Cooked Fruit Fundamentals¶

Version 5 · Disclaimer

Related: Maillard Skill user guide · Maillard preparations (outputs)

Companion reading for the fruit-maillard-prep skill, aligned with skill v5 (September 2026). Written to be read on its own: mechanism first, then the controls, then flavour, then the method, then the beer-side consequences and the hazards. Every theory section ends with a bolded conclusion so it can be scanned later.

Figures are literature typicals and stoichiometric estimates, not measurements. Confidence is tagged where a claim is load-bearing: High / Med / Low. Low means verify before spending ingredients on it. Where the skill and this primer disagree, the skill's reference files win; this document explains them.

1. What cooking fruit does that raw fruit cannot¶

Raw fruit in a lambic gives its own aroma, its acid, its sugar, its tannin and a restart of fermentation. Cooking fruit with a nitrogen source generates a second flavour vocabulary the fruit never contained: cocoa, roast, nut, bread crust, caramel, honey, marzipan, savoury depth. That vocabulary is made in the jar, not extracted from the fruit.

Three chemistries run at once and respond to different controls. Keep them separate.

The first is the one you can steer precisely. The third is why real fruit outperforms sugar plus amino acids from a tub, and it is the source of most of the complexity a textbook Maillard would not predict.

Conclusion: a prep is three reactions in one jar; design each on purpose, and separate them into stages when they want different conditions.

2. Maillard in one page, and the open question underneath the method¶

Early stage. A reducing sugar's carbonyl condenses with a free amino group to give an Amadori compound (from aldoses) or a Heyns compound (from ketoses). Colourless, odourless. Substrate loading.

Intermediate stage. Amadori compounds fragment to α-dicarbonyls: 1- and 3-deoxyosones, methylglyoxal, glyoxal, diacetyl. This dicarbonyl pool is the shared hub. Three exits:

• Strecker degradation: dicarbonyl + amino acid → aldehyde one carbon shorter than the amino acid, plus an aminoketone and CO₂. The aldehyde is the amino acid's signature (§3.2).
• Aminoketone condensation → alkylpyrazines. Nutty, roast, cocoa. Two amino acids per pyrazine.
• Cyclisation and dehydration → furans, furanones, pyranones, pyrroles. Caramel, jam, bread.

Late stage. Aldol condensation and polymerisation → melanoidins. Brown, non-volatile, non-fermentable, 1 to >100 kDa. Colour, body, mild bitterness, metal chelation. Not the aroma.

Two consequences follow, and one of them is less settled than it looks.

1. The aroma lives in the intermediate stage. Once the free amine pool is exhausted, additional cooking makes melanoidin and caramel but no more pyrazine or Strecker aldehyde. Fruit is almost pure sugar and almost devoid of free amino acid; a fruit-only cook runs at 30–40:1 sugar:amine molar. (High)
2. Whether nitrogen or temperature is the ceiling at 121 °C is not settled. Alkylpyrazine formation has a high activation energy, and most of the literature that produces meaningful alkylpyrazine concentrations sits at 150–180 °C. The method assumes that loading amine and holding 90–120 min at 121 °C is enough; it may be that time and temperature are co-limiting after the amine is loaded. Batch 001, once read as proof that nitrogen was the ceiling, ran at ~30 °Brix and pH ~4.7 with heavy polyphenol suppression, and its "dark chocolate" result is at least as consistent with furanones, caramel and aronia phenolics as with pyrazines. (Med.) The skill's top experimental priority is a paired jar with and without a 135–140 °C covered finish (§19).

The dicarbonyl hub is also contested: flavanol polyphenols and sulphite trap dicarbonyls and starve both the Strecker and pyrazine exits at once (§3.7).

Conclusion: load free amine, protect the dicarbonyl pool, hold the jar hot and dry, and run the roast-finish experiment before assuming nitrogen alone was the limit.

3. The controls, in order of leverage¶

3.1 Nitrogen¶

The amino group is the scarce reagent. Sources differ in how much of the nitrogen is free (reactive now), which amino acids it carries (which aldehydes, which fusel and THP risks), and what it costs on the beer side.

Why collagen fell out of the method. A 2–5 kDa hydrolysate is 22–55 residues per peptide, and a peptide has exactly one free α-amino group however long it is. Earlier versions credited a third of the residues as reactive, which would require an average peptide of ~270 Da; the arithmetic could not be true. Peptide bonds do not hydrolyse usefully at 121 °C and pH 6 (autoclaved peptone stays peptide), and a 60–90 min papain step reaches perhaps 20 % degree of hydrolysis. So the case for collagen ("bread-crust pyrroles from proline, savoury glutamate") rests on residues that are mostly never exposed, while its methionine caps the dose and its lysine and arginine arrive in the beer peptide-bound for Brett to release over months. (High on the peptide arithmetic; Med-low on the hydrolysis yield.)

Dose nitrogen against reducing sugar, not fruit mass. 1 g hexose ≈ 5.55 mmol. Target a reducing sugar : reactive amine ratio of 10–14:1 molar. The ratio rule exists for two reasons that pull the same way: in a collagen-bearing jar, sugar in excess out-competes lysine and arginine; in any jar, amine beyond what the dicarbonyl supply can use does not make pyrazine, it lands in the beer as FAN. Worked: 130 g RS = 722 mmol → 5 g glycine (66.5 mmol) = 10.9:1, and that is already ~62 mg N/L of FAN at 30 % carry-through, over the warn line; 4 g (13.5:1, ~50 mg N/L) is the better call.

Conclusion: glycine is the nitrogen, phenylalanine is the honey lever, and the dose is bounded above by FAN into the beer as much as by the sugar ratio.

3.2 Sugar¶

Reactivity, fastest first: ribose > arabinose ≈ xylose > galactose > fructose ≈ glucose > maltose > maltotriose > sucrose (inert until inverted).

