Putting water underground can create a valuable reserve. The difficult accounting begins when water moves, mixes, reacts and returns through a different boundary from the one where it entered.
The meter records an entrance, not a complete journey
At a managed aquifer recharge test site in Hongseong, Korea, a study published online on 4 February 2026 examined injection behaviour alongside water chemistry, sediment properties and modelling. Gyoo-Bum Kim and colleagues reported generally stable injectivity during approximately 2.7 months, with some declines and remaining questions about longer-term changes. That is a bounded field result, not a promise that a well will accept water indefinitely. The author abstract makes the observation period visible, an important qualification for infrastructure intended to operate much longer. Kim and colleagues
The study points to a deceptively ordinary difficulty. A flow meter can record water entering a recharge facility. It cannot, by itself, tell us how much usable water will be available somewhere else months later. The intervening system is not a sealed vessel. It includes a geological setting, water already present, changing operating conditions, and chemical interactions that may matter as much as the volume delivered.
Managed aquifer recharge, or MAR, is intentional replenishment of groundwater for later use or other defined benefits. That broad purpose can involve water entering through the ground surface or through wells. Aquifer storage and recovery is one form, in which water is introduced and later recovered through well infrastructure. The variety matters: an infiltration basin, a continuously operating replenishment project and a seasonal storage scheme need not share the same performance measure.
The appealing picture is a bank account. Deposit water when it is available; withdraw it when it is needed. The metaphor is useful only if the account has rules for movement across its boundaries, mixtures with other water, changes in quality and the timing of access. A deposit recorded at one point is not automatically a claim on an identical volume at another. Nor does a lower recovery fraction mean that all unrecovered water has vanished or that no environmental benefit occurred.
To see the engineering clearly, it helps to follow several questions separately. How much water reaches the intended formation? How does the receiving system respond? What water is later sampled or extracted? Is its quality appropriate for the proposed use under the applicable assessment? What would have happened without the project? The same programme can answer one question well and leave another unresolved. Treating them as one success metric makes both achievement and failure harder to recognize.
Give every volume a boundary and a date
Hypothetical water-accounting illustration, not site measurements. The second account partitions the same infiltrated volume; it must not be added to the first.
Consider a fictional recharge project, with all numbers in this section invented for explanation. During one accounting period, its source delivers one thousand cubic metres to a surface basin. Eight hundred cross the basin floor; one hundred remain in the basin at the cutoff time; one hundred are assigned to other measured or estimated losses. The first balance closes: the three destinations add to the delivered volume. But the eight hundred have crossed only one boundary.
Suppose a separate subsurface assessment estimates that, by the same date, six hundred cubic metres have reached the target aquifer and two hundred remain above it. The correct statement is not that two hundred were necessarily wasted. They occupy a different part of the system at the chosen time. If they arrive later, that later arrival belongs in the later account. Counting all eight hundred as immediately available groundwater would erase a delay that could matter during a shortage.
Now suppose the project subsequently extracts four hundred cubic metres from a recovery well. The extraction meter records four hundred, but that number does not identify the origin of every portion. Some could be recharge water and some pre-existing groundwater. A calculation of total extraction divided by earlier infiltration would therefore answer a different question from a calculation of recovered recharge water divided by water that reached the aquifer. Both ratios might be useful; neither should be labelled simply efficiency without a definition.
The cutoff dates are equally important. A deposit in the final week of a reporting year may still be moving through the system when the annual report is produced. Comparing that deposit with recovery during the same year could make the system look inefficient for a purely accounting reason. Extending the comparison too far can create the opposite problem, because later extraction may draw on other recharge events. A transparent account retains the time windows instead of choosing whichever window gives the most flattering fraction.
The distinction between a surface observation and arrival at the water table is not merely a thought experiment. A US Geological Survey report by Joseph Nawikas and colleagues, issued in October 2016, addresses monitoring water movement through unsaturated alluvium during managed recharge. Its accessible agency abstract notes that water-level observations do not always make the arrival of infiltrated water at the water table clear. The underlying report PDF was inaccessible during this research, so no detailed monitoring results are inferred from it here. USGS report record
An original project dashboard built around these distinctions would have separate columns for delivery, infiltration, estimated aquifer arrival, extraction and attribution. It would also retain unresolved balance terms rather than silently assigning them to beneficial storage. The purpose would not be to demand an impossible inventory of individual molecules. It would be to prevent a well-measured number at one boundary from being promoted into a different, less-supported claim.
