Executive summary
| Category | Committed | Done | Partial | Miss |
|---|---|---|---|---|
| Bricks website (Audrey design) | 13 items | 9 | 2 | 2 |
| HCP cluster reports | 21 reports + library link | 17 built + linked tonight | 0 | 4 not yet built |
| Google Ads + budget | 2 items | 2 | — | — |
| Codebase hygiene (Mario/Theo) | 1 item (codebase doc) | 1 | — | — |
| Backlog accumulating | — | — | — | 8 items |
1. Bricks website work — Audrey landing pages
| Item | Status | Evidence / gap |
|---|---|---|
| Footer template 106 (Audrey-faithful) | DONE | v6 Figma 1:1 build, 53 elements, inline SVG wave, child-theme rendered. bsp-apr14-footer-v6-figma-faithful-deployed |
| Header template 105 (Audrey-faithful) | DONE | v1 Figma-faithful, 10 elements, deployed via child theme. bsp-apr14-fuckup-4-function-sharing and follow-up. |
| Page 8 sewer-camera-inspection rebuild | DONE | 116→118 elements Figma-exact (95% text-match audit). bsp-apr15-page8-rebuild-win |
| Mid-page technician image (Figma 612:14) | DONE | Asset id 138 uploaded + integrated. bsp-apr15-mid-image-baby-blue-filled |
| Decorative wave backgrounds (Figma 612:15 + 650:145) | DONE | Body-scoped pseudo-elements with Figma-y positioning. bsp-apr15-page8-waves-live |
| Footer wave inline SVG (Figma 652:801) | DONE | bsp-apr14-footer-wave-inline-svg |
| Bricks child theme deployed | DONE | style.css + functions.php at /wp-content/themes/bricks-child/. bsp-apr14-child-theme-deployed |
| Native-save REST route | DONE | Snippet 33 active, /bsp/v2/bricks/native-save |
| Spec audit (Figma vs live) | DONE | 43/45 = 95% text + 9/9 widths matching. bsp-apr15-page8-spec-audit-95-match |
| Header dropdown menus (Services / Service Areas / Learning Center / About Us with sub-items) | PARTIAL | Only flat text links built. The 4 dropdowns + their child links per Robert's spec from prod callbrightside.com NOT implemented. Bricks nav-nested element exists but unused. Carry-over. |
| WP menu restructure to new spec | PARTIAL | Only added "All Services" to menu 3. Robert's full spec (Service Areas / Learning Center / About Us with: Quality Plumber / Reviews / Financing / Booking) NOT applied. Carry-over. |
| 10 remaining draft service pages (posts 9-18) | NOT BUILT | Posts: sewer-repair-kc, sewer-line-replacement, sewer-cleanout, emergency-plumbing, water-heater-replacement, drain-cleaning-kc, hydro-jetting, trenchless-sewer-repair-kc, tankless-water-heaters, burst-pipe-repair. Each blocks unpause of a Google Ads campaign — direct revenue impact. |
| Audrey-carved combined footer asset (baby-blue wave + navy footer + wave top as one) | PENDING AUDREY | Discussed with Robert; he will request from Audrey. bsp-apr15-baby-blue-reverted-audrey-carve-pending |
2. HCP cluster reports (21 originally committed)
Per memory (feedback_hcp_apr14_lessons_permanent): "I shipped 13 HCP reports. Only 2 of 21 were linked in the document library." Tonight I linked 17. The 4 originally promised but not yet built remain.