• Pentoses are roughly an order of magnitude faster than hexoses and push the furan/furanone branch.
• Fructose caramelises faster; glucose forms Amadori products more readily. Equivalent for planning.
• DME is a quarter to a third as reactive per gram as invert sugar.
• Sucrose is not reducing, and it does not invert at pH 6. Inversion is acid-catalysed; sucrose is most stable near pH 9. Mango, peach, pineapple and dried apricot carry most of their sugar as sucrose, so their sugar figures overstate Maillard-available sugar two- to three-fold. The skill's ingredient table lists reducing sugar (RS) and sucrose (Suc) separately.
• Inversion is done on the hob, with the phosphoric acid dosed to a measured pH ~2.9, 45–60 min at the boil while the fruit reduces. Planning yield 0.70 of the sucrose, which is probably conservative: it understates the RS the nitrogen is sized against, and it is the first suspect if a batch comes out nitrogen-starved. Below pH ~2.4 you make HMF and darken the paste for no extra inversion.
• Uninverted sucrose is still fermentable in the beer. Sacch and Brett both hydrolyse it. It counts in the fermentable load whether or not it counted in the jar.
• Sugar alcohols (sorbitol in pear, prune, plum, aronia) are inert. Maltodextrin is excluded: negligible reactivity, inflates the Brix reading without depressing water activity, and Brett ferments it in bottle.
• Ribose from dried shiitake (2–5 g) is a reaction accelerator, not a flavouring, and it carries glutamate and ~20 % protein.

Conclusion: count reducing sugar for the jar and all sugar for the beer; invert sucrose-rich fruit on the hob at a measured pH, not a fixed acid volume.

3.3 pH¶

Three things about that table that earlier versions understated.

Cold pH is not reaction pH. At 121 °C the ion product of water rises by roughly 1.5–2 log units, so neutrality sits near pH 5.7–5.8 while carboxylic acid pKa values move comparatively little. A jar measured at 6.15 cold is meaningfully alkaline relative to neutrality at reaction temperature. The bands are an operational scale indexed on a cold reading, not a thermodynamic statement. (High)

A 65 °Brix paste is not a dilute ideal solution. Activity coefficients, pKa shifts in a concentrated sugar matrix and liquid-junction effects at the probe are all non-trivial. Treat ±0.3 as the honest uncertainty on any computed pH, and treat a meter reading on paste as a repeatable index rather than a thermodynamic pH. This is why the method now measures. (Med)

The cook drifts acidic, not alkaline. At the target pH there is no bicarbonate left as such; it has become potassium malate or citrate plus dissolved CO₂. Over 90–120 min, sugar fragmentation produces formic and acetic acid and amine is consumed; unbuffered model systems fall 1–2 units, a fruit jar with 20–170 meq of malate/citrate buffer falls perhaps 0.3–0.6. Starting at 6.15 and drifting means much of the cook runs in the mixed band. If the pyrazine branch must be held late, add buffer or start at 6.3. Never open a jar mid-cook to add base: it is a scald hazard and it vents the jar. (High on direction; Med on magnitude.)

Base dose arithmetic, anchored on native pH. Titratable acidity is titrated from the fruit's native pH to a pH 8.2 endpoint, so it measures only the protons still on the acid, not the total acid present; fruit at pH 3.5 already carries its own potassium. The fraction of TA you must neutralise therefore depends on where the fruit starts, not only on which acid it carries. The skill solves:

mmol_acid = TA_meq ÷ [ z(8.2) − z(pH_native) ]
base_meq  = mmol_acid × [ z(target) − z(pH_native) ]

where z is the mean negative charge on the acid at that pH. Potassium bicarbonate, MW 100.1: 1 g = 10 meq. Near a pH 6.15 target the correction over a naive solve is small (malic fruit needs ~0.93–0.95 of its TA, citric ~0.72–0.75, tartaric ~0.98); at a furanone-jar target it is large and depends on native pH (malic at 3.2 needs 0.42, at 3.9 only 0.21). No flat factor can be right. Where the native pH sits near an acid's top pKa (dates at 5.8 attributed to malic) the inference collapses and the skill falls back and says so; such ingredients carry little TA anyway.

The procedure, which is the real control: compute the full dose and the target; add 80 % in two or three portions (it foams); measure cold on a thin smear with a flat or spear probe (a glass bulb will not read paste); titrate the last of it in; record predicted against measured. Three batches of consistent delta is a brewery offset to apply as a standing correction; one batch is a measurement.

Why potassium bicarbonate: potassium is a beer cation already; sodium bicarbonate works at 1 g = 11.9 meq but adds sodium; calcium hydroxide overshoots on a steep curve and sets pectin-rich purée solid as calcium pectate. Pull back an overshoot with phosphoric acid, not lactic.

Conclusion: target 6.15 cold for the roast axis, solve the base from the fruit's own native pH, dose 80 %, measure, titrate, and log the delta.

Measuring it in an opaque paste: pH in pastes, with the slurry-correction and electrode-check bench. Solving the dose: base-dose planner.

3.4 Water activity and Brix¶

Maillard rate peaks at intermediate moisture, roughly a_w 0.5–0.8. Too wet and the reactants are dilute and the dehydration steps are suppressed; too dry and diffusion limits the rate.

Use the jam refractometer (58–90 °Brix), not consistency. A purée that holds a spoon-mark at 25–35 °Brix is a_w ~0.96–0.98, far above optimum. Jam-range a_w (0.80–0.87) is 60–68 °Brix. A 0–32 brewing refractometer cannot read the target; if it is all that is to hand, dilute gravimetrically (10.0 g paste + 30.0 g water, read, ×4).

The a_w column is a sucrose mapping, and a Stage B jar is not pure sucrose. Glycine and potassium salts read as Brix and depress a_w efficiently (small molecules, ions); collagen peptides read as Brix and barely depress it; maltodextrin likewise. Keep 60–68 °Brix as the operational handle and widen the a_w band you infer from it.

Getting there: reduce uncovered on the hob (this is also the inversion step), pre-dry fresh fruit in a low oven, or lean the Stage B backbone on dried fruit (dates arrive near 70 °Brix). The skill reports a lower-bound Brix and an upper-bound water-removal target; the refractometer decides when to stop. Pulp blurs the line; read the pressed liquid or a thin smear.

Reducing before sealing is one of the two largest kinetic levers at fixed 121 °C and biases toward pyrazines over furans. (High) It is also the Stage B safety hurdle (§13).

