Spike test — dosing plan and dilutions

How every dose in the spike test was derived, and what was actually poured. Four analytes in cumulative ladders: quinine sulfate, rhodamine, tryptophan, turbidity. The test closed 2026-09-01 19:55 local and the dataset is locked, so this page is now the record of the method rather than a sheet to dose from.

Closed 2026-09-01 19:55 local. 14 doses over 4.7 h. The fields below still recompute, which is deliberate: they are how the pours were derived, and anyone checking the arithmetic can change an input and watch it move. The rungs marked poured carry the time and amount from data/spike_events.json, which is the record. Results are on /spike.

Stage 0Design basis — measured, not assumed

Every step size below is set against the noise actually on the instrument today, so that no step is chosen to be smaller than the barrel can resolve. These were computed from the last 12 hours of the live spike archive on 2026-09-01, with the chiller running.

what	value	how it constrains the plan
Lume sample interval	1.0 min	Per-unit median. Upload is every 20 min, so the tail always trails, but the time resolution inside a step is one minute.
TLF amplitude noise	0.19 % CV	Within a 30 min window, median unit. Burn-in TLF batch 0.18 %.
FDOM amplitude noise	0.14 % CV	The quietest channel in the barrel.
Chl-a amplitude noise	57 % CV	On the 3 of 8 units that fit at all. The Chl-a batch has no usable clean-water reference level yet, which is the strongest single reason to run this test.
Net mon2, production combo	FDOM 1205 · TLF 455 · Chl-a 0	Counts, fleet median, LED 512 at bias ~3000. All eight Chl-a units read a net zero at that one combo in clean DI, so their response has to be read from the full sweep, never from mon2 at one combo.
ToF scatter	44 sps, 1.07 % CV	Clean-water floor for the turbidity proxy.
Sonde turbidity	0.49 to 0.67 NTU	Jamison 0.674, Marina 0.490. LakeviewEast reads a flat 0.000 with zero variance and should be checked before it is trusted as a reference.
Sonde FDOM / Chl-a	0.014 to 0.058 / 0.000 to 0.006 RFU	SD 0.009 or less. Relative units, so shape only, but the floor is low enough that any real dose is unmistakable.
Specific conductivity	7.9 to 8.0 µS/cm	SD 0.03 to 0.18. Near-DI. This is the independent check that a dose physically entered the barrel and mixed, for anything that carries an ion.
Water temperature	14.96 to 24.75 °C	Range over 12 h since the chiller went to 60 °F, with turning points about every 16 min. This sets the minimum step length.
TLF ppb scale	7.2 counts/ppb	Sensor 50066, 2026-07-16 serial dilution. Against the 455-count fleet baseline, 1 ppb is 1.58 % of signal, so a 3σ step is 0.4 ppb. That is the floor of the tryptophan ladder.
Formazin anchor	0 to 109 NTU, R² 0.990	Sensor 50031, 2025-12-15 bench cal. The burn-in page records that no formazin reference was ever run on this fleet, so the turbidity ladder here is the first one they get.

Stage 1What each analyte is for

Excitation and emission maxima below are literature values for the pure compound, given only to explain why each analyte was chosen. They are not a prediction of what our channels will do. Which channels respond, and how much, is exactly what the test measures, and a response that is not on the expected list is still recorded and shown.

