docs(hhtl): plan-review savant preflight brief for Q-NEW-1 + Q-NEW-2#165
Conversation
Drafts the joint plan-review savant brief that decides:
- Q-NEW-1: GridLake slot — path (a) W2.5 dedicated 0.5wk vs path (b)
PR-X10 A13/A14 absorption
- Q-NEW-2: PR-X14′ placement — path (α) concurrent with GridLake at
W2.5 vs path (β) sequential at W3
Lays out the 2×2 cell grid (A-α / A-β / B-α / B-β), the inputs the
savant should load before deciding, the per-criterion lean for each
question, and the verdict output format (appended to
pr-arithmetic-inventory.md § Shopping-list addendum as a new dated
entry plus a follow-up schedule-amendment PR).
Pins three forbidden constraints on the savant:
1. Additive only (no deletions of prompts / inventory / linalg /
pillar / ogit_bridge modules)
2. Per-repo boundary preserved (lance-graph jc copies stay where
they are; no cross-repo cleanup)
3. No scope creep into deferred decisions (no SQL frontend work, no
touching lance-graph datafusion deps, no Lance storage format
changes)
Session-author notes the sweet-spot read as A-β or B-β (low fan-out
heat, clean per-PR scope) but explicitly defers final cell to the
savant's verdict.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 8105c8e095
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| | | Q-NEW-2 = α (concurrent) | Q-NEW-2 = β (sequential) | | ||
| |---|---|---| | ||
| | **Q-NEW-1 = (a)** W2.5 dedicated GridLake | **Cell A-α**: W2.5 holds 10 workers (4 GridLake + 6 contract/bridge sub-DAG after A1). Single coordinator + single savant cycle covers all 10. Arc = 8.5 weeks. | **Cell A-β**: W2.5 holds 4 GridLake workers; W3 first half holds 4 X14′ workers. Two coordinator cycles, two savant cycles. Arc = 9 weeks. | | ||
| | **Q-NEW-1 = (b)** X10-A13/A14 absorption | **Cell B-α**: GridLake = A13/A14 inside PR-X10 (W1-W2); X14′ spawns at A13/A14 merge boundary inside W2 (concurrent with A2-A12 tail). PR-X10 sprint balloons to 16 workers. Arc = 8 weeks. | **Cell B-β**: GridLake = A13/A14 inside PR-X10 (W1-W2); X14′ takes W3 first half. PR-X10 stays at 14 workers. Arc = 8.5 weeks. | |
There was a problem hiding this comment.
This row says the B-α option makes the PR-X10 sprint "balloon to 16 workers," but the same document defines path (b) as adding A13/A14 to PR-X10 (12 → 14). That inconsistency makes coordinator-capacity and risk comparisons unreliable when choosing the cell, because B-α may be evaluated against the wrong staffing load. Please align this count (or explicitly define what is included in the 16) so the matrix is internally consistent.
Useful? React with 👍 / 👎.
|
|
||
| | Path | Description | Workers | Schedule cost | Critical-path heat | | ||
| |---|---|---|---|---| | ||
| | **(α)** | Concurrent with GridLake at W2.5 (A1 MultiLaneColumn is the chain dep; A2-A10 fan out parallel) | 10 in W2.5 (heavy fan-out) | 0 vs path (a); -0.5 week if combined with path (a) | A1 critical-path slip cascades to 9 downstream workers | |
There was a problem hiding this comment.
The schedule-cost cell for path α currently states both "0 vs path (a)" and "-0.5 week if combined with path (a)," which are mutually inconsistent under the same condition. This ambiguity can lead the savant to apply different timeline math depending on interpretation, so the decision grid should use one unambiguous baseline for the α delta.
Useful? React with 👍 / 👎.
2 participants
Summary
Drafts the joint plan-review savant brief that decides the two open Q-markers from PRs #162 and #163:
The brief lays out the 2×2 cell grid (A-α / A-β / B-α / B-β) the savant picks from, with per-cell trade-offs on coordinator load, fan-out heat, A1 critical-path risk, schedule cost, and scope cleanliness.
What the savant produces
A verdict appended to
pr-arithmetic-inventory.md§ "Shopping-list addendum" in this shape:The verdict is the input to a follow-up "schedule amendment" PR that applies the diffs and bumps the inventory.
Forbidden constraints carried forward
Three pins on the savant from the 2026-05-19 design session:
jccopies stay where they are; no cross-repo cleanupTest plan
Docs-only PR.
13dfcf9dCross-references
.claude/knowledge/hhtl-gridlake-pre-sprint-prompt.md(Q-NEW-1 source, masterade8edb2).claude/knowledge/hhtl-pr-x14-substrate-contract-prompt.md(Q-NEW-2 source, master56b26716).claude/knowledge/pr-arithmetic-inventory.md§ "Shopping-list addendum (2026-05-19 substrate-pair update)"13dfcf9dGenerated by Claude Code