7.1The golden states something its own sources contradict
Fails the taskGolden Solution - Artifact Completeness
The mistake
The golden states a value its source, or its own page, contradicts, often one true at an earlier turn. Nothing graded catches it.
Do this instead
Before you complete Leg B, send Model B a cleanup turn that points at the area, never the value, then check every place the value appears.
Where it happened
2 real tasks. Each shows the spot and what it should have been; click any line to open the task there.
the words at faultthe row or field at faultmissing from the taskwhat is righta task file
Example 33Small Business & Personal Docs · 4 Leg A turns
What is wrongThe golden gives the Friday custody handover as 13:30 to 15:30. The calendar photo and the calendar service both put it at 18:30.
Leg A turn 2 hands over the school calendar, IMG_0814_calendar.jpg, and asks how many custody handovers clash with travel. Its margin notes give the handover time as 18:30.
on-site until 17:00 Fri, can't make the 18:30 exchangeand the same everywhere else the time appears
How it breaks the standardThe golden is the answer the subjective block is graded against, so every value in it has to match the inputs, graded or not. No criterion grades the time, and a 17:00 finish misses the handover at either time, so every criterion still passes and nothing flags it.
What is wrongThe golden page counts 16 label posts, in its header and in its cards, but a note further down still explains “why the count is 15, not 17”.
The export holds 17 posts, and two artist account posts come out, leaving 15. Leg A turn 2's shot of Marisol's queue adds a sixteenth.
…under P11 not ours to review, so not shown. That is why 15 of the 16 cards come from the export.
How it breaks the standardFifteen was right until Leg A turn 2. When the queue added a post, the header and the cards moved to 16 and the note did not, so the golden contradicts its own page. No criterion grades the note, so nothing flags it; a stale count like this costs points.
Point every ungraded value at its source line, and check each total or count against the same page.
Where the correct guidance is written
Section titles as they print, so you can search for them in the document.
QC spec · Golden Solution - Artifact Completeness
Subjective rubrics and pairwise scoring are graded relative to it, so a defective golden corrupts every subjective score on the task.
Guidelines · 7. Golden Solution - Leg B
Every factual claim is grounded in the actual inputs: images opened, videos watched, PDFs read. Nothing is asserted from a filename alone.
Including the values no criterion grades.
1 more source
QC spec · Golden Solution - Artifact Completeness
the golden asserts a finding the inputs do not support.
One of the shapes that fail the task.
Why it matters
The golden is the reference every subjective rating and the pairwise comparison are scored against, so a wrong value in it is graded as correct and a run that reads the source properly is marked down. Where a criterion grades the same detail, either your golden fails it or, if the criterion was written from the golden, a correct run does.
Example 33The golden states something its own sources contradictFails the task
Small Business & Personal Docs4 Leg A turnstask 6aa34dd1625bb5b220c6b1cb
From the taskFrom the Leg B conversation, with Model B
the mistake
Model B · Leg B turn 4
Everything runs: the matrix generates 42 school-year weeks, all 10 month calendars render, and collision cells light up correctly. The `ctx.SCH` error is just my test harness reaching a block-scoped `const` — the builders themselves executed fine and reference it without issue.
## What I read off IMG_0814 (Jaylen's calendar)
I cropped and enlarged every month to read the marks precisely, then cross-checked against the calendar's own holiday footer:
- **Carline mornings** (yellow highlight, 7:15–7:35 AM): every **Thursday + Friday** school day. After removing Labor Day, Fall Break (Oct 9–12), Thanksgiving, Winter Break, MLK, Presidents', Spring Break, Easter (Mar 29–Apr 2), and the May 21 last day → **71 mornings** (37 Thursdays, 34 Fridays).
- **Custody handovers** (red circles, "Terrance – Poplarville" 13:30/15:30 in the margin): **7**, all Fridays, biweekly — **Aug 7 & 21, Sep 4 & 18, Oct 2, 16 & 30**. Those are the only ones circled on the sheet (Nov–May had no circles), so that's what I counted.
