diff options
Diffstat (limited to 'docs')
| -rw-r--r-- | docs/design/wrapup-routing-spec-review.org | 172 | ||||
| -rw-r--r-- | docs/design/wrapup-routing-spec.org | 7 |
2 files changed, 178 insertions, 1 deletions
diff --git a/docs/design/wrapup-routing-spec-review.org b/docs/design/wrapup-routing-spec-review.org new file mode 100644 index 0000000..0a2fb8d --- /dev/null +++ b/docs/design/wrapup-routing-spec-review.org @@ -0,0 +1,172 @@ +#+TITLE: Review: Wrap-Up Inbox/Transcript Routing +#+AUTHOR: Claude Code (rulesets) +#+DATE: 2026-06-21 +#+STARTUP: showall + +* Scope reviewed + +Spec under review: =docs/design/wrapup-routing-spec.org= (read twice — once for problem/design/decisions, once for gaps). + +Code and conventions read, with the facts each contributes: + +- =.ai/workflows/wrap-it-up.org= — Step 3 is the attach point. Step 3 already runs four =todo-cleanup.el= passes (hygiene, =--archive-done=, =--sync-child-priority=, =lint-org=), the roam inbox sweep, and the *Inbox sanity check* (lines 209-233), which counts unprocessed =inbox/= files and marks the wrap incomplete if non-zero. The router would be a new sub-step after that check. +- =.ai/workflows/process-inbox.org= — the per-item disposition engine. Value gate (3 questions), Skeptical Review, Phase B classification (implement / fold / file as TODO), Phase B.1 priority-scheme check, Phase D apply, and the *reject-from-another-project* flow (Phase D, lines 172-182): write a response file, deliver via =inbox-send <sender> --file=, delete the local copy. This is the existing, tested recovery path for a mis-routed handoff. +- =.ai/scripts/inbox-send.py= — the cross-project handoff mechanism. =_is_project= (line 60-62) requires *both* =.ai/= AND =inbox/=. =discover_projects= scans =DEFAULT_ROOTS= (=~/projects/=, =~/code/=). Filenames follow =YYYY-MM-DD-HHMM-from-<source>-<slug>.<ext>= (lines 176, 196). =build_text_org= stamps =#+SOURCE: from <source>= — provenance is already baked into every inbox-send drop. +- =claude-rules/cross-project.md= — inbox-send is the *sanctioned* cross-project write path; doing work directly in another project's scope is otherwise stop-and-ask. A proactive inbox-send needs no confirmation; a cross-scope edit does. +- =.ai/scripts/todo-cleanup.el= — =tc--find-section= (line 141) matches the unique level-1 "Open Work"/"Resolved" heading, returning =nil=/='multiple=. =--archive-done= (=tc-archive-done-in-file=, line 207) is an *intra-file* subtree move: cut from "Open Work", reinsert under "Resolved" *in the same buffer*. There is no cross-file or cross-repo move primitive today — Phase 2 would be net-new. + +Live precondition check run during this review (=ls= over =~/code/=): +- =little-elisper=, =rsyncshot=, =winvm= — have =inbox/= AND =todo.org= with one "Open Work" heading. +- =chime=, =yt-sync= — have =inbox/= but *no =todo.org= at all*. + +* Implementation-readiness + +*Not ready.* One blocking architecture finding (H1: the direct-move design vs the inbox-route alternative) must be dispositioned before Phase 2 can be built, because the answer reshapes D2 and Phases 2-4. A second blocking finding (H2: the candidate-set marking mechanism, which the spec itself flags for spec-review to pin) has no decision recorded. The six =* Decisions= all read =DONE= ([6/6]), so the spec passes the decision-cookie gate — but D2 ("move atomically through a helper") is the decision H1 challenges, and the spec explicitly defers the candidate-set marking to "Phase 3/4" without choosing, which is the unresolved core of H2. + +* Overall assessment + +The problem is real and well-scoped: an accepted keeper filed into the wrong project's =todo.org= has no clean home, and the manual two-repo edit is error-prone. The interaction design (batch go/skip, no per-item triage, reviewable surfaced list, confidence labels) is sound and matches Craig's stated ask. The decision record is unusually thorough, and the Reading-A-vs-B resolution (router acts on filed keepers, not raw inbox files) cleanly separates the router from the inbox gate. + +The risk is the central mechanism. The spec chooses to build a *new* primitive — an atomic helper that edits another repo's =todo.org=, inserts an externally-authored task the destination never gated, removes the source, and leans on a "load-bearing, not decorative" provenance note (D3) to make