The Day
One technician, one Club, one trip, at the Sowing stage. No signal from the moment they leave the house until the moment they get back to a road head. The reconciliation argument is worth nothing if the round does not close inside a working day, so this tab puts the day on a clock and shows the arithmetic instead of asserting it.
What is real here, and what is not. The stop order, the leg distances and the arrival clock are computed by the routing engine at render. The per-stop facts are the fixture's own records: measured area, walk duration, fix quality, satellite count, vertex count, method. The tolerance band each variance is judged against is computed from plot size, device class and capture method at render. Battery is the one stated model on the page and every constant in it is exposed. The world itself is synthetic and every GNSS capture in it is simulated. SYNTHETIC DATA. Every company, village, club, farmer, plot boundary, capture, contract, rate and settlement in this file is fabricated for a product study. No real person, holding or agreement is represented. District and state names are real Indian administrative names used only to anchor the belt; village centroids are approximate synthetic points inside it and do not correspond to any actual settlement.
Hand-rolled SVG from each plot's own ring in local metres, north up. Click a plot to jump the timeline to that farmer. Stop order and leg distances come from planRoute at render: nearest neighbour, then 3 improving 2-opt swaps.
By sequence number, never by timestamp. Field handset clocks drift, and a log that reorders itself is a log that cannot be replayed.
The same instant on the handset. The full flow lives on the Field App tab; this is the one screen that is in front of the technician right now.
The queue screen from last night, still showing whatever did not go up. One tap on Sync, then a list draining top to bottom.
Drains the write-ahead queue in seq order. Each op carries its own id, so a batch that half-uploaded last night is deduplicated rather than written twice. Mean session for this handset: 275 s across 14 completed sessions in the record.
The drain reports source "offline" because the upload never completed. Nothing is lost, but the technician leaves with yesterday still on the phone.
Every op rolls back to pending and its attempt counter increments. The day proceeds; the backlog is a supervisor signal, not a blocker.
Loading the capture records. The route, the clock and the budget are already complete because they come from world.json; walk duration, fix quality, vertex count and method arrive with the separate capture chunk.
| Line | Minutes | Where the number comes from |
|---|---|---|
| Morning sync at home | 4.6 | Mean of this handset's 14 completed sync sessions |
| Pre-load and route check | 12.0 | Stated assumption |
| Route computation | 0.2 | Engine, at render |
| Travel, village centre and 21 legs | 13.1 | planRoute, 2,619 m at 12 km/h |
| Capture and conversation, 21 stops | 335.3 | serviceMinutes: fixed admin plus the boundary walk, weighted for Sowing |
| Lunch | 35.0 | Stated assumption |
| Sync attempt at the road head | 5.2 | Fixture session SYN-086, 310 s on 2G, retried |
| Evening drain, after hours | 3.2 | Fixture session SYN-083, 192 s |
| Handover and notes | 5.0 | Stated assumption |
| Planned day | 413.6 | against a 480 minute day, leaving +66 min |
A stop has 19.1 minutes and the plan needs 16.0, so the round closes with 1h 06m of slack. That slack is not comfort. It is exactly what the day's exceptions eat. A re-walk under the tree line costs a stop's worth of minutes on its own, and the capture records for this club are still loading. The number that moves this most is not effort. It is plot size: the boundary walk is proportional to perimeter, so the anchor holding in this club costs several times what a half-acre plot costs.
planRoute: 4,393 m in roster order against 2,619 m after 3 2-opt swaps, at 12 km/h.
Limit. Great-circle legs. There is no road graph in this model, and inventing one would be a fiction. The ORDER is robust to that; the absolute metres are not.
Zero network operations between the pre-load at 05:35 and the road head at the end of the round. The queue is where a capture becomes durable, so the confirm screen returns in the time a local write takes rather than in the time a 2G round trip would take if there were one.
Limit. The minutes saved cannot be measured from this fixture, because no online-required variant of the day exists in it. What the fixture does show is the cost of assuming connectivity: 43 captures across the fleet are still on handsets, the oldest for over a month.
Buys. The boundary to walk against, the contracted area, the survey number, the crop mix.
Cost of leaving it out. Nothing to capture against. The trip cannot happen.
Buys. Names, tenure, which plots belong to whom, who the club lead is.
Cost of leaving it out. The technician cannot tell which plot in the block is whose.
Buys. Which vertices two neighbours hold in common. It is the constraint that makes six captures worth more than six times one capture, and it is what lets an overlap be caught in the field rather than in a settlement meeting.
Cost of leaving it out. Overlaps and gaps are only found back at the office, after the technician has left the village and the farmer has gone home.
Buys. Last stage's boundary drawn under the live trace, so a technician walking the wrong field sees it immediately.
Why not. Stage-jump errors survive the whole season. Six of them are already in this world.
Buys. Stage definitions, crop rates, the handset's own sigma, the local origin.
Cost of leaving it out. No tolerance band can be computed on the device.