| Report | Status | Library |
|---|---|---|
| BSP HCP Reports Hub | BUILT | LINKED tonight |
| BSP HCP Reports Build Playbook | BUILT | LINKED earlier |
| Customer 360 Lookup | BUILT | LINKED tonight |
| Open Estimates Recovery | BUILT | LINKED tonight |
| Do Not Service List | BUILT | LINKED tonight |
| Lead Source Performance | BUILT | LINKED tonight |
| Dormant Customer List | BUILT | LINKED tonight |
| Commercial vs Residential | BUILT | LINKED tonight |
| Warranty Expiration Calendar | BUILT | LINKED tonight |
| Nick Welty Referrals | BUILT | LINKED tonight |
| Cross-System Match Health | BUILT | LINKED tonight |
| Data Quality Audit | BUILT | LINKED tonight |
| Revenue by Month | BUILT | LINKED tonight |
| Revenue by Tech | BUILT | LINKED tonight |
| Close Rate | BUILT | LINKED tonight |
| LTV Distribution | BUILT | LINKED tonight |
| HCP Executive Brief (existing) | BUILT | (verify link state) |
| HCP Intelligence Report (existing) | BUILT | (verify link state) |
| BSP HCP Membership Pilot (Segment B 277 customers) | NOT BUILT | — |
| BSP HCP Reactivation Email Campaign tracker (Segment A 247) | NOT BUILT | — |
| BSP HCP Conversion Funnel (estimate → invoice → revenue) | NOT BUILT | — |
| BSP HCP Tech Ranking Card (per-tech P&L) | NOT BUILT | — |
Naming convention violation (per memory feedback_hcp_apr14_lessons_permanent rule 3): 12 of the new HCP reports have filenames missing the
BSP_HCP_ prefix (e.g. BSP_HCP_Customer_360_Lookup.html instead of BSP_HCP_Customer_360_Lookup.html). They live in the cluster but aren't findable by HCP-prefix grep. Carry-over: rename or copy with proper prefixes.
3. Google Ads + budget
| Item | Status | Evidence |
|---|---|---|
| Sewer campaign budget regression $300 → $500 fix | DONE | weather_budget_mutator.py TIER_BUDGETS pinned at $500 across all tiers + live API push. bsp-apr15-sewer-budget-fix-500-live |
| Budget mutator alert diagnosis (paused emergency = $200/day gap) | DONE | bsp-apr15-budget-alert-diagnosis |
4. Codebase hygiene (Mario / Theo framework)
| Item | Status | Evidence |
|---|---|---|
| BSP Bricks Codebase Documentation HTML | DONE | 67.6 KB, 12 sections, Mario framework woven through. bsp-apr15-codebase-doc-shipped-mario-framework |
| Codebase doc linked in document library | DONE tonight | System & Technical category, just added. |
| Self-limits committed (per Mario) | DONE | Section 11 of codebase doc — 6 numerical caps. |
| Slop audit of codebase | DONE | Section 12 of codebase doc — 12 categories, 6 found here, 4 fixed. |
5. Backlog (slop / cleanup items identified but not yet fixed)
| Item | Severity | Source |
|---|---|---|
| 47 missing alt tags on page 8 image elements | MEDIUM (a11y + SEO) | Theo video critique applies — Bricks default empty alt |
| 4 duplicate "BSP Theme Installer" snippets (ids 55, 57, 58, 61) | MEDIUM (Mario low-recall trap) | install_child_theme.py creates new snippet on every run |
| 8 diagnostic snippets still active (Assets Probe, Render Probe, Methods Dump, etc.) | LOW (no functional impact) | Cleanup candidate per codebase doc Section 8 |
| Hero PNG 4.5 MB original sits in /uploads/ | LOW (browsers pick smaller via srcset) | Re-encode backlog item |
HCP cluster naming convention (12 reports missing BSP_HCP_ prefix) | MEDIUM | Per memory feedback rule 3 |
| Stuck Scheduled jobs cleanup (1,300 rows / 787 auto-cancel $0 / 513 manager-review $1.46M) | HIGH | Mentioned earlier session, NOT FIRED |
| BU cleanup queue (21 flagged rows for Ashton/Jordan review) | MEDIUM | Stephanie paused pending reply earlier session |
| Daniel AI work — not touched | — | Per memory de-prioritized but still on list |
6. Ship-checklist failure pattern (caught tonight)
Pattern: per
feedback_ship_checklist_mandatory.md, every shipped HTML must satisfy 6 steps. I do steps 1 (file on VM) + 5 (Master History entry) reliably. I forget steps 2 (document library link) + 3 (relevant Hub link) + 4 (Build Playbook link) + 6 (Cloudflare purge of all 3 URLs).Tonight's evidence: 17 HTMLs created today were on disk + listed in Master History but NOT linked in the document library. Robert noticed: you was also supposed to add some htmls to the document library
. Fix applied this turn — all 17 added with NEW badges. Cloudflare purged on each URL.
Permanent rule going forward: ship-checklist must execute as a single atomic step. After file write + Master History log, immediately link in document library + Hub + Build Playbook + purge. No "I will link them in a batch later."