## The exact collision numbers
Travel days aren't on the calendar, so I made them the interactive control. The decisive fact: **100% of your carline mornings and handovers fall on Thursday or Friday.** So the count is entirely about whether the regional travel block hits those two days:
| Travel block | Carline collisions | Handover collisions |
|---|---|---|
| Front-loaded · **Mon–Wed** | **0 / 71** | **0 / 7** |
| Mid-week · **Tue–Thu** (default) | **37 / 71** | **0 / 7** |
| Back-loaded · **Wed–Fri** | **71 / 71** | **7 / 7** |
So, concretely: at 3 travel days/week, if the block ever includes **Thursday you lose 37 school mornings**; add **Friday and it's all 71 mornings plus every one of the 7 Poplarville handovers**. Only a strictly Mon–Wed block keeps both at zero.
## What I added to dashboard.html
A new **"Schedule conflict"** section (financial dashboard untouched above it):
- **Weekday travel selector + presets** (Mon–Wed / Tue–Thu / Wed–Fri), defaulting to Tue–Wed–Thu — everything recomputes live.
- **KPI cards**: 71 carline mornings, 7 handovers, and live carline/handover collision counts with %.
- **Collisions-by-pattern table** (the three presets side by side).
- **Week-by-week matrix** — all 42 school weeks, Mon–Fri, each cell tagged C / H / T with collisions filled red (holiday days greyed, "marked-but-closed" days flagged `*`).
- **Conflict calendar** — all 10 months as mini-grids: carline shaded amber, handovers ringed, travel days blue, collisions red.
The assumptions panel documents the extraction rules and holiday removals. One honest flag carried in the UI: handovers were only circled Aug–Oct on your sheet, so the 7 is what's *marked* — if custody actually continues biweekly all year, the Friday exposure is larger, which the Wed–Fri preset already hints at.
From the taskInput files
IMG_4412_call_notes.jpg.jpg138 KB
IMG_0814_calendar.jpg.jpg194 KB
IMG_0819_travel_policy.jpg.jpg177 KB
From the taskLeg A prompt, turn 2, to Model A
Leg A turn 2
I dug up Jaylen's school calendar for this year and marked it up (see attached IMG_0814_calendar.jpg). My carline mornings are highlighted on it and the custody handovers with Terrance are circled, with the Poplarville details in the margin. Since the regional job runs 60% to 70% travel, about three days a week across the five sites, update dashboard.html with a schedule conflict calendar and a week-by-week matrix across the school year. Map the travel load against what is marked on that calendar, and tell me exactly how many school mornings and how many handovers end up with a travel collision.
Task fileIMG_0814_calendar.jpg
Look at The photographed calendar. The margin notes under August, September and October give the handover as 18:30.
IMG_0814_calendar.jpginput file
Click the image to see it at full size, and again to fit it back.
Task filedashboard.html
Look at The golden dashboard, which gives the handover as 13:30 to 15:30.
dashboard.htmlgolden artifact
Example 34The golden states something its own sources contradictFails the task
Creative & Media3 Leg A turnstask 6a99296813124786224a01ef
From the taskFrom the Leg B conversation, with Model B
Model B · Leg B turn 2
All four points are handled. Verified: **15 cards** now (both artist-account posts gone), all `@fretandfathom`, calls **KILL 4 · HOLD 3 · AMEND 4 · PUBLISH 4**, and the "P11" left in the code is only the one line in the notes explaining why the count dropped.
Here's what changed:
**1. Artist posts removed.** C16 and C17 (Sunny's own handle) are off the page entirely — not listed, not counted. Total drops 17 → 15.
**2. Un-killed the one that didn't trip its rule — R02.** Re-reading P1: it bans money *figures*. "Best quarter we've ever had" quotes no figure, so it doesn't land on P1, and the rest checks out (Sunny *was* in Room A through April; address/rooms are ours under P2). Now **PUBLISH** — and it's already live, which is fine.