The useful capacity is the overlap of several calendars
An empty basin has no recharge value during a period when no suitable source water is available to it. A large seasonal water volume has limited value to a particular project if it arrives faster than that project can accept it. A nominal pumping capacity is not the same as an annual delivered volume. The units and calendars must meet before the capacity can become a meaningful resource estimate.
Here is a fictional scheduling calculation. A source can supply ten volume units per day for ten days. A recharge facility can accept six units per day over that interval. Without an additional holding arrangement or another destination, the overlap permits sixty units to enter the facility, not one hundred. Multiplying the facility's daily rate by every day of the year would give a much larger number, but the source is not available every day of the year in this example.
Now shift the source's ten-day window to a period when the facility is unavailable for five of those days. The overlapping opportunity drops to thirty units. The annual source total and the nominal facility rate have not changed. Timing has. Conversely, if a project can redistribute delivery safely within its evaluated arrangement, it may capture more of the available volume. The arithmetic identifies a planning issue; it is not a recommendation for how to operate a real recharge system.
The same logic applies to several facilities sharing a source. Imagine two proposed projects each claiming all one hundred units from the same fictional seasonal supply. Each proposal can look feasible when read alone. Together they claim two hundred units from a source that provides one hundred. Resolving the inconsistency requires a shared allocation and timing account, not merely a better infiltration estimate for each individual site.
Nicola Ulibarri and colleagues' 2021 study examined California recharge proposals through content analysis of groundwater sustainability plans. It identified implementation concerns involving available water, legal arrangements, funding and land requirements. The unit of analysis was proposed projects and their documentation, not a measured inventory of completed recharge. Its findings are therefore evidence about planning feasibility, not a current legal opinion or a declaration that the proposals were subsequently built. Ulibarri and colleagues
This makes an apparently administrative question part of the physical story. A project cannot replenish an aquifer with a volume that is counted elsewhere, unavailable during its operating window, or not deliverable through the infrastructure assumed in its plan. Good geology is valuable, but it does not repair a missing source. A realistic capacity statement therefore needs both a rate and an opportunity to use that rate.
A falling water surface can conceal different explanations
A basin's water surface gives an immediate visual impression of whether water is disappearing into the ground. Interpretation is less immediate. A decline can be affected by how inflow and outflow are recorded, when the measurement begins, and which phase of an operating cycle is being examined. The useful quantity is not simply that the basin emptied; it is the relationship between the measured change and the explicitly defined water balance.
Roy Elkayam and Ovadia Lev's March 2024 preprint examined a decade of records from fifty recharge basins at Shafdan in Israel. They reported a consistent, approximately head-independent infiltration pattern during the drainage phase for most analysed events. The opened paper is a preprint, and the reported regularity belongs to the studied system and event definition. It does not establish that adding water depth has no hydraulic significance in every recharge basin or that all soils follow one universal operational rule. Elkayam and Lev
An original comparison helps separate two possible summaries. Suppose a fictional basin's surface level falls by ten centimetres in each of three equal observation intervals after inflow stops. Describing the measured decline as approximately constant over those intervals is reasonable. It is a different claim to say that the same rate will persist after a long period of operation, under another source-water composition, or in a neighbouring basin. Those extensions introduce variables absent from the three-interval description.
Suppose instead that the fall is ten, seven and four centimetres. A single average of seven centimetres per interval captures the total decline but suppresses its time pattern. If a planner needs to know when the basin will be available again, the pattern can matter. Neither fictional sequence is presented as a measured Shafdan result. They illustrate why the phase of measurement and the definition of rate belong beside any reported performance number.
Comparing seasons also requires comparable denominators. A programme could report total annual infiltration, average rate while active, or rate averaged across active and inactive days. These describe different features of the system. A high active-period rate can coexist with a modest annual total if opportunities are infrequent. A lower rate maintained for longer can deliver a larger volume. The distinction is not semantic; it changes what an apparent improvement means.
The lesson for a reader is to look for the actual curve or interval behind the adjective stable. Stability over a measured phase is useful evidence. It does not mean that the geological system has become timeless. An honest account can celebrate regularity because it helps a defined calculation while remaining open about the conditions under which that regularity was observed.
