# Deck & walkthrough — provenance, the version, and what is not in them
**Build pictured:** `v36-b11` · portal cache tag `makki-portal-v36-b11` · the portal's demo date is 2026-08-22.
**Build on disk today:** `v37-b5` (batch 5, time & crew — built and independently graded after the pictures were taken).
**Build published:** `v32-nb6`. Nothing in either artifact is live, and nothing was published, deployed or sent.
**For:** finish line 1, presentation readiness. Separate from, and not blocked by, launch readiness.
**On the folder name.** This work sits in `materials/deck-v37_2026-10-10/`. The `v37` in that name is the disk tag on the day the folder was made, not the build the deck shows. **The deck shows `v36-b11`.** The folder was not renamed because scripts and the frozen tree reference its absolute path; every delivered file carries `v36-b11` in its own name instead.
---
## The version statement — where it now appears
Omar's words: *"It uses the frozen v36 demo, so clearly identify that version and any later Batch 5 features it doesn't show."* The statement is in four places, three of them asked for:
| Where | What it says |
|---|---|
| **The deck, slide 2** ("How to read this") — a bordered strip under the lede, before the three label columns | *Every screenshot in this deck is of private build v36-b11, on the demonstration machine. The published site is a different, older build, v32-nb6. Nothing in this deck is live. Work built after v36-b11 is on disk today and is not pictured; the honest ledger names it under a fourth label, built since.* The slide title now reads **"Four labels. Three on every claim in this deck, one about the build."** |
| **The deck, slide 27** ("Built since this build, not pictured") — a new slide straight after the honest ledger | The headline is the version: *Every picture in this deck is private build v36-b11. The build on disk has already moved on.* The left column repeats that the published site is v32-nb6 and nothing is live; the right column is the "built since" list below. Deck is now **30 slides** (was 29). |
| **The walkthrough, top** — a paragraph in "Before you start", before the first timed section, headed *"The build you are showing — say this out loud before the first click"* | Same statement, in the presenter's voice, plus: if the presenter runs from the current disk build rather than the frozen one, Start Day and Payroll will show more than the deck's pictures. The label list under it is now **four labels**, with **BUILT SINCE** defined. The walkthrough's masthead carries a one-line version too. |
| **This file**, this section | — |
Also: the Start Day section of the walkthrough (0:20) carries a one-line aside saying it is the v36-b11 Start Day; the limits section (0:50) has a new heading *"Built since this build, not pictured"*; and the Q&A at the end has a new answer for *"Is this the latest version?"*
---
## Built since this build, not pictured — the fourth label
Since `v36-b11` was frozen at 17:10 UTC, batch 5 (`batch5-time-crew`, disk tag `v37-b5`) was built and independently graded: round 1 `SCORE ok=128 fail=1` (the one fail a defect in the Builder's own test, not the app), round 2 after a test-only repair `SCORE ok=129 fail=0` — the round-2 figure is the Builder's claim and the Checker's r2 grade is the word. Read from `hive/checks/2026-10-10_batch5-time-crew.md` (the criteria) and `hive/checks/2026-10-10_batch5-time-crew.r1.grade.md` (the grade), these exist on disk today and **are not in any screenshot in the deck or the walkthrough**:
1. **Start Day — a signature on the timesheet itself.** The site lead signs a man's line with a PIN. The PIN is checked and never kept; the record shows who signed, when, and the last two digits only (*"Signed Dale Ferris · 2026-10-10 18:26 · PIN ending 17"* in the grade). A wrong PIN leaves the row unsigned. (F13; rows 58–63; browser-proved in the Checker's `r1c` probe and the Builder's r2 suite.)
2. **Start Day — the bulk crew-time control** with the company's cost codes on it: one control puts the hours and a cost code on every man on the sheet; the codes offered are `LAB, EQP, DSP, ABT, SUB, PRM, OTH`; another crew's sheet is refused. (V39; rows 40–57.)
3. **Start Day — the three head starts come through the gated reader.** From the board, from the previous day, or flat: the men, their hours, their cost codes and the basis they came from, printed on the page. The board's per-slot man list is now data in a seed (`portal/seed-time-crew.js`), gated like the roster, rather than a note kept in one device's localStorage. (GL-001 · M-HEAD · F24 · V38; rows 24–39.)