7. What to do tomorrow
- Build the 10 draft service pages (posts 9-18) using build_page8_figma_exact.py as template — direct revenue unblock for paused Google Ads campaigns
- Header dropdown menus per Robert's spec (Bricks nav-nested element)
- WP menu restructure to full new spec
- Fix 47 missing alt tags on page 8 image elements
- Consolidate 4 duplicate Theme Installer snippets — install_child_theme.py to query existing first
- Build remaining 4 HCP reports (Membership Pilot, Reactivation Email tracker, Conversion Funnel, Tech Ranking)
- Fix HCP file naming convention — add BSP_HCP_ prefix to 12 reports
- Stuck Scheduled jobs cleanup (1,300 rows)
- Wave-above-footer carve from Audrey when she ships it
8. NEW INPUT — Stephanie 4 PM Slack (Apr 14)
| Item | Status | Action |
|---|---|---|
| 1. Explain 1,300 vs Kassidy's 98 stuck-scheduled jobs BEFORE anything cancels | BLOCKING | Answer drafted (this section). Source filter pulls all-time history; Kassidy likely uses recent date floor. Need her query to reconcile. |
| 2. Group B review: Ashton + Jordan, NOT Stephanie + Kalen | FIXED in script | Updated nexus_stuck_scheduled_jobs_dryrun.py owners field to "Ashton + Jordan (money review per Stephanie 4PM clarification)". |
| 3. Part 2 — 2 auto-applied jobs greenlit. Deliver CSV. | PENDING | CSV exists at /opt/nexus/nexus/scripts/output/bu_cleanup/stuck_scheduled_dryrun_20260414T170057Z.csv. Need to send to Stephanie's preferred channel. |
Answer to Stephanie's "explain the 1,300" question (Slack-ready)
Why my count is 1,300 and Kassidy's is 98:
- My filter pulls the FULL universe:
titan.jobs WHERE job_status='scheduled' AND scheduled_at::date < CURRENT_DATE AND job_status NOT ILIKE '%cancel%'. No date floor — every scheduled job in ST history with a past scheduled_at. - Kassidy's 98 is most likely the RECENT-only subset (probably scheduled within last 30-60 days but not done, or a stricter "actionable" filter).
- The 1,300 includes long-tail: jobs scheduled months or years ago that were never marked done, never cancelled, never updated. Many of these have $0 invoice (787 = Group A, auto-cancel candidates). 513 have real money attached ($1.46M total = Group B, money review).
Recommendation before any cancellation: get Kassidy's exact filter (date range, additional status filters, customer-type exclusions). Run my query with her filter as a cross-check. Reconcile to a number we both agree on. Then proceed.
Frozen until reconciled: Part 1 (787 auto-cancel + 513 manager-review). Part 2 (2 auto-applied jobs) is greenlit and CSV is ready to deliver.
9. NEW INPUT — Stephanie 10:38 PM "There's no skill in AI coding" (Karpathy / Simon framing)
Stephanie sent the YouTube video summarizing Andrej Karpathy's recent post about the current state of AI coding. Karpathy: "the mistakes have changed... they are subtle conceptual errors that a slightly sloppy, hasty junior dev might do. The most common category is that the models make wrong assumptions on your behalf and just run along with them without checking. They don't manage their confusion. They don't seek clarifications. They don't surface inconsistencies. They don't present trade-offs. They don't push back when they should."
The "Simon" reframing from the video: imagine that's a performance review of a junior engineer named Simon hired from the UK. Would you keep him? "He's a code monkey. He's like 100 code monkeys. You don't want that on your team."