the boundary crossing legible after the fact. That primitive is data-loss-adjacent (source deletion + foreign-file write across two repos), it bypasses the destination project's value gate / priority scheme / Skeptical Review, and it duplicates a capability the codebase already has in a tested, convention-sanctioned form: =inbox-send= + =process-inbox=. The spec's own *Alternatives Considered* section names "Reuse process-inbox's file-as-TODO with a destination argument" and rejects it on cadence grounds (per-item mid-session vs batch-at-wrap) — but that rejection argues against reusing process-inbox's *caller*, not against reusing inbox-send's *delivery channel*. The two are different. Inbox-routing keeps the batch-at-wrap interaction and swaps only the delivery: drop a file, let the destination's own next session file it. That alternative was not evaluated as such, and the live precondition facts (below) tilt the decision toward it. + +* High-priority findings + +** H1 — Inbox-route supersedes the direct-move helper; D2 and Phases 2-4 should be reshaped [BLOCKING] + +*Spec text.* D2 ("Move atomically through a helper, never hand-edit two repos") and the Implementer "move helper" / "cross-project write discipline" notes build a new atomic primitive that writes into project X's =todo.org='s Open Work section and removes the source, authorized by the batch "go", made legible by a provenance note (D3). + +*Why it matters.* The inbox-route alternative — for each routable keeper, =inbox-send <destination> --file= (or --text) a handoff into the destination's =inbox/=, then *leave the source where it is* until the destination's own next session runs =process-inbox= and files it per that project's value gate, priority scheme, and =todo-format.md= — beats the direct-move design on five of the six axes the review weighed, and the deciding axis (precondition) was discovered to run opposite to the spec's assumption. + +1. *Safety / data-loss.* Direct-move is a cross-repo write to a foreign =todo.org= plus a source deletion, in one helper — the worst-case failure (a confidently-wrong destination, which the spec names as "the worst failure" in Readiness dimensions) corrupts another project's tracker with an externally-authored task and removes it from the source. Inbox-route's worst case is a file dropped in the wrong inbox/, which the receiving session simply *rejects via process-inbox's existing reject-from-another-project flow* (process-inbox.org Phase D, lines 172-182) — a tested recovery path that already exists. Dropping a file is additive and non-destructive at the destination; editing its =todo.org= is not. The source-deletion half can also wait: under inbox-route the source keeper stays put until the destination confirms by filing, so a mis-route never both corrupts the destination and loses the task. + +2. *Reuse / does inbox-route make Phase 2 unnecessary.* Yes. =inbox-send.py= already does cross-project delivery with from-prefix, date stamp, and =#+SOURCE:= provenance — the exact provenance D3 calls "load-bearing" is already emitted by =build_text_org= (inbox-send.py line 152-160) and the filename convention (line 176/196). =process-inbox= already does the per-item filing with the destination's conventions. Phase 2 (the new atomic move helper) and the load-bearing-provenance design (D3) both become unnecessary under inbox-route — the spec would build and test a new data-loss-adjacent primitive to do what two tested tools already do together. This is the strongest argument: the codebase already sanctions this exact path (=cross-project.md=: "Dropping a proposal in its inbox is the sanctioned alternative" to cross-scope work). + +3. *Conventions.* Direct-move inserts a task the destination *never gated* — it skips that project's value gate (does this advance a TODO / improve the project / serve the mission), its Phase B.1 priority scheme (an ungraded TODO in a project with a legend is a defect per process-inbox.org Common Mistake 3), and its Skeptical Review. The provenance note (D3) is an honest admission of exactly this gap: "the destination's next session sees an externally-authored task it didn't file." Inbox-route closes the gap by construction — the destination's own session, which understands its context, files the item through its own gate. + +4. *Latency.