assets/evals/fixtures/ethnographic/gold.md
Gold Mapping — Ethnographic/Contextual Lens (in-situ)
Observation: senior picker at an auto-parts online store warehouse, picking order #48213 (three line items: bearings, brake pads, oil filter) in the WarehousePro WMS system. One-off session, the respondent herself flags the atypicality of the day (warehouse after an inventory count, line 4) — an open question for high-stakes decisions: whether further observation is needed.
Layer 1 — Operational
-
N1 Environment and artifacts — picking terminal, monitor with WarehousePro WMS, a hand-drawn grid-paper notebook with her own warehouse map, sticky notes on the monitor. Quote: "here's my terminal, I scan the bin barcode, then the item barcode" (line 6); observation: "on the desk a battered grid-paper notebook and sticky notes on the monitor" (line 5).
-
N2 Observed sequence of actions — opened the picking task for order 48213 (line 8), scanned bin A-12-3 (line 9), picked the bearings (line 23), went to pick the pads, found a count discrepancy at V-4 (line 32), located the item on a pallet via the chat (line 41), picked the filter in the fast-pick zone without scanning the bin (lines 52—57), closed the box and stuck on her own label (lines 63—64). Quote: "I'm opening the picking task, here, order number 48213, three line items" (line 8).
-
N3 Homegrown adaptations (workarounds) — at least three:
- hand-drawn warehouse map notebook layered over the system's bin addresses — "here's my own map of the warehouse, I drew it myself two years ago and touch it up in pencil whenever stock gets moved around" (line 16);
- reminder sticky note about the terminal freezing — "don't scan the box twice!! froze on 3/14" (line 19);
- her own handwritten sticker over the box barcode — "the printer prints it so small and crooked that the packers sometimes mix up orders... so I write it big by hand" (line 66).
-
N4 Think-aloud commentary during action — tied to a specific step in N2, prompted by the observer's contextual-inquiry question "What are you doing right now?" (line 7), answer: "I'm opening the picking task, here, order number 48213, three line items" (line 8); also at the fast-pick-zone step the observer asks "I noticed you didn't scan the bin... is that on purpose?" (line 53), answer: "the fast-pick zone doesn't require scanning the bin... I grab it and scan the item right away instead of the bin, it's faster that way" (line 54).
-
N5 Says-vs-does gap — at least two cases:
- says (line 10): "Normally I do everything strictly by the system, it tells me the route through the warehouse, I walk it and pick, no improvising" — does (line 11, observation): "without waiting for the terminal to display the route, the respondent immediately turns toward rack B-7".
- says (line 28): "Normally I try to stick strictly to the order route and not get distracted, we're strict about that" — does (line 29, observation): "the respondent was twice sidetracked into unrelated actions outside the order route" (see also lines 25/27 — straightening other people's boxes along the way).
-
N6 Interruptions and switches — a coworker sends a voice message in the chat about the count discrepancy (line 39, observation: "a coworker replies in the chat with a voice message, the respondent listens through an earbud, nods"); coworker Sasha comes over twice with a question about a neighboring rack (line 48, observation: "a coworker approached the respondent twice with a question about a neighboring rack, the respondent answered briefly and kept picking").
-
N7 Social context — the shift's group chat as a collective resource for locating items: "someone on shift usually knows where the item actually is, and they tip me off quickly" (line 37); coworker Sasha personally comes over for help — "Sasha just came by to ask where to find parts for a Daewoo, it's like that here all the time, we cover for each other" (line 49); the supervisor is mentioned as a figure of social control — "I'm not sure he'd officially approve of it, we just don't talk about it out loud" (line 59).
Layer 2 — Analytical
-
A1 Adaptation type (N3) — (unstable cell)
- Warehouse map notebook (N3.1) — PROCESS-FAILURE WORKAROUND: compensates for the mismatch between the system's bin addresses and the item's actual location. Basis: "in the system the bin addresses periodically don't match what's really on the shelf" (line 14).
- Terminal-freeze sticky note (N3.2) — PROCESS/SYSTEM-FAILURE WORKAROUND: compensates for a known terminal bug. Basis: "if you scan the same box twice in a row, the terminal freezes" (line 21).
- Handwritten box sticker (N3.3) — ambiguous: can be read either as a workaround for a label printing failure (the printer prints small and crooked), or as an excessive precaution/ entrenched shift habit ("all of us senior pickers do this... it just sort of caught on" — line 67). Cannot be classified definitively without data on the packers' error rate — flagged as an open question.
-
A2 Significance of the says↔does gap (N5) — (assumption, not fact) For the route case (lines 10—11), it looks more like the respondent genuinely doesn't notice the gap: the phrase "no improvising" is delivered confidently and without hesitation right before the observer recorded the opposite, which points more toward an automated habit that the respondent herself doesn't flag as a deviation. For the fast-pick-zone case (lines 55—56, 61), the gap is, by contrast, conscious and socially calibrated: the respondent directly admits "with the supervisor around I'd probably scan the bin just to keep things proper, but without him not always" (line 61) — this is closer to a deliberate adjustment for the observer/authority figure than to an unconscious discrepancy. The distinction between the two cases isn't always sharp given the context — keeping this as a flagged assumption, not a fact.
-
A3 Process fragility — the picking process visibly runs on homegrown adaptations: the respondent herself assesses the dependency directly — "if you took my notebook away and stripped off the stickers, I'd probably work way slower for the first week, until I relearned where everything is" (line 69); observation confirms that over the session she "consulted the notebook seven times and the monitor stickers twice, never once opened the official picking instructions" (line 70). Replacing the respondent or removing her personal artifacts is a risk of a sharp drop in picking speed.
-
A4 Design implication — at least three mismatches between the official process and actual use: 1) bin address records in the WMS diverge from the item's actual location (lines 14, 32—33) — address updates aren't keeping pace with restocking moves; 2) the official path for fixing discrepancies (a report through the inventory clerk) is too slow relative to the pace of work — "Through the report that's about a two-hour wait, through the chat it's usually five minutes" (line 38) — the shift's informal chat effectively replaces the accounting procedure; 3) the no-exceptions "bin → item" scanning procedure doesn't account for the fast-pick zone, where scanning the bin is physically redundant (lines 54—56).
-
A5 Key insight — In actual practice, the picker runs picking off her own hand-drawn warehouse map and the shift's informal chat instead of the formal route and the WMS discrepancy report, because the system's bin addresses regularly diverge from the item's actual location, and the official procedure for correcting discrepancies (a report through the clerk) is far slower than what's needed to hit the picking quota; the systemic consequence is that WMS records chronically lag reality ("we find about ten of these discrepancies per shift", line 44), and the resilience of the entire picking process effectively rests on one employee's personal artifacts and informal channels rather than on the system itself.
Limitation (Hawthorne / single observation)
The respondent herself directly names the observer effect: "I'm working more carefully than usual today, because you're standing right here, normally I'd probably cut a few corners that I felt too embarrassed to show just now" (line 74), and gives a concrete example of normally hidden behavior — picking two orders in parallel instead of one (line 77), which was not observed live at the time of the conversation (observation, line 78: "the respondent is standing with one box in hand, no second box visible nearby" — meaning the example was given retrospectively, not shown in action). This confirms that a single observation session systematically underestimates the real scale of deviations from procedure — an open question for any high-stakes decision based on this observation.