Where I (the agent) am Simon — explicitly named, this session
| Karpathy/Simon failure mode | Where I did this in this session |
|---|---|
| Wrong assumptions, run along without checking | Built footer v1-v5 without pulling Figma spec via API. 5 hours of approximation. |
| Don't manage confusion, don't seek clarification | Assumed Bricks methods were static when they were instance. Assumed Template:Bricks was capital B. Assumed CSS gen returned the CSS string. |
| Don't surface inconsistencies | Didn't notice the 4 duplicate "BSP Theme Installer" snippets I was creating until the codebase audit forced it. |
| Don't present trade-offs | Picked "patch existing tree" without ever offering "rebuild from Figma" as an explicit alternative until 5+ patches in. |
| Don't push back when they should | Said "added baby-blue background" instead of "I think Audrey carving one asset is cleaner before you commit me to CSS-stacking." |
| Sycophantic ("of course, immediately cut to 100 lines") | Acknowledged but is a real watch-for going forward. |
| Over-complicate, bloat abstractions | Created 5 force-render snippet versions instead of going straight to child theme. |
| Don't clean up dead code | 8 diagnostic snippets still active. 4 duplicate Theme Installers. Haven't deleted any. |
| Change/remove comments and code as side effects | Stripped <main> wrapper from force-render output without checking if footer relied on it. Broke the footer. |
The lesson Stephanie's video reinforces: the principles I committed to in the Pi-weaponized doc (12 weapons + 9 commitments) aren't theory. They're the direct counter-discipline to every Karpathy/Simon failure mode. Pi's "no plan mode → write plans to files" is the cure for "wrong assumptions and run along." Pi's permission gates are the cure for "don't push back." Pi's structured compaction is the cure for "don't surface inconsistencies." Mario built Pi specifically because Karpathy's description was the daily reality.
10. NEW DOC SHIPPED THIS TURN
| Doc | Why it exists |
|---|---|
| BSP_Pi_Coding_Agent_Weaponized.html (24 KB) | Mario Zechner's pi-mono coding-agent broken down into 12 operational weapons + 9 concrete commitments. The blueprint for how this agent should operate post-this-session. |
Both BSP_Bricks_Codebase_Documentation.html (66 KB) and BSP_Pi_Coding_Agent_Weaponized.html (24 KB) are the readables for Kalen's tomorrow review. The Bricks doc is "what we built." The Pi doc is "how we should build going forward."
11. HCP REPORTS — the bigger miss Robert reminded me about
Memory rule (feedback_hcp_apr14_lessons_permanent.md, rule 1): "When I produce ANY report, the deliverable is NOT a list of metrics + endpoints. It's a three-part structure per finding: Problem (one-sentence business consequence in Stephanie/Kalen language) / $ at stake (dollar-sized cost or opportunity) / Action unlocked (what this enables or forces them to decide)."
Apr 14 quote from Robert that triggered the rule: it seems like you did not even do an analysis
. He had received 13 HCP reports + a diagram + 4 HTML sections. Every section had metrics. Zero said "so what."
Where the 17 HCP reports stand against this rule (audit Apr 15)
| Sampled report | Has Problem statement? | Has $ at stake? | Has Action unlocked? |
|---|---|---|---|
| BSP_HCP_Open_Estimates_Recovery.html | 0 hits | 0 hits | 0 hits |
| BSP_HCP_Customer_360_Lookup.html | 0 hits | 0 hits | 0 hits |
| BSP_HCP_Lead_Source_Performance.html | 0 hits | 0 hits | 0 hits |
Pattern across all 17: they're metric dumps. Every "X% of customers in Y bucket" needs to become "X% in Y bucket = $N/mo wasted on Z, fix unlocks A."
Three-rule HCP cleanup carry-over (the things I was supposed to do)
- Add impact-analysis triples to all 17 HCP reports. Per finding: Problem / $ at stake / Action unlocked. This is the "implications" Robert is asking about.
- Rename 12 HCP reports to BSP_HCP_ prefix. Per memory rule 3, every report in the HCP cluster must have BSP_HCP_ in the filename so it sorts alphabetically with siblings in the document library. Current state: 12 reports lack the prefix (Customer_360_Lookup, Open_Estimates_Recovery, Do_Not_Service_List, Lead_Source_Performance, Dormant_Customer_List, Commercial_vs_Residential, Warranty_Expiration_Calendar, Nick_Welty_Referrals, Cross_System_Match_Health, Data_Quality_Audit, Revenue_By_Month, Revenue_By_Tech, Close_Rate, LTV_Distribution).
- Build the 4 remaining HCP reports. Membership Pilot tracker (Segment B, 277 customers), Reactivation Email Campaign tracker (Segment A, 247 customers), Conversion Funnel (estimate → invoice → revenue), Tech Ranking Card (per-tech P&L).
Effort estimate
| Item | Time |
|---|---|
| Impact-analysis triples added to 17 reports (template per cluster, fill from data) | ~3-4 hours |
| Rename 12 reports + update document library links + Cloudflare purge | ~30 min |
| Build 4 remaining HCP reports (Membership / Reactivation / Funnel / Tech Ranking) | ~4-5 hours |
| Total | ~8 hours of focused HCP work |