* This is the *only* axis direct-move wins. It lands the task in =todo.org= immediately; inbox-route defers filing to the destination's next session. But the task being routed is, by definition, *not* this project's work — it's homeless backlog headed elsewhere. There is no actor in the current session who needs it filed now. Immediacy buys nothing here and costs the safety, reuse, and convention wins above. (The startup workflow auto-runs process-inbox when a destination's inbox is non-empty — Phase C step 2 — so the deferral self-resolves on the destination's next launch with no manual step.) + +5. *Precondition — discovered fact, reverses the spec's assumption.* Direct-move requires the destination to have a =todo.org= with a clean "Open Work" heading (D1, via =tc--find-section=); a destination lacking it is "surfaced and skipped, never guessed at" — i.e. *unroutable*. Inbox-route requires the destination to have an =inbox/=. Live check today: =chime= and =yt-sync= have =inbox/= but *no =todo.org= at all*. Under direct-move, a keeper whose home is chime or yt-sync is silently unroutable — the discovery filter excludes them. Under inbox-route, it is delivered fine. The spec assumed "destination has an inbox" was the fragile precondition; the opposite is true — inbox/ was just created for these projects, while todo.org is the one that's missing. Inbox-route degrades gracefully where direct-move drops the task on the floor. + +*The one counter to weigh honestly:* inbox-route requires the destination's session to *eventually run*. A project Craig never opens accumulates unfiled handoffs in its inbox. But that is strictly better than direct-move's failure mode (a project Craig never opens gets a foreign task silently inserted into its tracker that no session gated), and the inbox sanity check on *that* project's next wrap surfaces the backlog. Stale-inbox is visible and recoverable; a wrongly-filed foreign todo is neither until someone notices it. + +*Recommendation.* Reshape D2 to "deliver routable keepers to the destination's inbox via =inbox-send=; do not edit the destination's =todo.org= directly." Drop Phase 2 (the atomic move helper) and D3's load-bearing-provenance burden — inbox-send provides provenance for free. Phase 4 wiring calls =inbox-send= per surfaced item instead of the helper. Keep Phase 1 (widened discovery) but change its filter to "has =inbox/=" (matching =_is_project=) rather than "has =todo.org= with Open Work" — though see H1a below on whether discovery needs widening at all. Keep Phase 3 (recommendation engine) unchanged; it is orthogonal to the delivery mechanism and remains the interesting, uncertain part. + +*** H1a — file-per-task vs one batch file: file-per-task fits process-inbox's per-item disposition + +If H1 is accepted: =process-inbox= dispositions *per item* (Phase B: read each, value-gate each, classify each). One file per routed task maps one-to-one onto that loop — each becomes its own act/fold/file/reject decision at the destination. A single batch file forces the destination session to split it back into items before disposition, re-deriving the boundaries the source already knew. Emit one =inbox-send --file= (or --text) per routable keeper. The batch "go" in the wrap interaction stays a single keystroke; it just fans out to N inbox-send calls instead of N todo.org edits. + +*** H1b — under inbox-route, "widened discovery" may be unnecessary + +The spec widens discovery from ".ai marker" to "todo.org with Open Work" so a plain code repo with a todo.org is routable. But =inbox-send='s existing =discover_projects= (=.ai/= + =inbox/=) is already the destination set for inbox-route, and it is tested. If the goal is "any project Craig keeps a todo.org in," note that today every code project with a todo.org checked here also has =.ai/= — so the existing filter likely already covers the real destination set. Phase 1 may reduce to "reuse =inbox-send='s discovery," eliminating new discovery code entirely. Confirm the destination universe before building a widened filter; if a real destination has =todo.org= but no =.ai/=, name it. + +** H2 — Candidate-set marking mechanism is unspecified [BLOCKING — spec flags this for spec-review to pin] + +*Spec text.