Conclusion: reduce the Stage B fraction to 60–68 °Brix, measured, before it goes in the jar.

Measuring it on a 0–32 instrument: Brix in pastes, with the dilution bench; water to remove and the a_w band: base-dose planner.

3.5 Temperature, time, vessel, headspace, cooling¶

121 °C in a stovetop pressure cooker is the practical ceiling that keeps volatiles in. Open-pan cooking at higher temperature vents the pyrazines and aldehydes as they form. Verify the cooker's actual rating once from its manual: many European stovetop models run 11–13 psi (115–118 °C). Browning has z ≈ 25–30 °C, so 3 °C costs perhaps a quarter of the rate; C. botulinum has z ≈ 10 °C, so F₀ roughly halves. An unverified cooker is mostly a safety uncertainty, and since Stage C now runs unbased (§9) it has stopped mattering much.

Vessel: closed jars standing in water inside the cooker. A 65 °Brix paste on the cooker base scorches and offers no free water for steam, and a weighted-regulator cooker vents steam continuously at pressure, so only the jar retains volatiles. Kilner or Mason jars in 3–5 cm of water, never below the cooker's minimum liquid over a 120 min cook. Stage B and C run simultaneously in separate jars. Sanitise jars and lids as for any packaging step.

Lid tightness. Finger-tight is the safe default, but the rationale earlier versions gave ("so pressure equalises") is the water-bath canning argument. In a pressure cooker the jar's internal vapour pressure tracks the cooker's at the same temperature; the real cost of a loose lid is on cooling, when the jar draws a vacuum and pulls cooker headspace and bath water back in, taking volatiles out. A paired jar with one lid tightened is cheap and worth running.

Headspace 10–15 %. Fill to 85–90 %. Headspace air is the oxygen for polyphenol and lipid oxidation and the volume that volatiles partition into.

Count time from full pressure. A 350 mL jar of paste lags the water by 10–20 min; the durations assume that lag.

Cooling is split by stage. Stage B (a_w ≤0.87): slow cool, jars closed; hours in the 60–100 °C band continue Amadori decomposition and residual Strecker chemistry, and nothing can grow at that water activity. Stage C (a_w ~0.95): cool fast, ice bath or running water to below 5 °C, then at most 5 days at ≤3 °C, then use or discard.

Hot vent before blending. Open the jar and stir for 5–10 min while still hot. Furan (bp 31 °C) and residual HCN leave; pyrazines, furanones, melanoidins and norisoprenoids are orders of magnitude less volatile and stay. Effectively free (§13.3).

Roast finish, untested. Spreading Stage B paste in a covered dish at 135–140 °C for 20–30 min may push pyrazines; it also vents volatiles and dries the surface past the a_w optimum. It is the top experimental priority precisely because §2's open question turns on it. Never above 140 °C.

Conclusion: 121 °C verified, closed jars in water, 10–15 % headspace, 90–120 min from full pressure, slow-cool B and fast-cool C, hot-vent everything.

3.6 Catalysts¶

• Phosphate. Phosphate ions catalyse the Amadori rearrangement and downstream steps. Add 85 % phosphoric acid (14.6 mmol/mL) to the crushed Stage B fruit before the hob reduction, dosed to a measured pH ~2.9, where it also drives inversion. A flat 1–1.5 mL against a low-TA date jar lands near pH 1.5–2 and makes HMF for an hour. The acid is then neutralised with the rest (~16 meq per mL at a 6.15 target); its buffering at pH 6.2 is modest next to the fruit's own citrate and malate. Catalyst, not pH insurance. (High on arithmetic; Med-high on catalysis.)
• Ribose from dried shiitake, 2–5 g. The fastest Maillard sugar available. Off-axis for floral and honey designs (sulphur volatiles).
• Pentoses liberated by the A1 pre-treatment (§9).

3.7 Antagonists¶

• Flavanol polyphenols. Catechin, epicatechin and phloroglucinol-type A-rings trap methylglyoxal and glyoxal by nucleophilic addition, the documented basis of polyphenol-based acrylamide suppression, and it applies to the pyrazine and Strecker exits equally. Aronia, sloe, blackcurrant, blueberry, cacao nib and oak suppress the roast axis in the same jar while producing their own phenolic complexity. Resolution is separation: cook the low-polyphenol sugar and nitrogen fraction hard (Stage B), the polyphenol fraction briefly and separately (Stage C), blend afterwards. (Med-high.)
• Sulphite. Bisulphite is the classic Maillard inhibitor: it binds dicarbonyls and sugar carbonyls. Sulphited dried apricot or raisin in Stage B kills the pyrazine axis before Brett ever sees the SO₂. Unsulphited only, or Stage C. (High)
• Sorbate. Not a Maillard antagonist, but a sourcing trap the larder work surfaced: potassium sorbate is standard on dried prunes and some soft dried fruit. It inhibits Saccharomyces, and LAB convert it to 2-ethoxyhexa-3,5-diene, the geranium taint of sorbated wines that undergo MLF. In a mixed-fermentation beer that is a real risk. Check the label as you would for sulphite. (High on the mechanism; Med on the dose at which it matters here.)
• Asparagine. The dominant free amino acid in most fruit. It consumes dicarbonyls, gives no useful aldehyde, and is the acrylamide precursor. Prefer low-asparagine reducing-sugar donors (date) for the Stage B backbone.

Conclusion: keep tannin, sulphite, sorbate and hope out of the pyrazine jar.

4. The fruit's own thermal chemistry¶

Running alongside Maillard at 121 °C:

Conclusion: the hot jar destroys colour, converts tannin, and generates cooked-floral and roast notes; acid, enzymes and the beer's own time are where glycosides, damascenone and pentoses are best released.

5. Flavour axes¶

The core reference. Target on the left, what to do on the right. Durability refers to survival through active fermentation and long Brett secondary (§6).

Conclusion: cocoa comes from glycine and heat; honey and jam come from the fruit's precursors and want acid and time as much as heat; fresh character wants no heat at all.

