For more than twenty years, sodium iodide crystals under the Gran Sasso counted slightly more events every year around the start of June. Rebuild DAMA's 2021 claim from its own plotted points, plant it at full size in two independent detectors' released data, then move the first day of the year and watch what background subtraction can create.
The claim reconstructed here: 2021, 2-6 keV
13.7σ as printed
In 2021 the DAMA collaboration reported, for the 2-6 keV window and 2.86 ton×yr of exposure, a modulation amplitude of (0.01014 ± 0.00074) cpd/kg/keV at 13.7 σ, a measured phase of (142.4 ± 4.2) days and a measured period of (0.99834 ± 0.00067) yr. In the collaboration's words, all these values are well in agreement with those expected for DM particles.
Bernabei et al. 2021, abstract and Table 2. The modulation is a pattern observed in DAMA's detectors; that dark matter causes it is the disputed inference. An amplitude divided by its error is not the probability that dark matter exists.
The claimant record, digitisedDAMA/NaILIBRA phase1LIBRA phase2this page's fit
156 points read from the vector drawing coordinates of two DAMA figures: 37 DAMA/NaI points (1995 to 2002) from the 2008 paper and 119 DAMA/LIBRA points (2003 to 2019) from the 2021 paper. Vertical bars are DAMA's plotted errors, horizontal bars the plotted time bins. The amber curve is this page's free fit. The shaded years hold three later cycles (2021-12-02 to 2024-07-09) that enter DAMA's 2026 2-6 keV fits, but whose residuals DAMA has plotted only in other energy windows.
Recomputed in your browser from the 156 points
0.01011 ± 0.00074 cpd/kg/keV, 13.63 standard errors. Period 0.99861 ± 0.00221 DAMA years; maximum on day 141.9 ± 4.3 of 2012, 22 May.
One year and June 2 held fixed: 0.00993 ± 0.00074 cpd/kg/keV, 13.41 standard errors, χ² 129.9 for 155 degrees of freedom.
Anchor reproduced: the printed amplitude, its error and their ratio all come back within the declared tolerances, for the free fit and for the fixed one.
Check the reconstruction: printed against recomputed, and three conventions read out of DAMA's own record
Printed against recomputed, with the declared tolerance wherever one is declared
Quantity
Printed
Recomputed
Gap
Verdict
Amplitude, period and phase free
0.01014
0.01011
−0.00003
within 0.00006
Its standard error
0.00074
0.00074
0.00000
within 0.00001
Amplitude / error (printed 13.7 σ)
13.7
13.63
−0.07
within 0.15
Amplitude, one year and June 2 fixed
0.00996
0.00993
−0.00003
within 0.00006
Its standard error
0.00074
0.00074
0.00000
within 0.00001
Amplitude / error (printed 13.4 σ)
13.4
13.41
+0.01
within 0.15
χ² of the fixed fit, 155 degrees of freedom
130
129.89
−0.11
reported, no tolerance declared
Period, in DAMA years
0.99834
0.99861
+0.00027
0.41 printed σ away
Phase, days
142.4
141.9
−0.5
0.12 printed σ away
Phase error, days
4.2
4.3
+0.1
1.02 times the printed value
Period error, DAMA years
0.00067
0.00221
+0.00154
3.31 times the printed value
Three conventions had to be read out of DAMA's own record
DAMA's time axis counts a year as 365 days. DAMA's own figures draw June 2 every 365 days. The 56 dashed and dotted guide lines in the two claimant figures sit within 0.94 days of 152.5 + 365k (and 335 + 365k for the minima); against a 365.25-day year they drift up to 6.07 days. With a 365-day year the fixed fit returns 0.00993 and χ² 129.9 against the printed 0.00996 and 130; with 365.25 days, 0.00976 and 135.7.
A time bin is averaged, not sampled. DAMA describe their fitting function as averaged over each time bin: The fitting function has been derived from eq. (3) integrated over each time bin. Sampled at bin centres instead, the free amplitude comes back as 0.00982 and the fixed one as 0.00964, about 3% low. (Bernabei et al. 2003, p. 29)
The printed phase fits a count from the record's centre, not from 1995. Read literally, the printed phase is counted from 1 January 1995; the digitised record then gives 150.5 ± 14.5 days. Read as the day of year of the maximum in the record's central year (2012), it gives 141.9 ± 4.3 against the printed 142.4 ± 4.2. The paper does not say which reading it uses.