* The Implementer "candidate set" note (and Phase 3) say the router considers only "keepers =process-inbox= filed this session whose inferred home differs from the current project," but explicitly leaves *how those are marked* to "Phase 3/4: either =process-inbox= tags a cross-project-candidate keeper at file time, or the router infers from a =CREATED= stamp dated this session plus content." The spec asks spec-review to pin it. + +*Why it matters.* Without a definite marking mechanism, Phase 4 cannot know which local tasks to surface, and the design constraint ("session-filed inbox keepers only, never the standing backlog") is unenforceable. The two options are not equivalent: + +- *Option A — tag at file time.* =process-inbox= Phase D ("File as TODO") stamps a property (e.g. =:ROUTE_CANDIDATE: <inferred-project>=, or a =:cross-project-candidate:= tag) on any keeper whose inferred home differs from the current project. The router then greps for that marker. *Pro:* precise — only genuinely-routable, this-session-filed keepers carry it; zero false positives from the standing backlog; the inference happens once, at file time, by the session that read the handoff (best context). *Con:* requires a paired edit to =process-inbox.org= Phase D (File-as-TODO), so the two workflows ship coupled. + +- *Option B — infer from CREATED-this-session + content.* The router scans =todo.org= for tasks with a =:CREATED:= (or =CLOSED:='s analog) stamp dated today and re-runs content inference. *Pro:* no process-inbox change. *Con:* fragile — depends on every filed keeper carrying a dated stamp (process-inbox.org Phase D does *not* currently stamp =:CREATED:= — verified: the File-as-TODO step adds priority/tags/body, no creation timestamp), so this option requires *adding* that stamp anyway, which is the same paired edit as Option A but with weaker precision; and "this session" is ambiguous across same-day sessions. + +*Recommendation — pin Option A.* Tag at file time. It is the most precise, it puts the inference where the context is richest (the session reading the handoff), and Option B's "no process-inbox change" advantage is illusory because process-inbox doesn't stamp =:CREATED:= today, so Option B *also* needs a process-inbox edit. Given both need the paired edit, choose the one with no false positives. Make this a new decision task (D7) and add the =process-inbox.org= Phase D edit to the implementation phases (currently absent — see M3). Under H1's inbox-route, the marker doubles as the inbox-send trigger: at wrap, any local keeper carrying =:ROUTE_CANDIDATE:= is the candidate set, and "go" inbox-sends each to its tagged destination. + +* Medium-priority findings + +** M1 — Recommendation-engine confidence has no defined threshold or low-match behavior [non-blocking] +Phase 3 and the Risks section lean on a "confidence label" and "low-confidence flagged" as the safety mechanism, but the spec never defines what produces high vs low confidence (project-name match in item text = high? topic-word overlap = low?) or what "go" does with a *no-match* item. Acceptance criterion "a low-confidence recommendation is visibly labeled" is untestable without the threshold. Define the tiers concretely: strong = destination project name or path appears literally in the item; weak = topic-word overlap only; none = stays put, never surfaced as a move. State whether "go" moves low-confidence items or only high-confidence ones. (This finding stands regardless of H1 — it's about the engine, which both designs share.) + +** M2 — "Empty routable set → router stays silent" is good; make it a tested case [non-blocking] +The Readiness dimensions say an empty routable set produces no prompt. Confirm Phase 4 surfaces nothing (not "0 items to route") so the common case — every wrap where nothing is routable — adds zero interaction. This mirrors =--archive-done='s real-mode no-op silence (=tc--emit-archive-report=, todo-cleanup.el line 399-409). Add it to the test surface. + +** M3 — Implementation phases omit the paired process-inbox edit [non-blocking, but fix before Ready] +H2's Option A (and Option B) require an edit to =process-inbox.org= Phase D to stamp the route-candidate marker at file time. No phase covers it. Whichever design wins H1, the candidate-set marker is filed by process-inbox, so a phase (or a line in Phase 3) must cover that paired workflow edit and its =.ai/= mirror sync. + +** M4 — Documentation plan names a cross-project.md edit; under