Water quality is also a property of the encounter
It seems intuitive that cleaner incoming water should produce only cleaner groundwater. That intuition can fail when it treats the aquifer as chemically inert. The receiving sediments and the water already present are part of the system. A change in the mixture or chemical environment can alter how substances move between solids and water. Evaluating source-water quality is necessary; it is not the whole compatibility assessment.
A 2015 study by Sarah Fakhreddine and colleagues, described in the Stanford Synchrotron Radiation Lightsource's own February 2016 research account, investigated arsenic release associated with recharge in Orange County, California. Laboratory work and spectroscopy were used to examine interactions with aquifer sediments. The facility account emphasizes that highly treated water can still alter sediment-water chemistry. This is a primary institutional account of the research, not full access to the journal paper or a prescription for adjusting recharge water. SSRL's research account
The distinction between concentration and mass can make the issue clearer without supplying a treatment recipe. Suppose a fictional water sample initially contains two units of a constituent in one litre. Mixing it with one litre containing none produces two units in two litres, or one unit per litre, if nothing reacts and the volumes add. The concentration has halved, but the mass of the constituent has not. Dilution and removal are different statements.
Now imagine that contact with the fictional sediment adds another two units to the water. The final concentration becomes two units per litre, even though the added water began with none. This is only an invented mass-balance example. It does not predict a real contaminant concentration. It demonstrates why measuring the incoming water cannot, by itself, identify all sources of a substance observed later.
A separate Fakhreddine-led paper published online in July 2020 investigated geochemical heterogeneity across an Orange County injection system. Its author abstract attributes arsenic release to different mechanisms in differently conditioned aquifer zones, combining water chemistry, aquifer analysis and reactive-transport modelling. The result is not that every recharge project mobilizes arsenic in the same way. It is that one well can interact with zones whose prior histories matter to the explanation. Fakhreddine and colleagues, 2020
A fictional monitoring design that combines all those zones into one blended sample might miss an important difference. If one contributing interval has a higher concentration and another a lower one, the mixture can remain unchanged even while the contributions shift. A stable composite number is useful for the mixture actually sampled. It is not automatic evidence that each contributing interval remained stable.
This is not an argument against recharge. It is an argument against evaluating compatibility using a slogan about purity. The technically useful question is how the introduced water, native groundwater and sediments interact under the relevant conditions, and how that interaction is checked. A favourable answer must be demonstrated for the system being considered; an adverse result elsewhere is a reason to investigate a mechanism, not a substitute for local evidence.
From a trial to a larger system, the question changes
A short trial can reveal a reaction or provide data for a model. A larger project introduces another set of questions about the spatial extent and duration of that reaction. Scaling the volume does not automatically preserve the ratio of water to reactive material, the distribution of flow, or the sequence in which different parts of the system encounter the introduced water. A model has to state which features it carries forward and which it changes.
Jing Sun and colleagues' October 2020 study examined large-scale geochemical assessment for groundwater replenishment in Perth. It combined a model developed from a field trial, observations from the early full-scale operation, and an upscaled reactive-transport representation. The paper keeps those evidence stages distinct. It investigates specified buffering and water-quality processes rather than claiming that a brief trial alone establishes every long-term consequence of replenishment. Sun and colleagues
To make the inference concrete, imagine a fictional test in which a small volume of introduced water passes through material with a finite capacity to take part in a reaction. An initial response may be consistent with abundant available capacity. Repeating that response indefinitely in a forecast would assume the capacity remains available. That assumption may be defensible, conservative or inappropriate, but it requires an account of the relevant quantities and replenishment, not only a good fit to the first observations.
There is a second, different issue when the small trial samples only one part of a heterogeneous formation. Increasing the project footprint may encounter regions absent from the original test. A larger operation can therefore be more than a longer version of the same experiment. It may test whether the conceptual description was representative. The appropriate comparison asks which new observations could challenge that description, rather than only whether the larger system produces more data points.
A useful original reporting format would separate measured trial behaviour, measured early operational behaviour and projected future behaviour into three clearly labelled groups. A common axis could help compare them, but the graphic should not use a continuous observed-data style to conceal where observation ends. The model's future line may be valuable. Its value depends on being recognizable as a conditional projection.