4. **Payroll — four more filters** on the payroll screen (`app-ops.js`): office, project type, verify status and approval status, beside crew; filtering by office narrows the list (browser-confirmed *"Airdrie yard (demo): 12 -> 4"*). Payroll by cost code now splits across codes instead of one `LAB` line. The 32 / 40-hour overtime thresholds are read from the seed and are overridable, instead of being literals in `payrollWeek`. (GL-103 · F11 · A5 / A11; rows 16–23, 85–94.)
5. **Job costing — two new owner-only columns**, *Temporary labour* and *Contracted labour*, partitioning the job's existing `LAB` spend by hours share — never an addition on top. (F18 · A10 / F10; rows 69–77.)
6. **The day link — a man enters his own hours** on his no-login page; another man's line is refused; an expired link still refuses; no money key reaches the page. (F54; rows 78–84.)
7. **A time record carries more fields:** cost code, break, per diem per man per day, asset time, labour kind, signature. (F14 · A8 · F8; rows 40–49, 64–68.)
8. **Readers with no screen of their own yet:** a per-man week reader (`myWeekFor`, still *"not sent"* / *"In-app only"*), and an estimated-hours fixture with a *"near estimated hours"* reader (`nearingBidHours`) — the fixture is labelled synthetic and F5 / F6 stay PARTIAL behind the owner's workbook.
Every one of those figures is a **synthetic demo fixture that says so on its own row**; the batch did not touch `portal/data.js`. The real crew assignments, cost codes per man, estimated hours, per-diem policy and signature policy are still the owner's to give. Nothing in batch 5 sends anything; GL-023's texting half is Omar's.
**The Start Day on slide 14 is the `v36-b11` Start Day** — the un-started screen with the board's men ticked and the three start buttons. It has the cost-code dropdown per man and the "same as previous day" head start that existed before batch 5; it does **not** show the signature panel or the bulk control, which only appear on the started day and only in `v37-b5`.
The slide and the walkthrough carry items 1–7 (the slide folds 7 into 6). Item 8 is in the walkthrough only — it has no screen to picture even on the current build.
---
## Where the screenshots came from
Every screenshot was shot from a **frozen copy**, never from `portal/` — a Builder was editing `portal/` throughout.
| | |
|---|---|
| Frozen source | `materials/deck-snapshot-v36-b11_2026-10-10_1710/` (219 files) |
| Tree actually served | `materials/deck-v37_2026-10-10/_tree/v36-b11/portal/` — a byte-for-byte copy of that snapshot **plus** `brand/`, which the snapshot did not carry |
| Proof it is the same build | `roles.js` md5 `a3cdce0e97b1ad9277805bb547381891` in both; `sw.js` reads `makki-portal-v36-b11` in both |
| Shooter | `materials/ali-guide/shoot.cjs`, reused unchanged, pointed at the frozen tree |
| Browser discipline | one Chromium at a time, `the box's browser lock` held for each run (see the lock note below for the one defect in that) |
**Why `brand/` had to be added.** The frozen snapshot has no `brand/` folder, so the first run rendered a **broken logo image and fallback system fonts** on every page. Those shots were discarded and the run restarted against a tree carrying `portal/brand/` (logo + the two brand woff2 files). Nothing in `portal/` was written to; the brand folder was copied out of it, read-only.
---
## Could not be captured, and why
| | |
|---|---|
| **The owner's Safety page at phone width** | ~50,000 px tall at 390 px wide. Beyond any full-page capture on this box. The shooter tiles and caps it; the deck uses the **desktop** Safety screen instead |
| **The tasks page at phone width** | ~62,000 px tall. Same limit. Not used in either artifact |
| **Full-page phone shots generally** | several needed viewport-tiled stitching, and the log records repeated "unpainted band" retries. Where a tall phone page was wanted, the deck uses a **section or the viewport**, not a full-page shot |
| **Anything live** | the published site is still `v32-nb6` and nothing has been published since 2026-10-04. Every shot is of the **private build on this box**, not of anything on the internet |
---
## The 404 in the render log — identified and removed
The first render log recorded `console: Failed to load resource: 404` on the deck's PDF render and nothing on the walkthrough's. A static walk of every `src=`, `href=` and `url()` in the deck HTML found every referenced file present. The render script was then given server-side logging of every path it answers with 404, and re-run:
```
server: 404 for /favicon.ico (requested by the browser; not in the page's own asset list)
```
**It was Chromium's automatic request for `/favicon.ico`.** The deck ships no favicon and the little render server had nothing to answer. It never affected a rendered page: a favicon is not drawn into a PDF, and the walkthrough page (same origin, loaded second) showed no 404 because the browser remembers a missing favicon per origin for the session. The request is made by the browser process, which is why the page-level `response` listener never saw a 4xx — only the server log could name it.