analyte	nominal ex/em	what it probes	reference instrument
Tryptophan	280 / 350 nm	The target analyte for the TLF channel and the whole E. coli chain. Ties this barrel to the existing ppb scale.	None on the sonde. Anchored instead to the 2026-07-16 and 2026-08-06 ladders.
Quinine sulfate	350 / 450 nm	The conventional FDOM/CDOM standard: 1 ppb quinine sulfate is 1 QSU, so this is what puts an absolute scale under the FDOM batch.	Sonde FDOM Fluorescence, RFU. Relative, so shape and linearity only.
Rhodamine WT	530 / 555 nm	A strong fluorophore well to the red of both of the above. Two jobs: a specificity probe that separates selective channels from photon counters, and the only live check the Chl-a batch gets. Commercial fluorometers make the same connection: Turner Designs' adjustable solid secondary standard, P/N 8000-952 (Red), is specified for Chlorophyll, Rhodamine, Phycocyanin and Phycoerythrin channels, one red standard serving all of them. That makes rhodamine a stability check for a chlorophyll channel, not a concentration calibration for chlorophyll.	None. The sondes carry no rhodamine channel, so it is read only through FDOM RFU, Chl-a RFU, turbidity and the Lume channels.
StablCal, rung 1	800 NTU StablCal as supplied. 484 mL of the 500 you have.	800 NTU	rung 1, to 2.3 NTU	Formazin-based, and the same standard family as the 2025-12-15 bench calibration on 50031, so these rungs tie to the existing record.
AMCO Clear, rungs 2-3	1000 NTU AMCO Clear as supplied. 865 mL of the 875 you have.	1000 NTU	rungs 2-3, 2.3 to 7.4 NTU	A styrene-divinylbenzene bead standard, a different matrix from formazin. Run on top of the StablCal rungs it takes the ladder to 7.4 NTU, which is 9.9 sigma per unit and 54 sigma on a 30-unit median. It also gives something the fleet has never had: a per-NTU ToF slope measured on two different standard matrices in the same water on the same afternoon. Caveat to state with any comparison: the AMCO rungs sit higher on the ladder than the StablCal ones, so matrix and any nonlinearity are not fully separable.

Stage 2Order, and why it is not reversible

The barrel is not drained between analytes, so every dose is permanent and every later analyte is measured on a background containing all the earlier ones. That makes the order a real experimental choice rather than a convenience. The clean background is a finite resource, and it is spent on the channels that have no data yet, not on the one that already has two ladders behind it:

Quinine sulfate first. The FDOM batch has no absolute scale at all yet, so it gets the cleanest background of the four. It is also the quietest channel in the barrel at 0.14 % CV, which is only worth having if nothing is in the water yet.
Rhodamine second. A specificity probe is only clean while the background is. Reading it before tryptophan goes in means any TLF movement it causes is unambiguously the rhodamine.
Tryptophan second to last. The TLF ppb response is already characterized on this hardware, twice, so it does not need to spend the clean background. Running it late converts it from a repeat calibration into an interference test: does the known 7.2 counts/ppb response still hold with quinine and rhodamine already in the water. That is the question the existing data cannot answer.
Formazin last. Scattering perturbs every optical channel at once and cannot be undone. Running it on top of a known fluorophore load is also the most field-realistic case, because real water carries both.

No handling null was run. The plan called for a sham addition before the first dose: 250 mL of the barrel's own water, poured the way a dose is poured, to measure what the act of dosing does on its own. It was skipped by operator decision on 2026-09-01, on the grounds that the barrel is well mixed. Recorded here rather than dropped, because it changes how the smallest rungs can be read: 1 ppb quinine and 0.5 ppb tryptophan have no measured handling baseline to sit above, so a response at those two rungs rests on the barrel being undisturbed by a pour rather than on evidence that it is.

Stage 3Bench prep — the stocks as made up

Sized for 167.4 L and the ladders below, with headroom. Total analyte actually entering the barrel is tiny: 8.4 mg of tryptophan, 24.9 mg of quinine sulfate, 16.7 mg of rhodamine. Everything else is dilution so that each rung is a pour rather than a pipette.

stock	make it	strength	ladder uses	why this strength
Quinine primary	500 mg quinine sulfate into 100 mL of the 0.1 N sulfuric acid. Confirmed dissolved 2026-09-01.	5000 mg/L	5 mL	Acid into the barrel is set here and nowhere else. At this strength the whole ladder carries 5 mL of acid, which is pH 5.5 and +1.3 µS/cm. That is why no acid blank rung is needed.
Quinine working	6 mL of the primary + 300 mL DI, 306 mL total.	98.04 mg/L	254 mL	Diluting with DI does not change the acid per milligram of quinine, so this step costs nothing and makes the low rungs pourable.
Rhodamine working	1.0 mL of the 20 % stock into 1 L of DI.	200 mg/L	83 mL	Neat 20 % is 200,000 mg/L. Poured directly the first rung would be 4 µL, which is not a dose, it is a rounding error.
Tryptophan working	50 mg into 1 L of DI.	50 mg/L	167 mL	At 100 mg/L the first rung is 833 µL. Halving the stock puts it at 1.7 mL and every rung becomes a pour.
Formazin	Use the bottle as supplied.	4000 NTU	4.3 L	To 20 NTU costs 0.84 L, to 50 NTU 2.11 L, to 100 NTU 4.27 L. Measure what you have and cap the ladder rather than running out partway up a rung.

