (Python interim) |
+| 2 | Inline Python markdown-fix | PR #542 (Python interim) |
+| 3 | Inline Python gh-JSON-parse | B-0028 (TS owed) |
+| 4 | **Bash markdownlint+grep** | **B-0030 (TS owed; this row)** |
+
+Four instances in one session is enough signal to *actually start the first sibling-migration*, not just queue more. The discipline is collapsing under repeated catches; the structural answer is the TS-tool that ships first and unblocks the rest.
+
+## What this DOES NOT do
+
+- Does NOT replace `markdownlint-cli2` — wraps it with repo-aware defaults
+- Does NOT auto-run on every commit — invoked explicitly when needed
+- Does NOT promise complete coverage of every lint scenario — only the recurring-use-case patterns
+
+## Owed work cluster after this row
+
+The post-install TS-migration batch:
+
+- B-0015 batch-resolve-pr-threads.sh → TS (P2)
+- B-0027 markdown-table-cell-count tool → TS (P3)
+- B-0028 gh-pr-state-summary tool → TS (P3)
+- B-0030 lint-with-exclusions tool → TS (P3, this row)
+- Rewrites of #541, #542
+
+Five-tool cluster. **First-migration unblock should happen now, not later** — the Otto-346 violations are accumulating proof that the queue is no longer queue but blocker.
diff --git a/docs/backlog/P3/B-0031-rename-references-directory-naming-clarity-avoid-upstream-collision-aaron-2026-04-26.md b/docs/backlog/P3/B-0031-rename-references-directory-naming-clarity-avoid-upstream-collision-aaron-2026-04-26.md
new file mode 100644
index 00000000..9fe8a233
--- /dev/null
+++ b/docs/backlog/P3/B-0031-rename-references-directory-naming-clarity-avoid-upstream-collision-aaron-2026-04-26.md
@@ -0,0 +1,99 @@
+---
+id: B-0031
+priority: P3
+status: open
+title: Rename `references/` directory — Aaron 2026-04-26 noted "upstream" naming was randomly chosen and collides with git-semantic meaning; rectify before language-wars/confusion compound
+tier: hygiene-naming
+effort: M
+ask: Aaron 2026-04-26 — *"references (not upstream that's proabalby a bad name i randomly chose, we should rectify to avoid wars/confusion becasue im using upstream incorrectly)"*. The `references/` directory holds vendored / mirrored upstream-codebase content; Aaron initially used the word "upstream" colloquially to refer to it, but "upstream" has specific git-semantic meaning (the parent branch / repo a fork tracks). Using "upstream" in this colloquial sense creates confusion and risks future agents/contributors interpreting it via the git-semantic. "References" is fine for the directory name; the issue is the colloquial vocabulary used around it.
+created: 2026-04-26
+last_updated: 2026-04-26
+composes_with: [feedback_otto_346_dependency_symbiosis_is_human_anchoring_via_upstream_contribution_good_citizenship_dont_blaze_past_2026_04_26.md, B-0030, docs/GLOSSARY.md, B-0010]
+tags: [naming-clarity, glossary, git-semantic-collision, vocabulary-discipline, otto-339-anywhere, references-directory]
+---
+
+# B-0031 — rectify "upstream" colloquial vs git-semantic naming around references/
+
+## Origin — Aaron 2026-04-26
+
+> *"references (not upstream that's proabalby a bad name i randomly chose, we should rectify to avoid wars/confusion becasue im using upstream incorrectly)"*
+
+Aaron self-corrected: he'd been using "upstream" colloquially to refer to the `references/` directory's contents (vendored / mirrored external code). But "upstream" in git semantics specifically means *the parent branch / repo a fork tracks*. Two different meanings; same word; recipe for confusion.
+
+## The naming problem
+
+Two distinct concepts conflated in current vocabulary:
+
+| Concept | What it is | Current name(s) used |
+|---|---|---|
+| Vendored external code | Code from other projects mirrored into `references/` for inspection / lineage / Otto-346 contribution-tracking | "upstream", "references" (mixed) |
+| Git fork-parent | The repo / branch the fork tracks | "upstream" (correctly) |
+
+The first usage ("upstream" for vendored mirrors) is **colloquial-but-incorrect**; the second ("upstream" for git fork-parent) is **git-semantic-correct**. They collide.
