Repository navigation
Releases
0.44.0: who makes what in a kit manual
the guide now says who makes a kit's reagents; tested on an eighteenth set from outside: eight manuals from five makers (promega, thermo fisher, cell biolabs, biovision, sciencell), each drafted five ways by five agents at once
- result: 38 of 40 drafts reach a rehearsed protocol, 37 at the first attempt, and the 38 do the manual's assay; 2 were handed back to a person (a 100-assay kit asked to fill 162 wells)
- guide: what the reagent preparation makes is a person's, before the run; standards, sample dilutions and what the numbered steps make are the robot's
- guide, after the write-up: a mix a numbered step makes is the robot's also when the text wants it fresh; a kit's contents list is not what stands on the deck, and a run that needs more than one kit says how many; wells asked for only under a condition are left out; a sample range made up to a total is half the total; an amount too small to pipette is diluted on the deck first
- after the fixes, the five manuals they touched were drafted again all five ways (not an outside test): 25 of 25 pass at the first attempt
- my mistake in this set: two sheets were registered with one page of two; the ten drafts from the cut text are void and were redone from the whole text
- totals: 232 of 267 drafts from 108 manuals stand as rehearsed programs with no hand edits; 218 tests
- caveats: five agents agreeing is one reading of one text, not five tests; nobody who has run these kits has read the drafts; details in evals/RESULTS.md
0.43.0: eight manuals drafted five ways
the widest test so far: eight kit manuals of different kinds, each drafted for the Flex, the OT-2 and the Hamilton with nothing decided, and for the Hamilton and the Flex with 24 samples in triplicate; forty drafts, by five agents at once
- result: all 40 reach a rehearsed protocol, 34 at the first attempt; 31 do the manual's assay; 9 did only its first half, because a rule of the guide cut two assays in two
- guide: alternatives (plate reader or flow cytometry) take the first; stages of one assay (a reaction and the ELISA that reads it) are written in full; culture over days happens before the run; a timed incubation at 37 C is a pause, however short
- guide: how much a deck holds (19 plates and troughs on the Hamilton, 9 on an Opentrons deck, each checked against its planner); when the samples need more, the file is written for as many whole plates as fit and says how many runs that makes
- check: a step that empties a range passes over wells that hold nothing, as long as another well of the step holds some; that one case cost six second rounds
- after the fixes, the three kits they touched were drafted again all five ways (not an outside test): 15 of 15 pass at the first attempt, and all write the whole assay
- totals: 194 of 227 drafts from 100 manuals stand as rehearsed programs with no hand edits; 217 tests
- caveats: one maker; five agents agreeing is one reading of one text, not five tests; nobody who has run these kits has read the drafts; details in evals/RESULTS.md
0.42.0: one step over several ranges or plates
wherever a protocol asks for wells it now takes a list of wells and ranges, in the order written: [plate_1/A1:D6, plate_2/A1:D2]; a wash of three plates is one step, and 8 wells of a 12-well plate are one line
- format: lists of wells for liquids, transfers and mixes; no well twice in a list; a dilution series still starts from one range
- examples:
cell_staining.yaml, fixing and staining cells on two 24-well plates; plans for all three robots - intake: a rehearsal problem names the step of the file it is about, and for liquid that is thrown away it offers a lower draw height; the Hamilton's deck refusal says when the plates alone fill the deck
- measured on a sixteenth set (four staining kits from ScienCell, Cosmo Bio and Wako): 4 of 8 drafts reach a rehearsed protocol (Flex 3 of 4, Hamilton 1 of 4); every draft that needed a list wrote one step over it
- the four handed back: two because 72 wells on 6- and 24-well plates do not fit a STARlet deck (my test sentence, not the format), two because a 1 mm layer of stain could not be taken off from the surface and the feedback did not say how; with the new feedback that kit passes on both robots in the second round
- totals: 154 of 187 drafts from 92 manuals stand as rehearsed programs with no hand edits; 215 tests
- details and caveats in evals/RESULTS.md
0.41.0: defaults for small plates, measured on five staining kits
the guide's defaults were written for 96 wells; three of them now cover plates with fewer than eight rows, and a fifteenth set tests them on kits from five makers
- guide: 8 samples go from A1 down each column in turn, or as many as the plate holds; a choice the text leaves between two ("in duplicate or triplicate") takes the first; liquid already in a well is what the text gives per well, else 500 uL, 1 mL or 2 mL by plate size; a volume worked out from a ratio is listed under