The research question is consequently not whether modelling should be trusted or rejected as a category. It is what each model component is constrained by, whether the relevant observations are reproduced for an intelligible reason, and how changes in assumptions affect the stated outcome. Those questions are especially important when a successful short demonstration becomes the public face of a much longer proposed investment.
A tracer identifies a relationship, not a universal water age
Finding a marker in a monitoring well can connect that sample to a source. The strength of the inference depends on what the marker does along the way. A substance that reacts, a temperature signal that exchanges heat with surrounding material, and an indicator of mixing do not necessarily measure the same aspect of movement. More indicators can help, but their different behaviour must be interpreted rather than averaged away.
Christian Moeck and colleagues' 2017 study compared seven tracer approaches at a managed recharge site in northern Switzerland. The accepted manuscript discusses how tracer properties and aquifer heterogeneity affect interpretation, including the need to account for thermal retardation when comparing heat with other transport indicators. The study's value lies in comparing complementary evidence at the same system, not in declaring that seven readings always produce a single unambiguous residence time. Moeck and colleagues
Here is an original mixing calculation using a deliberately fictional conservative indicator. Let the recharge source have indicator concentration ten units and the native groundwater have two. If a sample measures six, a two-source, nonreactive, constant-endmember model gives a recharge fraction of one half: six is halfway between two and ten. This is an inference under the stipulated model, not a direct label attached to half the molecules in a real well.
Change the source concentration from ten to fourteen while retaining the sample concentration of six and native concentration of two. The same calculation now gives one third: the difference of four between sample and native water is one third of the source-to-native difference of twelve. A changing source can therefore alter the interpretation even when the sample concentration is unchanged. The example shows why the input history belongs in the analysis.
Introduce a third source, or allow the indicator to react, and the simple two-endmember calculation may no longer determine a unique fraction. The right response is not to report the original half with more decimal places. It is to recognize that the model has acquired another unknown or lost an assumption. Additional observations can help constrain that unknown, but only if they contribute information not already built into the first calculation.
This makes sampling time important as well as sampling location. A late sample may show that a marker has arrived, but not when its first arrival occurred. A record with gaps can miss a short-lived change. Conversely, a dense time series at one location does not automatically describe the whole formation. A tracer programme is informative through the relationship between its measurements and the question being asked, not through the number of samples alone.
The input can change while the front is moving
Meredith Martinez and Mark Widdowson's paper, published online on 15 November 2024, studies transport in a multilayered aquifer receiving continuous replenishment. It uses several constituents and an analytical approximation that accommodates changing operations and input conditions. The study identifies a useful conservative indicator for its site rather than recommending the same indicator everywhere. The opened published PDF establishes the online date; the university repository's December date is not substituted for it. Martinez and Widdowson
An imagined step change shows the interpretation problem. Suppose a facility introduces water carrying a marker at one level, then changes to a lower level. A downstream sample collected later may contain contributions from both periods. Comparing it only with today's input can make a perfectly ordinary delayed response look like removal or an unexpected source. Comparing it only with the initial input can make the opposite mistake. The relevant reference is the time history reaching the sample, not whichever input number is easiest to retrieve.
Flow distribution adds another dimension. In a fictional well contributing to two layers, the share entering each layer might change between periods. A combined meter records the total but does not identify the layer-specific shares. If the layers have different transport behaviour, an unchanged total rate can coexist with a changed downstream signal. Again, the measurement is not wrong. It is answering a more aggregated question than the one the interpretation requires.
The distinction can be expressed without drawing a misleading underground movie. A conceptual figure can show a source history, two unspecified pathways and a sampled mixture, with the pathways explicitly labelled hypothetical. It should not present their lengths, velocities or shapes as measured unless those quantities are supported. The appealing smooth plume in a diagram is often more specific than the evidence allows.
This is particularly important when a project expands. The distance over which source water can be distinguished from background water, the cost of observing it and the variability of the source all affect what can be inferred. A monitoring approach that works well near an initial test may need a different interpretation farther away. The scientific task is to preserve the link between source history and observation as the scale changes, not merely to move the same instrument to a more distant point.
Recoverable water is not all the water that remains underground
In brackish or saline settings, water introduced for storage can mix with water already present. Density differences can also change where the introduced water sits. A quantity can remain in the subsurface yet become difficult to recover at the desired quality through the selected infrastructure. The relevant performance measure therefore includes a quality condition and a recovery arrangement, not just whether water entered a geological formation.