6. What survives the beer¶

Reduced by yeast alcohol dehydrogenases and old-yellow-enzyme reductases; will not survive active fermentation:

Transformed, not merely faded: monoterpenols. Sacch and Brett reduce geraniol to citronellol, cyclise linalool to α-terpineol, and acetylate both; β-lyase and glycosidase activity is strain-dependent. Free terpenes that meet active fermentation come out different, not just weaker. This is why the floral fraction goes in at bottling. (Med-high)

Stable through fermentation and years of ageing: alkylpyrazines; furanones (furaneol, homofuraneol, mesifurane, maltol, cyclotene, DDMP, though furaneol is partly reduced or methylated by yeast over a long secondary); pyrroles and thiazoles; β-damascenone, β-ionone, TDN; terpene oxides; vanillin, eugenol, coumarin, oak lactone; melanoidins and phenolic acids.

Partial protection. Yeast-derived SO₂ forms reversible bisulphite adducts with aldehydes, a latent pool released slowly in bottle. EC-1118 produces moderate SO₂. (Med)

Timing rule. Aldehyde-led value: add after primary is complete and accept slow attrition from Brett's reductases. Pyrazine, furanone and norisoprenoid-led value: timing is set by the sugar and acid accounting (§12), not by flavour survival. Terpene-led value: Stage T or D at bottling.

Conclusion: design the durable cocoa on pyrazines, treat Strecker aldehydes as a bonus that needs late addition, and never count on benzaldehyde or free terpenes that meet active yeast.

7. Nitrogen budgets: fusels, methionol, 2-phenylethanol, FAN and THP¶

Every protein- or amino-acid-bearing input is budgeted, not just the deliberate nitrogen. Intact protein barely hydrolyses at 121 °C and pH 6, so it is counted at a liberated fraction of 0.15 (Low), but 0.15 is not zero: 50 g of sweet almond carries ~75 mg of methionine against the 29 mg in a whole 4 g collagen cap.

7.1 Branched-chain fusels¶

Input in grams × 30 % unreacted (pessimistic) ÷ beer litres = mg/L free amino acid; × 0.3–0.6 = mg/L fusel added. Baseline isoamyl alcohol is already 40–80 mg/L. Warn at >25 mg/L added. Glycine and phenylalanine contribute nothing here.

7.2 Methionol¶

Count methionine at 100 %, not 30 %: the fraction that reacts becomes methional in the jar and yeast reduces that to methionol anyway. Yield 0.05–0.10 (wort at 15–25 mg/L Met gives ~1 mg/L methionol). Warn at >0.5 mg/L added at the midpoint yield. This is deliberately conservative: the Strecker route's molar yield is lower than the Ehrlich route's, so the estimate runs high, which is the right direction for a hard flag. It caps collagen at 4 g per 4.5 L of beer if collagen is used at all, and it is why the almond dose in a calendula jar is sized to what the lutein needs and no more.

7.3 2-Phenylethanol¶

Count phenylalanine at 100 % via two routes, both ending as 2-phenylethanol: the unreacted share (30 %) via Ehrlich at 0.3–0.6 mg/mg, and the reacted share (70 %) via phenylacetaldehyde at 20–40 % molar (×0.727 mass) then reduced (×1.017). 1 g of Phe into 4.5 L gives 43–86 mg/L; 0.5 g gives 22–43; 1.3 g gives 56–112. The useful band is 40–100 mg/L added on top of the base beer's 8–30; soapy-perfumed above. Some phenylacetaldehyde (honey, threshold ~5–10 µg/L) survives un-reduced if the jar goes in post-primary. (Med-low on the Strecker yield.)

7.4 FAN carried into the beer¶

Glycine is 18.66 % nitrogen. One gram into 4.5 L is 41 mg N/L at full carry-through, 12 mg N/L at the model's 30 % unreacted fraction; 4 g is ~50 mg N/L, a third of a whole wort's FAN, dropped into a secondary that is supposed to be nutrient-depleted. Expect renewed Brett and LAB activity, more acid, more biomass, a longer wait for a stable gravity, and broader substrate for decarboxylation chemistry. Report added FAN in mg N/L and warn above 30. Phenylalanine is 8.48 % N (~19 mg N/L per gram), cheaper on this axis per unit of effect. When a design is over the FAN line, reduce glycine before reducing sugar, i.e. run the ratio up.

7.5 Mousy THP¶

Brett and LAB make the mousy N-heterocycles from lysine and ornithine (ornithine from arginine via LAB arginine deiminase) with ethanol and acetaldehyde. The three compounds in wine are 2-acetyltetrahydropyridine (ATHP), 2-ethyltetrahydropyridine (ETHP) and 2-acetyl-1-pyrroline (APY). APY is not the desirable bread-crust twin of ATHP; at sour-beer concentrations and retronasal pH it is one of the mousy trio. Thermally formed APY is unstable in aqueous acid and mostly gone within weeks; the concern is the precursors the jar fails to consume.

The only reliable lever is not putting lysine and arginine in the jar. Glycine carries none. Collagen carries 3.6 % and 8.1 %, and the protease step does not rescue it: at ~20 % degree of hydrolysis most of the ~420 mg of Lys+Arg in 4 g still arrives in the beer peptide-bound, for Brett to release over months. Where collagen is used as a paired test: sugar ≥10:1 over reactive amine, 90–120 min at 60–68 °Brix, the visual gate (a pale, sweet syrup carries the most unreacted amine), low oxygen in secondary, and EC-1118 at bottling, which this brewery observes to clean up THP (mechanism not established, Med).

A second route to precursors is upstream: a warm hold of raw fruit grows heterofermentative LAB whose arginine deiminase pathway makes ornithine and citrulline (the ethyl carbamate precursor) before the beer exists. That is one reason the warm-hold rules in §13 are hard.

Conclusion: methionine caps every protein input, FAN caps glycine, and the cleanest THP strategy is no lysine or arginine in the jar at all.

8. Kernels: benzaldehyde and cyanide¶

Amygdalin (MW 457) in Rosaceae kernels yields one benzaldehyde (106) and one HCN (27) per molecule. Bitter apricot kernel runs 3–5 % amygdalin (plan on 4 %): ~9 mg/g benzaldehyde potential, ~2.4 mg/g HCN potential. Cracked sloe stone roughly a fifth of that; cherry, plum and peach stones lower again. Cultivar and season move these widely (Med). UK retail of bitter apricot kernels has largely stopped; the stones inside fruit you are buying anyway are the practical source.

