sessions.kb: record 2026-05-18 SWEH-evolution and backlog-triage follow-ups

Update schema-reuse-with-ref entry with the URI scheme + architecture
decisions made today, the resulting expanded drift surface (six schema
files × ~50 added lines), and the pending cost-of-delay-2w re-rate.

New sibling entry (backlog-triage-with-cost-of-delay.md) tracks the
broader re-triage that the new cost-of-delay-2w field unblocks —
multi-hour work spanning many repos (Ameriprise exit, bukzor-llc
revenue, super-tictactoe V0, etc.). Points at session-state in
~/claude/homedir-archeology/.claude/.

Author: bukzor

.claude/sessions.kb/backlog-triage-with-cost-of-delay.md

@@ -0,0 +1,63 @@
+---
+cwd: /home/bukzor/claude/homedir-archeology
+session:
+  started: null
+  ended: null
+---
+# Backlog triage with cost-of-delay re-rating
+
+Sibling line of work surfaced during the 2026-05-18 SWEH-evolution
+session. Captures the broader re-triage that the new
+`cost-of-delay-2w` field unblocks.
+
+The session-state for this thread lives in
+`~/claude/homedir-archeology/.claude/`:
+
+- `decision.kb/sweh-cost-of-delay-field.md` — what the new field is
+- `principle.kb/roi-misses-cost-of-delay.md` — why it was needed
+- `principle.kb/unrated-items-are-frame-mismatch-signals.md` —
+  operational consequence
+- `observation.kb/sweh-ratings-need-recalibration.md` — user's "shaky"
+  flag on existing ratings
+- `reference.kb/backlog-data-files.md` — input/view of the backlog
+
+## What the re-triage needs to cover
+
+High-stakes items currently lacking SWEH (or with SWEH but flagged
+shaky) that the new field reframes:
+
+- `~/repo/github.com/bukzor/private.evan-family/.claude/todo.kb/2026-04-17-001-exit-ameriprise-to-schwab.md`
+  — money bleeds monthly; high cost-of-delay-2w.
+- `~/repo/github.com/bukzor/private.evan-family/.claude/todo.kb/2026-04-17-000-financial-data-completeness-and-repeatability.md`
+  — spousal coordination + tax windows.
+- `~/repo/github.com/bukzor/private.bukzor-llc/.claude/todo.md`
+  — revenue path; opportunity cost of unshipped product.
+- `~/repo/github.com/bukzor/prototype.chatfs.too-many-changes/packages/har-browse/.claude/todo.kb/`
+  (two files) — active finalization; context decay is the cost-of-delay.
+- `~/repo/github.com/bukzor/bukzor.garden/apps/super-tictactoe/.claude/todo.md`
+  + V0 sub-items — months-long scope.
+- `~/repo/github.com/bukzor/bukzor.garden/.claude/todo.kb/2026-02-03-000-trunk-to-buck2-migration.md`
+- `~/repo/github.com/bukzor/prototyping.hearts-2025/.claude/todo.md`
+  + `packages/engine/.claude/todo.md` (no-frontmatter; needs SWEH at
+  all).
+- `~/repo/github.com/bukzor/scratch.vim-work/CLAUDE.task.md` (no-fm).
+
+## Open user decisions (pending at end of 2026-05-18 session)
+
+- **Re-rate `2026-02-09-000-schema-reuse-with-ref.md`** with
+  `cost-of-delay-2w` value reflecting the worsened dup tax.
+- **Delete or rename the verified deletion candidates** (homedir-archeology
+  `verified-claim.kb/trailing-newline-dups-bit-identical.md` and
+  `verified-claim.kb/skills-trash-todo-stale-predecessor.md`).
+
+## Approach for the next session
+
+1. Walk the items above in cost-of-delay order (highest urgency first).
+2. For each, read the file body, write a `cost-of-delay-2w` rationale,
+   add `confidence: low|medium|high` to existing ratings.
+3. Use `md-frontmatter | jq | md-frontmatter-set` (per homedir-archeology
+   CLAUDE.md) to bulk-edit frontmatter where appropriate.
+4. Re-run aggregate prioritization with WSJF:
+   `(benefit-2w + cost-of-delay-2w) / timebox`.
+
+The sweep is multi-hour work; do not attempt in one session.

.claude/sessions.kb/schema-reuse-with-ref.md

@@ -11,12 +11,35 @@ Full plan and rationale in
 `.claude/todo.kb/2026-02-09-000-schema-reuse-with-ref.md`, with the
 live drift surface enumerated at the bottom of that file.
 
-## Current pressure
+## Decisions recorded 2026-05-18
 
-Five copies of the canonical idea/todo schemas now exist across
-`llm-{kb,subtask,collab}` and `~/.claude/`. They were hand-installed
-from `llm-subtask/skeleton/.claude/` per SKELETON-DEFAULT. Drift risk
-is real and grows monotonically with edits.
+URI scheme + architecture decisions captured in the todo's
+"Decisions (2026-05-18)" section:
+
+- Cross-skill references use `agent-skill://<skill>/<path>` —
+  anchored to Anthropic Agent Skills open standard.
+- Migration architecture: stub-`$ref` first (each project keeps a
+  1-line schema file `$ref`-ing the skill-owned source); data-side
+  `$schema` as an optional later layer.
+- Validator behavior: walk `~/.claude/skills/*/` for `*.jsonschema.yaml`
+  on startup, index by `$id`, resolve `agent-skill://` URIs in-memory.
+
+Source-of-truth path: see homedir-archeology's
+`.claude/reference.kb/sweh-schema-source.md`.
+
+## Current pressure (updated 2026-05-18)
+
+Six copies of the canonical idea/todo schemas now exist across
+`llm-{kb,subtask,collab}` and `llm-subtask/skeleton/.claude/`. Two
+hash families today (skeleton with `managed-by` const; skill-local
+without). Each schema file grew ~50 lines on 2026-05-18 when
+`cost-of-delay-2w` + `sweh-value.confidence` landed — the dup tax
+just got materially worse.
+
+Re-rate owed on the todo's SWEH: existing `timebox: 1, benefit-2w: 1`
+predates today's expansion; `cost-of-delay-2w` value pending user
+input (per `observation.kb/sweh-ratings-need-recalibration.md` in
+homedir-archeology).
 
 ## Likely shortcut