H1 it changes [non-blocking] +The spec plans "a note in =cross-project.md= that the router is a sanctioned cross-project write path." Under direct-move that note documents a *new* write path (todo.org edit). Under inbox-route (H1) the router uses the *already-sanctioned* path (inbox-send), so the note becomes "the wrap-up router uses inbox-send to deliver routable keepers" — lighter, and consistent with the existing rule rather than extending it. Adjust per the H1 outcome. + +* UX observations + +- The batch go/skip with a reviewable surfaced list is the right shape and matches Craig's explicit ask; keep it under either design. The single keystroke fans out to N inbox-sends (H1) or N moves (as-written) — the user-facing interaction is identical, which is what makes H1 a clean swap behind the interaction. +- Surface the destination *and* the delivery mode in the list line so the user knows what "go" will do: e.g. =justin-onboarding → work (inbox handoff, high confidence)=. Under direct-move the parenthetical would read =(todo.org move)=; the user should see the consequence before the keystroke, especially since one writes a foreign tracker and one drops a file. + +* Architecture observations + +- The recommendation engine is correctly identified as the weak point and is well-isolated as a pure function =(item, project-list) → (destination, confidence)= (Phase 3). Keep it pure and unit-test it directly; it is the same under both designs. +- The spec's instinct to reuse =tc--find-section= and =--archive-done='s subtree logic is good *for an intra-file move* — but =--archive-done= moves within one buffer (todo-cleanup.el line 228-255 cut-and-reinsert in the same file). Extending it to cross-*file*, cross-*repo* moves is a larger change than "reuse the existing logic" implies: two buffers, two save points, two repos' git state, and atomicity across all of it. This under-stated delta is itself an argument for H1 — inbox-send is already a clean cross-repo file write with none of that coupling. +- Under H1, the only new code is the recommendation engine (Phase 3) + thin wrap-up wiring that shells out to =inbox-send= (Phase 4) + the process-inbox marker stamp (M3). That is materially less new surface than the as-written design (new atomic cross-repo move helper with ERT coverage for partial-move/rollback), and it reuses two tested tools. + +* Robustness and performance observations + +- Performance is a non-issue (inbox single-digits, projects in the tens, no hot path) — the spec's read is correct. +- Robustness flips decisively to inbox-route: the as-written design must prove "no partial move on failure" (acceptance criterion + Phase 2 ERT) for a two-repo write+delete. inbox-route has no partial-move state to defend — a file either copied or didn't (=shutil.copy2=, inbox-send.py line 198), and the source is untouched until the destination files it. The hardest test case in the spec (atomic no-op on failure) largely evaporates under H1. + +* Test strategy recommendations + +Under the recommended (H1 / inbox-route) design: + +- *Unit — recommendation engine (Phase 3):* strong match (destination project named literally in item) → high confidence + correct destination; weak match (topic-word overlap only) → low confidence; no match → stays put, not surfaced. Boundary: item naming *two* discovered projects (tie-break / lowest-confidence). Error: empty project list → all items stay put. +- *Unit — candidate-set marking (H2/Option A):* a keeper whose inferred home differs from current is stamped =:ROUTE_CANDIDATE:=; a keeper whose home *is* the current project is not; a standing-backlog task (no this-session marker) is never a candidate. +- *Integration — wrap-up wiring (Phase 4):* router surfaces N candidates → "go" issues N =inbox-send= calls landing N dated =from-<source>= files in the right inboxes, sources untouched; "skip" leaves everything in place, wrap completes clean; empty candidate set → zero interaction (M2). +- *Integration — destination lacking todo.org:* a candidate routed to chime/yt-sync (inbox/ but no todo.org) still delivers (proves H1's graceful-degradation win). +- *Recovery path (no new code, but assert it):* a mis-routed file at the destination is rejectable via process-inbox's reject-from-another-project flow — document this as the recovery story in the spec rather than testing