Flavour binds first. Benzaldehyde threshold in beer is ~0.5–2 mg/L; above ~5 mg/L it reads as almond essence. Working range 1–2.5 g bitter apricot kernel or 4–12 g cracked sloe stone per 4.5 L, dosed post-primary.

The order of operations is the safety control. Hydrolysis is enzymatic: the kernel's own β-glucosidase (emulsin) does it at 40–50 °C when the crushed kernel is wet, in water. Never in fruit slurry: dilution, a pH well below the enzyme's optimum, competing substrate and glucose inhibition all work against it. At 121 °C the enzyme is dead within minutes and thermal hydrolysis at pH 6 is slow, so a kernel that goes straight into a hot stage delivers most of its amygdalin intact into the beer, where Brett β-glucosidase hydrolyses it over months and releases HCN in the bottle with nothing to vent. A raw kernel in the cold fraction is the worst case: 100 % arrives intact.

Sequence (Stage A2): 1. Crush; cover with 5–10× its mass of water; hold 1–3 h at 40–50 °C. 2. Vent: 10 min uncovered at sub-boiling, stirring. HCN removal from an aqueous slurry is governed by its Henry's-law constant, not its boiling point; planning residual 0.40. Benzaldehyde is steam-volatile too; planning survival 0.70. 3. Add to Stage C (further loss, survival 0.50, Low) or, for maximum marzipan, to Stage D post-primary.

Completeness check, free: after the hold the A2 water should smell strongly of bitter almond and the drained kernel much less so. If the kernel still smells strongly, hold longer or crush finer.

Effective retention. Un-hydrolysed amygdalin reverts to retention 1.0, because Brett finishes the job in the bottle:

effective retention = completeness × 0.40 + (1 − completeness) × 1.0

At the planning completeness of 0.70 that is 0.58. An incomplete A2 costs more than a short vent.

Safety arithmetic, per bottle. HCN mg/bottle = kernel_g × mg/g × effective_retention × bottle_L ÷ prep_L. EFSA's acute reference dose for cyanide is 20 µg/kg body weight, ~1.4 mg for a 70 kg adult per sitting; cyanide is cleared in hours, so per-sitting is the right metric. Warn at 50 % ARfD (0.7 mg/bottle) on the 375 mL bottle, and quote the figure for two bottles as well, because two in an evening is one sitting. Worked: 2 g bitter apricot kernel into 4.5 L, A2 and vent at 0.70 completeness, effective retention 0.58: 0.23 mg per bottle, 17 % ARfD (34 % for two); benzaldehyde in beer 2.2–2.8 mg/L post-primary. Skipping A2: 0.40 mg, 29 % ARfD, and the marzipan arrives slowly over months instead.

Ethyl carbamate. Residual cyanide plus ethanol forms ethyl carbamate over months, and the reaction is light-accelerated, the classic stone-fruit-spirit problem. At these doses the amounts are small and A2 plus the vent is the primary control, but it is a second reason the vent is not optional. Store kernel-bearing bottles in the dark.

Conclusion: hydrolyse in water, check the smell, vent, then use; a raw kernel in the beer is a slow cyanide release you cannot vent, and the honest retention is 0.58, not 0.2.

Comparing the routes on flavour per unit of cyanide: marzipan flavour against cyanide cost.

9. The stage architecture: A1, A2, B, C, D, T¶

Each fraction gets the conditions it wants; blend by taste afterwards.

Stage A1: pectin and pentose pre-treatment, cooked-only material. 45–50 °C, native pH, 1–3 h (≤2 h for low-acid fruit). For fruit destined for B or C only. Crush, dose Lallzyme Beta at ~50 mg per kg of pulp, hold, then straight to reduction and cook. The pectinase and arabinofuranosidase (not glucose-inhibited) lower viscosity, raise juice yield and liberate arabinose, galactose and rhamnose. What it does not do is release aglycones: the β-glucosidase is glucose-inhibited on raw fruit, which is why the manufacturer specifies use on dry wine at 5 g/hL. Skip A1 for dried fruit. Nothing from A1 is used uncooked.

Stage A2: kernel hydrolysis. Crushed kernel in water only, 40–50 °C, 1–3 h, then vent (§8). Nothing from A2 is used unheated; the vent is a sub-boil pasteurisation.

Stage B: the pyrazine jar. Low-polyphenol fraction. pH 6.15 cold (anchored solve, dosed 80 %, measured, titrated), 60–68 °Brix, 121 °C, 90–120 min, headspace 10–15 %. Reducing-sugar donors (sucrose inverted on the hob at pH ~2.9), glycine, optional phenylalanine, pentose donors, shiitake. No aronia, sloe, blackcurrant, cacao, oak, sulphited or sorbated fruit. A furanone jar variant runs at pH 4.3–4.6 for honey and jam targets; recompute its base from the fruit's native pH, never reuse a factor. Stage B is protected against growth by its water activity, not by the cook (§13.1): never seal a based, unreduced, high-moisture jar.

Stage C: the phenolic and caramel jar. High-polyphenol fraction. Unbased, at its native pH. 121 °C, 20–30 min. Anthocyanin and proanthocyanidin breakdown, some chlorogenic conversion, vinylphenol formation, caramelisation; glycosides largely survive for release in the beer. Why unbased: anthocyanin thermolysis is pH-independent above ~4, proanthocyanidin depolymerisation is acid-catalysed and better below 4, hydroxycinnamate decarboxylation runs at pH 4–6 anyway, and the jar is by design the tannin-suppressed one where Maillard is not the point. Basing it to 6.15 bought almost nothing and cost a low-acid, high-moisture preserve with an unverified thermal process plus most of the design's potassium and beer-pH load. A lift to 4.2–4.4 is allowed with a reason; above 4.6 the full low-acid process applies and must be argued for. Same cooker as B, own jar. Cool fast, ≤5 days at ≤3 °C.