Lilli Witt and colleagues' April 2021 laboratory study visualized freshwater storage and recovery in a brackish model aquifer. The team combined tank observations with numerical interpretation and found that the three-dimensional flow context mattered to understanding visible patterns. The experiment also examined the role of well infrastructure in movement. These are controlled laboratory observations, not measured recovery percentages promised to a real municipality. Witt and colleagues
A fictional recovery test makes the role of the endpoint explicit. Suppose one hundred volume units are introduced. Later, the first forty units extracted satisfy a chosen analytical quality criterion, while subsequent extraction does not. A quality-limited recovery fraction defined as those forty units divided by the introduced hundred is forty per cent. It does not mean that exactly sixty units have disappeared. Some may remain, move elsewhere or be present in mixtures outside the defined recovery condition.
If a different assessment uses a different permitted endpoint, its reported recovery fraction may differ even for the same physical sequence. This is not permission to relax a real water-quality requirement. It is a warning that comparative percentages need common definitions. A study using one cutoff cannot be compared casually with another using a different cutoff, storage period or accounting convention.
Simon Kreipl, Mark Bakker and Boris van Breukelen's February 2026 paper compares well configurations numerically under selected saline-aquifer conditions. The simulations target situations in which conventional vertical configurations perform poorly and examine the consequences of mixing and buoyancy. Their reported advantage for horizontal wells belongs to that modelling design and parameter selection; it is not field validation that horizontal wells are always preferable. Kreipl and colleagues
The comparison illustrates why a design headline needs its denominator of circumstances. Was the test meant to represent typical sites, a difficult subset, or a range chosen to expose a mechanism? A method that improves a difficult subset may be valuable without being universally best. Conversely, a modest overall average can conceal a useful response in a clearly defined setting. The interpretation depends on how the comparison population was constructed.
A recovery report should also explain whether a cycle begins with residual water from earlier cycles. Otherwise, improvements across repeated cycles can be mistaken for an independent fresh start each time. In an original accounting exercise, an initial inventory of twenty units plus a new deposit of one hundred is a different starting condition from a deposit of one hundred into an otherwise identical system without that inventory. The balance should carry the initial stock forward explicitly rather than treating it as free unexplained recovery.
Fast entry can create a different protection problem
A formation that accepts water readily is attractive for recharge capacity, but rapid hydraulic connection is not automatically favourable for every water-quality objective. The project has to evaluate the path between the surface, the recharge location and the places where groundwater is used. A high acceptance rate does not establish the treatment or protection achieved along that path. Volume and vulnerability can be related without being interchangeable.
Julian Xanke's 2017 dissertation examines recharge associated with Jordan's Wala reservoir and a karst groundwater system. Its opened author-repository abstract describes a conceptual and numerical assessment of recharge, abstraction and protection in a heterogeneous setting. The work includes historical analysis and projections; a forecast extending to a later year is not evidence that the projected outcome was subsequently observed. The complete dissertation PDF exceeded the access tool's size limit, so the account here remains within the abstract. Xanke's dissertation record
A hypothetical comparison clarifies the tradeoff. Site A transfers a signal to a monitored point quickly. Site B transfers it more slowly. If the only objective is to observe prompt hydraulic response, A may look attractive. If a separate objective relies on processes requiring contact time, speed alone does not establish success. Neither site can be ranked overall until the objectives, constraints and relevant observations have been specified.
The same issue arises when a project seeks an environmental benefit rather than a later withdrawal. Water moving out of a narrowly drawn storage boundary may support a different intended outcome. Calling it a loss without reference to the objective would be misleading. Calling it recoverable supply for the original user would also be misleading. An account needs both the physical destination and the beneficiary or purpose attached to it.
This is why the reservoir metaphor eventually fails. A tank has an engineered enclosure and a comparatively direct inventory. An aquifer project operates within a larger hydrological system. Its boundaries are analytical and managerial choices as well as physical features. Those choices can be reasonable, but they must be stated. Otherwise, water can be counted as a success whenever it enters the chosen picture and ignored whenever it leaves.
A recharge project needs a counterfactual, not just a total
Imagine a fictional aquifer region that receives an additional sixty units through a new project during a reporting period. If extraction elsewhere in the same account rises by eighty units, the new recharge does not by itself demonstrate a net gain of sixty in stored water. Other inflows, outflows and changes still matter. The project's gross contribution can be real while the regional balance continues to decline.