Buys. Nothing this stage needs. At 10 m a sub-half-hectare boundary is not separable, so a basemap cannot verify what the walk measures.
Why not. Deliberate. Shipping tiles would add tens of megabytes to buy a picture that cannot settle a boundary question.
The whole club, roster included, is smaller than a single photograph. That is the argument for pre-loading everything rather than paging it: there is no bandwidth to save that is worth the risk of a technician standing in a field without the record they need. The one line deliberately left at zero is satellite imagery.
At 55% health the round never gets near the floor. Drag the slider down: the day starts failing somewhere in the fifties, which is where a three-year-old cell sits. That makes the low-battery risk a handset-refresh question, not a field-discipline one, and it points at the same conclusion as the accuracy argument.
Every constant in this model
| Stop | Farmer | What happened | Contracted | Measured | Difference | Band | Unexplained |
|---|---|---|---|---|---|---|---|
| 19 | Srisailam C.310/3 | Measured under contractRe-walk once at the next visit. If the second walk agrees, this is a contract question for the supervisor, not a field dispute with the farmer. | 2.625 | 1.927 | -26.6% | ±7.6% | Rs 29,954 |
| 12 | Odelu R.69/9 | Measured under contractRe-walk once at the next visit. If the second walk agrees, this is a contract question for the supervisor, not a field dispute with the farmer. | 3.550 | 2.921 | -17.7% | ±6.1% | Rs 24,816 |
| 5 | Nagamani M.225/7 | Sown area over contractIssue inputs against the contracted area and open a contract amendment for the excess. Do not top up in the field. | 4.450 | 5.042 | +13.3% | ±5.1% | Rs 21,811 |
| 2 | Latha G.385/8 | Measured under contractRe-walk once at the next visit. If the second walk agrees, this is a contract question for the supervisor, not a field dispute with the farmer. | 4.125 | 3.567 | -13.5% | ±5.4% | Rs 20,041 |
| 14 | Lakshmi D.246/3 | Measured under contractRe-walk once at the next visit. If the second walk agrees, this is a contract question for the supervisor, not a field dispute with the farmer. | 1.050 | 0.729 | -30.6% | ±14.8% | Rs 9,947 |
| 20 | Pochamma L.465/8 | Measured under contractRe-walk once at the next visit. If the second walk agrees, this is a contract question for the supervisor, not a field dispute with the farmer. | 0.525 | 0.268 | -49.0% | ±24.3% | Rs 7,765 |
| 6 | Devender G.473/8 | Measured under contractRe-walk once at the next visit. If the second walk agrees, this is a contract question for the supervisor, not a field dispute with the farmer. | 0.650 | 0.387 | -40.5% | ±20.9% | Rs 7,643 |
| 7 | Srinivas L.274/4 | Measured under contractRe-walk once at the next visit. If the second walk agrees, this is a contract question for the supervisor, not a field dispute with the farmer. | 9.875 | 9.536 | -3.4% | ±2.8% | Rs 3,471 |
| 8 | Srisailam P.256/2 | No Sowing captureSchedule a revisit inside the sowing window. Without this stage the plot reconciles from five captures instead of six and its settlement band stays wider all season. | 0.475 | - | - | ±26.1% | - |
| 10 | Renuka Y.351/5 | No Sowing captureSchedule a revisit inside the sowing window. Without this stage the plot reconciles from five captures instead of six and its settlement band stays wider all season. | 0.675 | - | - | ±20.3% | - |
| 17 | Praveen L.121/2 | No Sowing captureSchedule a revisit inside the sowing window. Without this stage the plot reconciles from five captures instead of six and its settlement band stays wider all season. | 0.875 | - | - | ±16.8% | - |
| 18 | Mahesh T.361/9 | No Sowing captureSchedule a revisit inside the sowing window. Without this stage the plot reconciles from five captures instead of six and its settlement band stays wider all season. | 4.675 | - | - | ±5.0% | - |
The band column is the whole argument in one place. Judge these variances against zero and every row looks like a dispute; judge them against what a 5.0 m single-frequency fix can resolve on a plot that size and 9 of 21 stops close themselves.
The counter-intuitive part. A better handset raises more exceptions here, not fewer. Switch to the club on the dual-frequency phone and the band collapses to about a percent on a large holding, so differences that a 5 m fix could honestly blame on itself move into the column marked “the contract is wrong”. That is the instrument doing its job: it does not create disputes, it stops hiding them. It is also why device class is a first-class data attribute in this model rather than a procurement footnote.
Limit. The band is computed from the engine's device-class sigma, not from each capture's own sigma, which runs higher under canopy and a poor sky. On a weak-fix capture the band shown here is therefore slightly tighter than the truth, which makes this list marginally over-inclusive rather than under-inclusive. That is the right direction for an error to run when money is downstream of it.
What the round produces is not a folder of boundaries. It is one more weighted observation of every plot in the club, plus the shared vertices that tie them to each other. Reconciliation is where those observations become a single boundary with an interval on it, and Offline is the queue that gets them there without a network.