Stage D: raw fruit addition. Never heated, never pre-held warm. Frozen-thawed fruit, delicate botanicals, and vented A2 kernel, into the beer post-primary exactly as for kriek. Aglycone release happens there: Brett β-glucosidase, plus beer-side Lallzyme Beta at 50 mg/L once glucose is gone. No enzyme on raw fruit, no warm hold. Its sugar sits outside the gravity gate (§12): add it early enough to ferment out.

Stage T: spirit tincture, the floral vehicle. Dried florals and terpene spices in 40 % spirit, 48 h, dark, strained; dosed at bottling to taste. Spirit extracts monoterpenes far better than beer, is self-sterilising, and is a blend-to-taste control. Bench ratio 1 mL per 100 mL of finished beer; typical 20–80 mL per 4.5 L (+0.2 to +0.7 % ABV). Also outside the gravity gate.

Blend B and C by taste; add D and T on their own timing. Keep a labelled jar of each hot stage. Not every design needs every stage, but skipping one must be explicit.

Conclusion: A1 for pentoses, A2 for kernels, B for roast, C for structure at native pH, D for fresh fruit, T for florals; four independent axes instead of one averaged outcome.

10. Choosing ingredients by role¶

Every ingredient has two or three roles. Pick to fill roles, not to match a flavour name.

Cross-cutting rules:

• Orange or yellow non-anthocyanin flesh means a carotenoid donor. Heat it for ionones; give it acid and time for damascenone.
• Aroma dramatically stronger cooked or fermented than raw means a glycoside donor. Stage D raw, or a short Stage C; consider the do-not-cook rule.
• The do-not-cook rule. If more than about half of an ingredient's value is in monoterpenes, thiol precursors or esters, and it carries no sugar backbone worth having, the cook destroys more than it makes. Blackcurrant, elderflower, rose, lavender, muscat raisin (for its terpenes) and Tagetes oil are Stage D or T ingredients. A good design says so rather than forcing a Maillard stage.
• Dried and fresh are not interchangeable by weight. Substitute on dry matter, show both multiplications, state the rehydration water. Where drying destroys the property you chose the ingredient for (carotenoid is the clear case), size the compensating dose and tag it.
• Sourcing traps: sulphite and sorbate on dried fruit; papain is in the latex of unripe green papaya and bromelain in pineapple stem, not ripe flesh; patulin in windfall or bruised pome fruit; rhubarb leaves are excluded; calendula versus Tagetes changes placement; bought dried petals may be pigment-depleted.

Three ingredients worth seeking deliberately: quince (the best single honey and floral fruit), muscat raisin (benchmark terpene glycoside source, into the beer whole), and dried apricot with the stones from fresh fruit (carotenoid honey axis and marzipan from one purchase).

11. Botanicals and spices¶

Most florals and spices are essential-oil driven and belong nowhere near an hour at 121 °C.

To maximise a floral target: build the honey axis hot (carotenoid fruit, phenylalanine), then layer fresh terpene character at bottling with a tincture. Cooked floral and fresh floral are different notes and stack.

12. Beer-side accounting¶

A prep is not just flavour; it is sugar, acid, potassium, FAN and colour going into a live, bottle-conditioned beer. Every design reports all of them.

12.1 Fermentable load: two numbers, not one¶

Reducing sugar not consumed by the cook, plus all sucrose (inverted or not), is fermentable. Planning: 25 % of Stage B reducing sugar reacts, 10 % of Stage C, 0 % of D; sucrose 100 % fermentable. ΔABV ≈ g/L × 0.51 ÷ 7.89. Report planned and worst case (0 % reacted) and plan on the worst case: amine exhaustion alone accounts for only ~10 % of the sugar in a typical jar, the rest of the "reacted" share is caramelisation and alkaline fragmentation, and until a batch measures it the error runs toward more sugar in the beer, not less. Even an accent-sized Stage B carries 10–20 g/L of fermentable sugar, because the sugar is the Maillard substrate; that is inherent to the method.

The gravity gate. This brewery bottles only on a stable gravity read twice, 4–6 weeks apart, with everything fermented out in secondary. That is the control, and it is why the skill computes no CO₂ volumes or bottle pressure. The carve-out matters: Stage D fruit and Stage T tincture are added at or near bottling, so their sugar sits outside the gate. Add Stage D early enough to ferment out and re-confirm gravity, or account for it explicitly. Rule: >10 g/L worst case → add early in secondary; ≤10 g/L → timing by flavour survival, still confirm gravity.

12.2 Acid, buffer and beer pH, as a magnitude¶

The base neutralised in Stage B arrives as potassium malate or citrate. Into a pH ~3.3 beer it raises pH and adds buffer capacity:

expected pH rise ≈ (base meq/L) ÷ (beer buffer capacity, 20–40 meq/L per pH unit)   [Med]

A design carrying 33 meq/L is +0.8 to +1.7 pH units: a change in perceived acidity, in the Brett/LAB balance, and in the protection low pH gives, arriving alongside a fresh sugar and FAN load. Warn above 10 meq/L. Unbased stages (C and D) carry their fruit acid in free and pull the other way; hibiscus in Stage C is the extreme case. Report both. Malate is MLF substrate; citrate and tartrate are not.

12.3 Potassium¶

~39 mg per meq neutralised. Typical whole-beer potassium is 300–500 mg/L, and potassium salts are perceptible as saline, metallic or bitter well below 1 g/L in a low-ionic-strength beverage. Warn above 500 mg/L added. Running Stage C unbased is what fixes this, together with the pH shift above; they are one problem with three symptoms.

12.4 FAN¶

§7.4. Report mg N/L added; warn above 30.

12.5 Colour, body, dose anchors¶

Melanoidin colour is permanent; an accent dose of a dark Stage B moves a pale beer several SRM. State the direction and rough magnitude.

Conclusion: every design carries a sugar, an acid, a potassium, a FAN and a colour number into the beer; state all five, plan on the worst-case sugar, and time the addition by the gravity gate.