it here (it's process-inbox's existing behavior). + +If the as-written (direct-move) design is kept instead, the test surface must add: successful cross-repo move, missing-destination-heading refusal, source-removal-only-on-success, and no-partial-move-on-failure across two repos — the harder surface H1 removes. + +* Documentation and tooling recommendations + +- =wrap-it-up.org= Step 3: add the router sub-step prose after the Inbox sanity check, stating clearly it is *optional* (skip is a clean wrap) versus the sanity check which *gates*. The spec's D1 already calls for this; make sure the prose names the gate-vs-optional split so a future reader doesn't collapse the two inbox-touching steps (the spec's own Risk #2). +- =cross-project.md=: per M4, adjust the planned note to the H1 outcome. +- =make test= picks up new ERT/ bats files by glob (confirmed in CLAUDE.md) — no Makefile edit needed for the new test files. Note this in the spec so the implementer doesn't add one. +- If Phase 3's engine is Python (to sit beside =inbox-send.py=) vs elisp (to sit beside =todo-cleanup.el=), the spec should pick — the language choice follows H1: under inbox-route the wiring shells to a Python script, so a Python engine co-located with inbox-send is the natural home; under direct-move it's elisp beside todo-cleanup. Name it. + +* Suggested spec edits + +1. Reopen D2 as a decision between direct-move and inbox-route, record the H1 analysis, and (recommended) resolve it to inbox-route. If inbox-route is chosen: mark the old D2/D3 SUPERSEDED, drop Phase 2, rewrite Phase 1 (discovery filter) and Phase 4 (wiring calls inbox-send), and update Acceptance criteria (the "moves into Open Work section / removes from source / no partial move" criteria become "delivers a from-<source> handoff to the destination inbox / source untouched / destination files via its own process-inbox"). +2. Add D7 pinning the candidate-set marking mechanism to Option A (tag at file time), per H2. +3. Add a phase (or fold into Phase 3) for the paired =process-inbox.org= Phase D marker-stamp edit + =.ai/= mirror sync, per M3. +4. Define the confidence tiers and no-match behavior concretely in Phase 3 / Acceptance criteria, per M1. +5. Add the "empty routable set → zero interaction" acceptance criterion, per M2. + +* Agreed decisions + +None reached interactively — this is a fresh-context review. The recommendations above are proposals for the author (spec-response) to accept/modify/reject. + +* Open questions + +1. *Direct-move or inbox-route?* (H1) — the one answer that reshapes the spec. Recommendation: inbox-route. +2. *Candidate-set marker: tag-at-file-time (Option A) or CREATED-inference (Option B)?* (H2) — recommendation: Option A. +3. *Does the real destination set include any project with =todo.org= but no =.ai/+inbox/=?* (H1b) — if no, drop the widened-discovery phase and reuse =inbox-send='s discovery. + +* vNext candidates + +- Transcript routing (already D4-deferred in the spec; carry to =todo.org= as =[#D]=). +- Per-item destination correction loop (spec's Alternatives "Neutral" note) — only if batch-only proves too blunt. + +* Implementation tasks (drop-in for todo.org) + +Reflecting the recommended inbox-route design (H1). If the author keeps direct-move, swap Phase 2 back in and adjust Phase 1/4 accordingly. + +#+begin_example +** TODO [#B] Wrap-up routing — destination discovery :feature: +Reuse inbox-send's project discovery (=.ai/= + =inbox/=) as the destination set; confirm whether any real destination has todo.org but no =.ai/= before building a widened filter. Spec: [[file:wrapup-routing-spec.org]] (Implementation phases, phase 1; review H1b). + +** TODO [#B] Wrap-up routing — candidate-set marking in process-inbox :feature: +Stamp =:ROUTE_CANDIDATE: <inferred-project>= on any keeper process-inbox files whose inferred home differs from the current project (Option A). Paired edit to process-inbox.org Phase D + =.ai/= mirror sync. Spec: [[file:wrapup-routing-spec.org]] (review H2/M3). + +** TODO [#B] Wrap-up routing — recommendation engine :feature: +Pure function (item, project-list) → (destination, confidence) with concrete strong/weak/none tiers; no-match stays put. Unit-tested. Spec: [[file:wrapup-routing-spec.org]] (Implementation phases, phase 3; review M1). + +** TODO [#B] Wrap-up routing — wrap-up step