+
+## What this row addresses
+
+1. **Audit current usage** of "upstream" across `docs/`, `memory/`, `tools/`, and code comments — distinguish git-correct uses from colloquial uses
+2. **Define replacement vocabulary** for the colloquial sense:
+   - Candidates: `mirrored-references/`, `vendored-deps/`, `external-source-of-record/`, `inheritance-references/`, just `references/` with explicit definition in glossary
+3. **Update `docs/GLOSSARY.md`** to formalize the distinction
+4. **Sweep documentation** for misuses; replace colloquial-"upstream" with the chosen term
+5. **Code-comment audit** for the same pattern
+
+## Why this matters per Otto-339
+
+Per Otto-339 anywhere-means-anywhere: vocabulary collisions in substrate cause wrong-state-vectors when AI agents (or humans) read the substrate. "Upstream" interpreted via git-semantic when colloquial-sense was meant produces:
+
+- Wrong assumptions about repo relationships
+- Confused contribution direction (Otto-346 upstream-contribution gets confused with `references/` write-back which doesn't make sense)
+- Documentation drift as later contributors interpret per their own assumed sense
+
+Aaron's catch is preventive-discipline: **rectify before the language-war compounds**. Cheaper to fix at 2026-04-26 than after another 100 substrate references encode the colloquial sense.
+
+## Composes with prior
+
+- **Otto-346** (dependency symbiosis; upstream-contribution discipline) — uses "upstream" in the OSS-contribution sense (canonical repos like bcgit/bc-csharp); the colloquial conflation contaminates the precision Otto-346 requires
+- **B-0010** (memory-index-conventions doc) — sibling naming-discipline backlog row
+- **`docs/GLOSSARY.md`** — the right home for the formal distinction
+- **Otto-339** (anywhere-means-anywhere) — vocabulary precision applies to directory/file/concept naming
+- **Otto-286** (definitional precision changes future without war) — Aaron's *exact* phrase here is "to avoid wars/confusion"; this is preventive Otto-286 application
+- **B-0030** (lint-with-exclusions tool) — paired concern from the same Aaron message; the lint tool needs to know which directories to exclude AND those directories need clear names
+
+## Programming-language-as-religious-choice connection (Aaron's framing)
+
+Aaron added in the same message:
+
+> *"people literraly say your programming laganguage choice is like a religious choice, and there are programming language wars that resemble religious wars"*
+
+This composes with the naming-discipline at a meta-level: vocabulary collisions create the same religious-war pattern at the substrate-naming layer that programming-language choice creates at the implementation layer. Both are tribal-identity-via-shared-vocabulary patterns. Otto-286 (definitional precision changes future without war) explicitly names "without war" — vocabulary discipline is anti-religious-war discipline.
+
+## Effort sizing
+
+- **Audit**: M (~half-day) — grep for "upstream" across docs/memory/tools; classify each instance
+- **Decision on replacement vocabulary**: S (Aaron-decision; agent provides candidates)
+- **Sweep**: M (~day) — replace colloquial uses; preserve git-correct uses; update glossary
+- **Pre-commit lint candidate**: future tooling could flag colloquial-"upstream" in non-git-context (out of scope for this row)
+
+## What this DOES NOT do
+
+- Does NOT remove "upstream" from git-correct uses — those stay
+- Does NOT rename the directory itself necessarily — could just clarify vocabulary around it
+- Does NOT mandate immediate execution — research-grade backlog
+- Does NOT eliminate all naming-collision concerns — this is one specific instance; sister concerns surface separately
+
+## Operational implications
+
+Going forward (even before sweep lands):
+
+- When using "upstream" in substrate, default to git-semantic (fork-parent) unless explicitly clarified
+- For colloquial "vendored external code" sense, prefer "references/" + the eventual replacement vocabulary
+- Otto-346's "upstream contribution" means contributions to **canonical OSS project repos** (bcgit/bc-csharp etc.) — that use is git-semantic-aligned and stays
+
+## Cross-references for sweep
+
+When sweep happens, files most likely to need updates:
+
+- `docs/POST-SETUP-SCRIPT-STACK.md` (mentions upstream in OSS-contribution sense — git-aligned, keep)
+- `references/` README or similar (defines what's there)
+- Otto-NNN substrate referring to "upstream" — most uses are Otto-346-style git-correct contribution sense, but audit each
+- BACKLOG rows (B-0007 Bayesian primitives upstream — git-correct, keep)