Barrel working volume

First dose at (local, MT)

Used to build the clock and the event log below.

Minutes per rung

Mixing time before the window opens

Written into each event as settleMin. Measure it on the first dose from the conductivity trace and correct it.

Carry-forward at the end of the run

Each analyte is diluted by every addition that follows it. This is what is actually in the barrel when the last rung is reached.

Stage 5Timing, as run

The ladders were cut on 2026-09-01 to shorten the run. Tryptophan went from five rungs to three, set by the operator at 1, 10 and 50 ppb; turbidity from six to three at 2.3, 5.0 and 7.4 NTU. The turbidity rungs dropped were 0.6 and 1.4 NTU, which sat under 2 sigma per unit and inside the barrel's own 0.7 NTU wandering. The tryptophan rungs span 50x and straddle the 7.9 ppb that corresponds to the 126 CFU/100 mL criterion, with 1 ppb below it and 10 ppb just above.
Rungs are 20 min, set by the operator on 2026-09-01. At 1 sample per minute and a 10 min settle that is about 10 usable readings per rung. The units also sync every 20 min, so one rung is exactly one upload interval: a rung's data lands in a single burst after it closes, which is why each verdict arrives about a rung behind the pour. Water temperature is swinging across 14.96 to 24.75 °C with turning points every 16 min, so a 20 min rung spans roughly two thirds of one chiller cycle and its mean carries a temperature phase rather than averaging over it. Read these rungs on the temperature-corrected series, not the raw one, where the quench is removed per reading instead of relied on to cancel across the window. The uncorrected panels show the chiller, not the dose.
Do not touch the chiller setpoint during the test. The quench correction is fitted on the pre-dose window and extrapolated forward. Moving the setpoint mid-run moves the thing the correction is anchored to.
Leave 40 minutes undisturbed before the first dose and after the last, so both ends of the ladder have a clean flank to be judged against.
Watch specific conductivity to confirm mixing. The barrel sits at 7.9 µS/cm with an SD under 0.2, so any ionic addition shows up immediately. The time from pour to a flat conductivity plateau is the real mixing time, and it should replace the assumed settleMin after the first dose.
The amount actually added was logged, not the amount planned. Three rungs differed from the sheet: rhodamine rung 3 was 70.0 mL rather than 67.1, and both AMCO Clear rungs were rounded to 460 and 375 mL. The event log carries what was poured. It is the record of the experiment.

Six units were already down before the window opened and two more dropped out on 09-01. Check the report-age cards at the top of /spike immediately before the first dose. A unit that is silent through a rung cannot be recovered later, because the barrel cannot be returned to that concentration.

Stage 6The event log

Paste these into the events array of data/spike_events.json and redeploy. Nothing on /spike is hard-coded, so the dose lines, the response windows and the specificity matrix all follow from this file. Correct the times and amounts to what actually happened before pasting.

Stage 7Still open after the run

The turbidity ladder reaches 7.4 NTU on two standards. 500 mL of 800 NTU StablCal carries rungs 1 to 3, then 875 mL of 1000 NTU AMCO Clear carries 4 to 6. The top rung is 9.9 sigma on a single unit against the measured 1.07 % ToF noise and about 54 sigma on a 30-unit median, so this fleet finally gets a real formazin anchor. Two things it still cannot do: reach the tens of NTU where field turbidity actually lives, and separate the StablCal and AMCO matrices cleanly, because the two standards occupy different parts of the same cumulative ladder.
The Chl-a batch has no chlorophyll-a analyte in this experiment. Tryptophan targets TLF, quinine targets FDOM, formazin is a scatterer. Rhodamine is the closest thing to a control it gets, and on commercial fluorometers one red solid secondary standard does serve the chlorophyll and rhodamine channels together, so it is a fair liveness check. It is not a chlorophyll concentration standard. That matters because the Chl-a batch also reads a net zero at the production combo, so a flat Chl-a line is exactly what a dead channel and a perfectly specific channel both look like, and this run cannot separate them. Its flatness through the quinine ladder, 0.003 %/ppb against FDOM's 0.60, is a genuine specificity result and is not evidence the channel works. Rhodamine is the only dose in the set with any chance of exciting it, so that rung doubles as the batch's only live check. A real answer needs a chlorophyll-a active dose, a dilute algal suspension or a chlorophyll standard, on a later run.