wiring :feature: +Add the optional router sub-step to wrap-it-up.org Step 3 after the inbox sanity check: surface candidates (destination + delivery mode + confidence), batch go/skip, "go" fans out one inbox-send per candidate, source untouched. Empty set = zero interaction. Sync the =.ai/= mirror. Spec: [[file:wrapup-routing-spec.org]] (Implementation phases, phase 4; review M2/M4). + +** TODO [#D] Wrap-up routing — transcript filing (vNext) :feature: +File a meeting recording into the destination assets/ per working-files.md, gated on the source-location decision. Spec: [[file:wrapup-routing-spec.org]] (Decisions D4, Implementation phases, phase 5). + +** TODO [#B] Wrap-up routing — test surface :test: +Unit: recommendation engine (strong/weak/none, two-project tie, empty list); candidate-set marker (cross-project keeper tagged, local keeper not, standing backlog never). Integration: go issues N inbox-sends to correct inboxes with sources untouched; skip leaves all in place; empty set = no interaction; candidate routed to a destination with inbox/ but no todo.org still delivers. Spec: [[file:wrapup-routing-spec.org]] (Acceptance criteria; review Test strategy). +#+end_example + +* Rubric + +*Not ready.* Two blocking findings (H1 reshapes the central mechanism; H2 pins the candidate-set marking the spec itself flagged) must be dispositioned before implementation. The most useful outcome of spec-response is to resolve H1 (recommend inbox-route) and H2 (recommend Option A), at which point the spec moves to Ready — the remaining findings (M1-M4) are author-discretion improvements that don't gate the rubric. diff --git a/docs/design/wrapup-routing-spec.org b/docs/design/wrapup-routing-spec.org index 5749491..08cdf1f 100644 --- a/docs/design/wrapup-routing-spec.org +++ b/docs/design/wrapup-routing-spec.org @@ -8,7 +8,7 @@ #+end_src * Metadata -| Status | ready for review | +| Status | reviewed 2026-06-21 — Not ready (H1/H2 blocking; see -review.org) | |----------+-----------------------------------------------------| | Owner | Craig Jennings | |----------+-----------------------------------------------------| @@ -183,3 +183,8 @@ Only after the transcript-scope decision resolves. File a recording into the des - What: resolved all three open decisions. The router's input is Reading B (filed keepers that belong elsewhere, not raw inbox files), so D1 keeps it a separate sub-step from the inbox gate and D5 keeps it distinct from the defer-and-stage router; D4 defers transcript routing to vNext. Reworked the design (input definition, a candidate-set note bounding the router to session-filed keepers) and Phase 3 to match. Cookie now [6/6]; Status moved to ready-for-review. - Why: Craig chose Reading B after the A-vs-B input ambiguity surfaced as the root under D1 and D5. Reading B keeps the inbox gate, the router, and defer-and-stage each simple instead of entangling three mechanisms. - Artifacts: this spec; the candidate-set marking mechanism is the one detail flagged for spec-review to pin. + +** 2026-06-21 Sun @ 01:58:41 -0400 — Claude Code (rulesets) — reviewer +- What: spec-review pass. Rubric *Not ready*, two blocking findings. H1: the inbox-route alternative (inbox-send each routable keeper to the destination's inbox/, let its own process-inbox file it) supersedes the direct-move design — reshape D2, drop Phase 2 and D3's provenance burden. H2: pin the candidate-set marking to Option A (tag =:ROUTE_CANDIDATE:= at process-inbox file time). Four medium findings (M1 confidence tiers, M2 empty-set silence, M3 paired process-inbox edit phase, M4 cross-project.md note). Full review + drop-in implementation tasks in the review file. +- Why: Craig challenged D2 directly (why edit a foreign todo.org rather than use the sanctioned inbox-send path). The review confirmed it: inbox-send already emits the exact provenance D3 reinvents, process-inbox already files per-item with the destination's own gate, cross-project.md sanctions the inbox path, and a verified precondition reverses the spec's assumption — chime and yt-sync have inbox/ but no todo.org, so direct-move silently drops keepers headed there while inbox-route degrades gracefully. +- Artifacts: [[file:wrapup-routing-spec-review.org][review file]]. Next: spec-response to disposition H1/H2 (recommend accept both), which moves the rubric to Ready. |