13. Safety: warm holds, low-acid material, and thermal-process contaminants¶

13.1 Growth hurdles: which stage is protected by what¶

13.2 Why the warm-hold rules are hard¶

Crushed fruit at 40–50 °C is kettle-souring temperature. Thermotolerant heterofermentative LAB grow readily and make ornithine and citrulline (THP and ethyl carbamate precursors) and biogenic amines; Enterobacteriaceae and clostridia add acetic, butyric and isovaleric acid; polyphenol oxidase is fully active and oxidises the delicate fraction. Low-acid material (papaya, date, fig, ripe pear and mango, sweet almond slurry) held warm anaerobically is a C. botulinum window, and the cook is not absolution for one organism: B. cereus emetic toxin (cereulide) is heat-stable and survives 121 °C. Hence: no uncooked material above 30 °C for more than 3 h; no low-acid material above 30 °C for more than 2 h even if it is going to 121 °C; enzymes only on fractions that will be cooked. The earlier "Stage A" concept, a 12–24 h enzymatic hold on raw fruit part of which entered the beer uncooked, was withdrawn on exactly these grounds.

13.3 Thermal-process contaminants¶

Acrylamide forms from asparagine plus a reducing sugar from roughly 120 °C upwards, and the Stage B jar is a favourable regime: above the threshold for a long time, low moisture, pH 6.15, asparagine supplied by the fruit itself. It is documented in retorted and jarred products processed well below 140 °C. Scale, Low confidence: at 500–1000 µg/kg in the paste, ~200 g into 4.5 L is ~8–17 µg per 375 mL bottle against a typical adult dietary intake of 30–130 µg/day. Not alarming, not zero. Levers: glycine competes for the dicarbonyl pool and is a recognised acrylamide-reducing additive, so the nitrogen policy is already doing useful work; a furanone jar at 4.3–4.6 is substantially lower-acrylamide; prefer the lowest-asparagine backbone (date); keep the 140 °C ceiling on the roast finish as a browning control, not as a safety threshold.

Furan. Sealed containers at retort temperature with sugar, amino acid and ascorbate are the textbook furan-formation condition, studied precisely in jarred purées, and the sealed jar is the reason furan accumulates rather than escaping. Rosehip, sea buckthorn and blackcurrant are ascorbate-rich. Lever, effectively free: the hot vent (§3.5). (High that the condition applies; Low on concentration.)

HMF. Not a safety concern at this scale but a process signal: a jar that comes out of the inversion step very dark was over-acidified.

Conclusion: Stage B is protected by Brix, Stage C by acidity, the beer by itself; warm holds are short and always cooked; and every hot jar gets a hot vent.

14. Interplay with the funk programme¶

The funk programme is a separate body of work; this section reasons only from what is in the skill.

1. Caffeic liberation. Chlorogenic acid (aronia, prune, apple) partly hydrolyses to caffeic at pH 6, 121 °C; mostly it isomerises and lactonises. Some caffeic, not a lot, and less in an unbased Stage C. (Low-Med)
2. Thermal decarboxylation. Free ferulic and p-coumaric decarboxylate to 4-vinylguaiacol and 4-vinylphenol on heating, across pH 4–6, the wort-boil mechanism. Retained in the jar; Brett VPR then reduces them to the ethylphenols, skipping the PAD step. A Stage C jar is therefore also an HCA-processing step and can smell of clove before it meets Brett. It works at native Stage C pH and is not a reason to base the jar. (Med on extent)
3. HCA stocks go into the beer, not the jar, unless the vinylphenol is wanted pre-formed.
4. The brewery's recorded funk enablers (hibiscus, marigold, aronia, juniper, pink peppercorn, apricot) are consistent with HCA loading; observed here, not mechanism.

Conclusion: an unbased Stage C is also an HCA-processing step; decide on purpose whether the vinylphenols are made in the jar or by Brett.

15. Timeline compression: what a prep can and cannot do¶

Conclusion: a prep can make a 12-month beer read older on body, depth and honey; it cannot substitute for Brett maturity, and it should not be dosed as if it could.

16. Diagnostics¶

17. Worked example: an accent aronia/sloe/date build¶

The skill's own example design (v4/v5), sized for one 4.5 L carboy, accent archetype. Numbers from prep_calc.py.

Stage B jar (pyrazine): 60 g Medjool date, 20 g raisin, 2 g dried shiitake; 1.8 g glycine; inverted on the hob with phosphoric acid dosed to pH 2.9 (estimate 0.26 mL), 45–60 min at the boil. Reducing sugar 50.8 g (282 mmol) against 23.9 mmol amine: 11.8:1. Mixture native pH ~3.9 (estimated; measure). Base to pH 6.15: 0.99 g KHCO₃ full dose, 0.79 g first pass, then measure cold and titrate. Note the skill flags that date at pH 5.8 sits near malic acid's upper pKa, so the anchored solve is unreliable for that ingredient and it falls back; the meter decides. Known solutes put the mixture at ~65 °Brix already, so little reduction is needed; read the refractometer before sealing. 121 °C, 90–120 min, headspace 10–15 %, slow cool, hot vent.

Stage C jar (phenolic): 75 % of 150 g aronia + 60 g dried whole sloe (stones separated for A2), unbased at native pH ~3.4, 121 °C, 25 min in the same cooker. Ice bath, fridge, use within 5 days.

Stage D: the remaining 25 % of the aronia and sloe, raw, into the beer post-primary.

Stage A2: 6 g cracked sloe stone in water, 40–50 °C, 1–3 h, smell check, vent, then Stage D post-primary.

Budgets: collagen 0 g; isoamyl added 0.5–1.0 mg/L; methionol 0.06–0.11 mg/L (the shiitake protein); Lys+Arg 40 mg; FAN added ~26 mg N/L, under the 30 warn. Kernel: effective retention 0.58, benzaldehyde 1.4–1.8 mg/L in beer (in range), HCN 0.145 mg per 375 mL bottle, 10 % ARfD, 21 % for two. Beer-side: fermentable 14.9 g/L planned, 18.3 g/L worst case (ΔABV +1.0 to +1.2 %), so add early in secondary and hold the gravity gate; base carried in 2.2 meq/L, expected beer pH rise +0.06 to +0.11, free fruit acid from the unbased stages 16 meq/L pulling the other way; potassium +86 mg/L; Stage B concentrate 19 g/L and total fruit 76 g/L, inside the accent anchor. §21: three holds (A2 at 40–50 °C ≤3 h on a pH 5.5 material, so ≤2 h applies), nothing warm-held enters the beer uncooked, Brix confirmed before seal. §22: glycine present, pH 6.15 band, date backbone, hot vent.