**Fixed, not just explained:** both HTML sources now carry ``, which tells the browser there is nothing to fetch. The final render log reads, for both files, *"no console errors, no failed requests, no HTTP status ≥ 400"*, with the server logging no 404 at all.
---
## For Omar — two findings in `v36-b11`, both `portal/` edits for the Builder on his word
Neither was fixed here: this work writes only under `materials/`, and `portal/` has another writer. Both are present unchanged in today's `v37-b5` (grep-confirmed; the line numbers below are identical in `portal/` as of 18:48 UTC).
### (a) "Fieldflo" and "EazyLife" in user-facing text
A competitor's name and the agency's name appear in the app's own copy, on screens a client would be shown. **Twelve strings in source, fifteen name occurrences** (comments excluded). A browser scan of the frozen build — every login, every nav page, every nested tab, plus the site lead's started day — rendered **21 occurrences in 19 distinct sentences**, because the same string prints on more than one page. The scan is in `build/wordscan.md` / `wordscan.json`.
| # | File : line | Text (abridged) | Where it renders — browser-verified |
|---|---|---|---|
| 1 | `app-dash.js:283` | *"Fieldflo's four tabs · what is running today, the safety register, the machines and the pipeline"* | **Owner › Dashboard**, the sub-line under the page title |
| 2 | `app-board.js:114` | *"Nothing is texted — Fieldflo texts him, here the office phones him."* | the roll-call panel on **Owner › Dashboard**, **Site lead › Dashboard** and **Site lead › Start Day** |
| 3 | `app-board.js:125` | *"Fieldflo texts the no-show; here the office phones him."* | the **Roll Call modal** (opened from that panel; the scan does not open modals, attribution is from source) |
| 4 | `app-ops.js:352` | *"Text-out is simulated. Fieldflo texts each man his week… Texting is switched on later by EazyLife."* — **two names** | **Owner › Schedule Board** and **Site lead › Crew Board**, under the week grid |
| 5 | `app-ops.js:385` | *"When EazyLife switches texting on, this same button sends these and logs the replies."* | the text-preview sheet opened from the board's text-out button (source attribution) |
| 6 | `app-ops.js:393` | *"In-app only — the text version is switched on later by EazyLife."* | **Crew › My Schedule** |
| 7 | `app-ops.js:219` | *"This is the question Fieldflo sends by text… Texting comes later, once EazyLife switches it on"* — **two names** | **Crew › End of Day**, the band over the question |
| 8 | `app-ops.js:177` | *"In-app demo. Fieldflo texts the man… Texting comes later, once EazyLife switches it on"* — **two names** | **Site lead › Start Day, once the day is started**, the End-of-day confirmations panel |
| 9 | `app-ops.js:158` | *"32+ is the early overtime flag (same rule as Fieldflo's)."* | **does not render** — `app-day.js` replaces that timesheet panel with its own on both the un-started and the started day. Dead copy, still worth deleting |
| 10 | `app-ops.js:578` | *"Same map Fieldflo showed on the call — plus what our seats do with it overnight."* | **Owner › Money › QuickBooks**, "What would sync" panel head |
| 11 | `ops.js:167` | *"CSV import … works without the link (same as Fieldflo's fallback)."* | **Owner › Money › QuickBooks**, the Fallback line |
| 12 | `ops-forms2.js:67` | *"The asbestos-specific set Fieldflo showed: the manometer, the two-stage waste load-out and the AHERA credential check."* | **Owner › Safety** (inspection sets) and **Site lead › Start Day** |
Fieldflo 11 occurrences, EazyLife 4 — fifteen. One further Fieldflo mention was in the **walkthrough script itself** (the presenter's line at Team & Access, *"This is the Fieldflo 'toggle it live' move"*); it was reworded today to *"the 'toggle it live' move from the sales call we sat through"*. The deck carries neither name (grep: 0).
**What the deck does about it:** the owner's dashboard is cropped below the sub-line, the Crew Board is cropped above its note, the End of Day shot is cropped to the question block, and the QuickBooks screenshot was dropped entirely (slide 23 makes the point in words). The Start Day shot is the un-started screen, which carries none of them.
### (b) An oversized icon in `.muted` notes — five screens, one cause
An inline SVG from `icon()` (`app.js:60`: `