**3. The cleared post that got the record wrong — R09.** The proof one-sheet put Ninebark at track 9; your own mixes-for-master email to Ade says *"Ninebark, track 7."* Moved to **AMEND** with the corrected value: Track 9 → **Track 7**.
**4. I did the checking instead of handing it back:**
- **R11 drummer:** not "confirm who played" — it's **Reggie Alcott** (session drummer, Room A tracking Feb–March). Corrected on the card. Cassidy Rhodes is a songwriter/guest.
- **R12 show night:** pulled the venue's run-of-show for Aug 14 — **Friday, on at 9:00 PM, 50-minute set** (doors 8 was right). The set length now reflects what happened on the night, not the 45–50 booking range. "Sold-out" is dropped — there's no settlement or attendance record showing a sell-out. (The stray "Tuesday" traces to a Ticketmaster confirmation that misprints the weekday.)
- **R18 how it was made:** specifics, not just "digital" — mixed in the box and delivered as **24-bit/48k WAV files** (11 mixes, −14 LUFS); the Otari MX-5050 is your bench restoration with parts still on order, never in the chain.
- **R20 "nine years":** verified, not flagged — your email to the broker says *"we are the tenant, nine years in the building."* Now a clean **PUBLISH** with no open question.
The HOLDs (R10 Wendell, R14 Ade, R15 clean test-pressing ref) still route to a named clearer — that's the rule doing the work, not me handing it back — and each already carries the corrected value (one episode/≤90s; −14 not −9, Ade not Teodora; recut pending). Everything you said read right is untouched.
One thing worth surfacing while I was in the records: the proof one-sheet is wrong on more than the track count — it also lists the single as Riverwide, when your sign-off email says **Longhand is the single**, and credits mastering to you when Ade did it. I used the delivered record over the proof throughout; it's all in the "How this was judged" panel at the bottom.
Model B · Leg B turn 5
Done. The two posts are separated and each judged on its own words-and-picture, and the shape check now appears on all seven picture-bearing cards. Verified: **16 posts, KILL 6 · HOLD 3 · AMEND 5 · PUBLISH 2**; the take-down banner is still just the two live ones.
**The two "nine years" posts are now separate cards, and they fail for different reasons:**
- **C12 — the room post** (*"Room A, 620 Main Street, East Nashville. Nine years of records made in here."*), still in the queue exactly as exported, TikTok, Oct 2. **Words: clean** — the room, address and nine years are all ours to say (P2), and nine years checks out. **Picture: the problem** — the file on it, room_a_wall.jpg, is actually the $214.86 earnings statement (P1). So it's a **KILL** purely on the picture; swap in a real Room A wall shot (tall, for TikTok) and the words are good. I flagged the wrong-shape too (it's landscape).
- **The landlord post** (*"…finally talking to the landlord about buying the place…"*), a **new** post that only exists in her live queue, Instagram, Sep 30. **Words: the problem** — renting and talk of buying the building is our position on it, which P2 keeps private, and that's the whole post, so there's no tidy of it. **Picture: none supplied.** Also a **KILL**, but on the words, not the image.
Same verdict, opposite fault — one killed by its picture, one by its words.
**On the shape checks:** every post that came with a picture now states the check explicitly on its card — R03 (wide, needs square), R10 (square, needs wide), R12 (landscape, needs square), R15 (landscape, needs square), R18 (portrait, needs square), C12 (landscape, needs tall), and R14 (loudness_card.png wasn't in the pack, so I said so rather than leaving it blank). The landlord post notes no art was supplied.
Nothing else was touched — the takedowns tonight are still **C11 ($9,500)** and **C07 (the comps giveaway)**.
From the taskLeg A prompt, turn 2, to Model A
Leg A turn 2
ok Marisol finally sent the artwork over, it's all in asset_pack/ - there's a shot of
her actual queue in there too. fold it into the same page, i think a couple of these
change the picture.
and can you tell me plainly which of the ones already out there need to come down
tonight, rather than sitting in a list with everything else
Task filesocial_audit.html
Look at The golden page. The header counts 16 posts and the page shows 16 cards, while the paragraph explaining the count still says 15.