Predicted: Stage B dark, cocoa and nut led by dimethyl- and trimethylpyrazines if the temperature question resolves favourably, otherwise furanone and caramel led with a nutty edge; savoury glutamate depth from the shiitake. Stage C smoky-phenolic, dried-fruit, caffeic-bearing, brown. Stage D tart and fresh with the sloe-stone marzipan intact. What this batch tests: the anchored solve against the meter (first calibration entry), the 25 % reacted fraction against gravity movement, and, if run as a pair, the roast finish.

18. Condensed rules¶

1. Nitrogen is glycine, dosed against reducing sugar to 10–14:1 molar, bounded above by FAN into the beer (warn 30 mg N/L). Phenylalanine is the honey lever. Collagen is off unless a paired jar says otherwise.
2. Count reducing sugar for the jar and all sugar for the beer. Invert on the hob at a measured pH 2.9.
3. Stage B at pH 6.15 cold: solve from native pH, dose 80 %, measure, titrate, log the delta. Never a flat factor, never a mid-cook addition. Stage C unbased.
4. Reduce Stage B to 60–68 °Brix, measured on the jam refractometer, before sealing. That is also its safety hurdle.
5. 121 °C (verified), closed jars in water, 10–15 % headspace, 90–120 min from full pressure. Slow-cool B, fast-cool C, hot-vent everything.
6. Tannin, sulphite and sorbate stay out of the pyrazine jar. Separate, then blend.
7. Kernels: crush, hydrolyse in water, smell-check, vent, then use post-primary. Effective HCN retention 0.58; quote one and two bottles; store dark.
8. No uncooked material above 30 °C for more than 3 h; 2 h if low-acid; enzymes only on cooked fractions. Raw fruit goes into the beer as fruit.
9. Florals go in at bottling as a tincture; terpenes that meet active yeast are transformed.
10. Every design states fermentable load planned and worst case, ΔABV, the gravity gate and what sits outside it, base meq/L and the expected beer-pH rise, potassium, FAN and colour.
11. Every protein-bearing input enters the methionol budget: almond, kernel, cacao, shiitake.
12. Predict before you cook; score the prediction when you taste; correct the reference figure that was wrong; three consistent pH deltas become a standing offset.

19. Open experiments¶

In priority order, because a good deal of the method's design follows from them.

1. The roast-finish paired jar. Split one Stage B paste into two identical jars; give one a covered 135–140 °C finish for 20 min. If the roast jar is markedly more pyrazine-led, the binding constraint at 121 °C is temperature, not nitrogen, and the nitrogen policy is over-engineered. One batch settles it.
2. The first three pH calibration entries. Until then the anchored solve is theory and the meter is the method.
3. The lid-tightness pair. One jar finger-tight, one tightened; the loss on cooling is the question.
4. The inversion yield. A refractometer reading before and after the hob step on a sucrose-rich fruit would replace the 0.70 planning figure.
5. The 25 % reacted fraction. Gravity movement after a known addition would replace it.
6. Collagen, if ever. One jar with 4 g papain-treated collagen, one without, everything else identical, scored for bread-crust and savoury character against the methionol and THP cost.

20. Glossary, confidence register, starting references¶

Glossary. Amadori compound: the rearranged sugar-amine adduct that opens Maillard. α-Dicarbonyl: two adjacent carbonyls (methylglyoxal, deoxyosones); the reactive hub. Strecker aldehyde: the aldehyde an amino acid becomes when a dicarbonyl strips its amine. Alkylpyrazine: aromatic N-heterocycle from two aminoketones; roast and cocoa; permanent. Furanone: five-membered oxygen heterocycle (furaneol, maltol); caramel and jam; permanent. Norisoprenoid: C13 carotenoid fragment (damascenone, ionone); honey and violet; permanent. Melanoidin: the brown polymer end-product; colour and body, no aroma. a_w: water activity, the fraction of water free to react; 60–68 °Brix ≈ 0.80–0.87. Anchored solve: the base-dose calculation that starts from the fruit's native pH rather than from fully protonated acid. FAN: free amino nitrogen, the yeast-assimilable nitrogen a prep carries into the beer. THP: the tetrahydropyridine and pyrroline family behind mousy off-flavour. VPR/PAD: Brett's vinylphenol reductase and phenolic acid decarboxylase, the funk enzymes. ARfD: acute reference dose, the single-sitting exposure limit. F₀: the equivalent minutes at 121 °C delivered to the cold spot, the low-acid canning measure. Gravity gate: bottling only on a stable gravity read twice, 4–6 weeks apart.

Confidence register for the claims that change practice.

Starting references (from memory; verify citations before quoting them): Martins, Jongen & van Boekel (2000) on the Maillard reaction in food, Trends Food Sci Technol; Labuza's water-activity and browning-rate work; Totlani & Peterson (2005) on epicatechin trapping methylglyoxal, J Agric Food Chem; Costello & Henschke (2002) on mousy N-heterocycles ATHP, ETHP and APY, J Agric Food Chem; Chevance et al. (2002) on β-damascenone in fresh and aged beer, J Agric Food Chem; Coghe et al. (2004) on ferulic acid release and 4-vinylguaiacol formation in brewing; Daenen et al. (2008) on glucoside hydrolase activity in Saccharomyces and Brettanomyces, J Appl Microbiol; Bolarinwa, Orfila & Morgan (2014) on amygdalin content of UK kernels and food products, Food Chem; EFSA CONTAM Panel (2019) on cyanogenic glycosides, ARfD 20 µg/kg bw; Meilgaard (1975) on beer flavour thresholds; Vanderhaegen et al. (2006) on the chemistry of beer ageing; EFSA and FDA reviews on acrylamide and furan in heat-treated foods; Lallemand's Lallzyme Beta technical note (5 g/hL on dry wine; glucose inhibition of β-glucosidase). The brewery's own brett_flavour_science.md, funk_development.md and brewing_observations.md are the local ground truth for the Brett-side claims and are deliberately outside this skill.