Now imagine that the project's source would otherwise have infiltrated downstream within the same regional boundary. Diverting it to a new basin may alter timing, location or quality without adding the entire diverted volume to regional recharge. Alternatively, the source might otherwise have left that boundary. The two situations have different implications for additionality even when the new basin's inlet meter records the same number.
These are explicitly hypothetical comparisons, not assertions about a named project's effect. They show why a before-and-after total is not automatically a causal estimate. Changes in rainfall, pumping, other infrastructure or the accounting boundary can affect the comparison. A useful assessment asks what alternative trajectory is being used and how uncertainty in that trajectory affects the claimed benefit.
The counterfactual should match the purpose. A project intended to shift water availability from one season to another may create value without increasing the annual regional total. A project intended to improve a particular groundwater condition needs evidence about that condition, not merely a seasonal storage story. A project intended to reduce a deficit needs an account of the deficit. Using the same headline volume to claim every possible benefit would hide these distinctions.
For a fictional programme report, the most informative presentation would show gross delivery, estimated contribution to the target system, the selected comparison without the project, and the outcome relevant to its purpose. The quantities would not necessarily be equal. Their differences would be part of the explanation rather than an embarrassment to remove from the headline.
The result has to remain useful in a dry year
A recharge programme can perform well during a favourable observation window and still face a difficult reliability question. What happens when the source is scarce at the same time that demand is high? An annual average can conceal that coincidence. A resource intended as a buffer should be evaluated in relation to the periods when the buffer is needed, not only the periods when filling it is easiest.
An original three-year illustration makes the difference visible. Suppose hypothetical recharge contributions are ninety, ninety and zero units, while demand for the stored supply is thirty, thirty and one hundred. The average annual contribution is sixty units and average demand is a little over fifty-three. Those averages look compatible. Whether the system can meet the final year's demand depends on carryover, recovery limits, other outflows and the inventory available at that time. The average alone cannot answer it.
Assume, solely for this simplified calculation, that every unused unit can be carried over without any loss or other claim. After satisfying the first two years' demand, the fictional account accumulates one hundred and twenty units, enough for the third year's hundred. Remove that perfect-carryover assumption and the conclusion may change. The exercise shows exactly which assumption makes the reassuring answer possible; it supplies no estimate of a real system's retention or usable yield.
A second scenario might receive all the water after the period of greatest demand. Its annual totals could be identical while its ability to help at the critical time is different. The planning question is consequently temporal as well as volumetric. It needs a sequence of availability and use, not a single large number of cubic metres on a publicity graphic.
This is also where operational evidence should be reported in ordinary terms. How much of the evaluated opportunity was used? Which periods could not be observed? Which quantities are measured and which are estimated? A gap in the record is not automatically a failure, but it should not be silently converted into full availability. The same care that protects a scientific conclusion protects the credibility of a public investment claim.
A more demanding kind of water reserve
Managed aquifer recharge is compelling because it can connect water available at one time with a need or benefit at another. The research does not reduce that possibility to a single yes-or-no verdict. It reveals why the connection has to be demonstrated across several boundaries: entry, transport, chemical compatibility, recovery and the wider water account. None of those boundaries is adequately described by a photograph of a full basin.
The most useful number is therefore not always the largest one. A modest, well-defined estimate of usable supply under stated conditions can be more valuable than a much larger inlet total with no account of timing or destination. A measured improvement in one groundwater condition can be meaningful without being advertised as proof of every other benefit. Precision about the claim makes a successful project easier, not harder, to recognize.
For the next recharge announcement, the revealing questions are practical. What boundary does the volume cross? Over what period? What happens between that boundary and the claimed benefit? How is quality assessed after interaction with the receiving system? Which part of the result is measured, and which is modelled? What changes in the comparison without the project? The answers turn an underground-storage metaphor into a technically interpretable proposition.
An aquifer is not an empty tank waiting for a deposit. It is an existing system into which a new flow is introduced. The engineering opportunity is real precisely because that system can move, store and transform water. Those same characteristics make the accounting demanding. The goal is not to eliminate them from the story, but to understand them well enough that the reserve being promised is the reserve the evidence actually supports.