chose - measured on a fifteenth set (senescence and Oil Red O staining kits from Cayman, BioVision, Cell Biolabs, Abcam and Cosmo Bio, on 6-, 12- and 24-well plates): 7 of 10 drafts reach a rehearsed protocol on the Flex and the Hamilton, 5 at the first attempt; 3 were handed back to a person
- one of those three and both second rounds were the guide's, and are fixed: how much a text says to make or have ready ("prepare 3 ml per well") is not the amount on the deck, and a pause is under 200 characters; the two Flex drafts pass at the first attempt when done again
- totals: 150 of 179 drafts from 88 manuals stand as rehearsed programs with no hand edits; 209 tests
- caveats: staining kits are one kind of protocol; 8 wells on a 12-well plate are two ranges, so every step is written twice (next); details in evals/RESULTS.md
0.40.0: small plates on an Opentrons deck
6-, 12-, 24- and 48-well plates can now stand on a Flex or an OT-2; they were left out because Opentrons marks their spacing as irregular, but they are pipetted one well at a time, where spacing does not matter
- opentrons: five plates with fewer than eight rows are on the list (51 plates and reservoirs now)
- the four corning costar plates are the same product as the Hamilton's, so one file that names them plans for all three robots; a test compares every well
- measured on a fourteenth set of one (oxidized LDL uptake, microscopy protocol on a 24-well plate): passes at the first attempt on the Flex, the OT-2 and the Hamilton; the apoptosis kit drafted again for the Opentrons robots now uses a 24-well plate, not a 96-well one
- totals: 143 of 169 drafts from 83 manuals stand as rehearsed programs with no hand edits; 208 tests
- caveat: one manual; 6-, 12- and 48-well plates were only run from a hand-written file; details in evals/RESULTS.md
0.39.0: take all of it
a transfer to waste can now say volume: all; steadyhand works out what each well holds when the step comes, so nobody (and no model) does the arithmetic for a removal; a well that holds nothing by then is reported with the step named
- format:
volume: allon a transferto: waste; refused for liquid that is moved on - guide: removals are written with it; drawn from the surface over cells or a pellet, else from 0.5 mm
- measured on a thirteenth set (4 cayman manuals that empty wells again and again): 8 of 8 drafts reach a rehearsed protocol at the first attempt on the Flex and the Hamilton; all 46 removals use
volume: all - totals: 140 of 166 drafts from 82 manuals stand as rehearsed programs with no hand edits
- caveat:
allis what the model of the well holds, not a sensed level; 4 manuals from one maker; details in evals/RESULTS.md
0.38.0: drafting for the OT-2's two pipettes, and cells put in a new medium
the OT-2 guide now says the robot holds two pipettes, with ranges taken from the planner's own table, so a draft keeps its mixes within what two can do; a file that still needs three is told which step ties it to each; cells put "in" a new medium have the old one taken off first, also where the text does not say so; measured on seven fresh manuals: on the first five, nine of ten drafts rehearse (Hamilton 5 of 5, OT-2 4 of 5); the tenth was refused three times because the first version of the pipette note gave a wrong range; with that corrected, two more fresh manuals and both redone ones pass at the first attempt; details in evals/RESULTS.md
0.37.1: a removal split into small draws is judged like one draw
on the OT-2 a single well is emptied by the 20 uL pipette in parts; the last part found the liquid already below the tip and failed as drawing air, where one draw of the same amount is only noted as leaving a little behind; both are noted now; found by carrying a Flex draft from the tenth set over to the OT-2
0.37.0: five assays on cells, read against their manuals
the rule behind the two wrong passes of the last set (wells the text empties are not treated as empty) is in this release, and was tried on a tenth set: five fresh manuals, all assays on cells in a plate; ten of ten drafts rehearse, eight at the first attempt; nine do what the manual asks; one adds a glucose-free treatment on top of the old medium, where the text only implies taking it off, and says so; also found: on the OT-2 a removal split into 20 uL draws fails where the same removal in one draw is only noted; details in evals/RESULTS.md
0.36.0: the rules the drafting kept tripping on, said once each
shaking and timed mixing are one rule now: a short shake is a pipette mix, also as the last step, and a long one is an incubation; a liquid no trough holds is split into named parts; every time on the magnet gets its wait, and a draw that empties a well goes down to 0.5 mm; a premix you make yourself is not made by the robot; warming to room temperature happens before the run; measured on a ninth set of five fresh manuals: ten of ten drafts rehearse at the first attempt, eight match their pages, and two add to wells the manual says to empty first (they say so in their notes; that rule is next); details in evals/RESULTS.md