The one printed number that does not come back under any convention tried here is the period's error. Propagated from DAMA's plotted errors, the record gives 0.00221 DAMA years, and the interval over which the profile χ² rises by 1 gives ± 0.00195, against the printed 0.00067. The printed period errors of the three nested 2021 fits (0.00067, 0.00069 and 0.00075 years) barely change although the records they come from span 24, 16 and 8 years. Propagated the same way, this page's grow from 0.00221, 0.00282 and 0.00670, as errors computed from the data must when the record shrinks, so the printed ones come from a procedure this page cannot see.
The deciding control, in your hands
Would an independent detector have seen it?
Two collaborations built the same kind of detector to test DAMA directly: ANAIS-112 at the Canfranc Underground Laboratory in Spain, nine sodium iodide crystals taking data since August 2017, and COSINE-100 at Yangyang in South Korea. ANAIS-112 released its rates with simulated background templates, COSINE-100 its background-subtracted residuals. The button adds DAMA's claimed waveform, at its printed size, to a copy of those released records and runs the same fit again, ANAIS's background terms included. Nothing else changes: not the dates, not the errors, not the fit.
ANAIS-112, public rate reconstruction: nine detectors, 441 released 45-day rates, each detector with its own constant and simulated-background coefficient, refitted every time. It reproduces ANAIS's printed fit but is not a run of their RooFit χ² fit.
released record, rate minus its fitted backgroundthe doctored copy, refitted
Every released row enters the fit. For display only, ANAIS residuals are averaged across the nine detectors in each 45-day bin and COSINE residuals into 60-day bins, each after subtracting that fit's own background. The vertical range is the same as for DAMA's record above, −0.07 to +0.07 cpd/kg/keV.
Released record
0.00105 ± 0.00249 cpd/kg/keV, 0.42 standard errors
Selected copy
0.00105 ± 0.00249 cpd/kg/keV, 0.42 standard errors (nothing planted)
Change from the plant, plant alone
0.00000 cpd/kg/keV, 0.00 standard errors
Nothing planted. The untouched ANAIS record gives 0.42 standard errors, below the threshold of 3: this test does not demonstrate a modulation at that threshold.
Distance from the 3-standard-error threshold: −2.58 standard errors.
The untouched ANAIS amplitude sits 3.44 combined standard errors below DAMA's printed fixed-phase claim, 0.00996 ± 0.00074.
Grade A: the claim, at its printed size, planted in the control's own released data and run through the unmodified fit.With the floated waveform planted (365-day year, printed phase counted in the record's central year), the plant alone accounts for 3.95 standard errors in ANAIS-112 and 3.12 in COSINE-100; the doctored records as a whole read 4.37 and 4.77, because the untouched records already give 0.42 and 1.66, and only the plant's own part decides the test. At the declared 3-standard-error demonstration threshold, ANAIS clears it by 0.95 and COSINE clears it by 0.12; with the printed phase counted from 1 January 1995 instead, COSINE's plant alone gives 2.92. Neither clears 5, so under a 5-sigma standard this test is INCONCLUSIVE. A deterministic plant shows what this rate fit can see; it is not an empirical probability of detection, and it cannot vouch for event selection, quenching factors or the full detector response. COSINE's rows are already background-subtracted.
ANAIS's own 20,000 simulations with DAMA's modulation put the spread of their fitted amplitude at 2.50 ± 0.01 cpd/ton/keV with a bias of 0.01 ± 0.02, and the collaboration states its sensitivity to DAMA's signal in 2-6 keV as (4.1 ± 0.3) σ, the DAMA amplitude it used over that spread. This page's reconstruction has an error of 2.49 cpd/ton/keV, and the same ratio here is 4.10.
Exactly what is planted, and when
The floated plant is DAMA's printed free fit, run continuously and never restarted at New Year:
Here t is the control's date on DAMA's time axis, in days from 1 January 1995 with 365-day years, and 6347.4 is day 142.4 of 2012, the central year of DAMA's record. ANAIS dates count from 3 August 2017 and COSINE dates from 1 January 2016; both are converted before the waveform is evaluated. Because the printed period is a little shorter than a year, the planted maximum drifts earlier: 2016: 20 May; 2018: 19 May; 2020: 18 May; 2022: 17 May. Counting the printed phase from 1 January 1995 instead puts the 2018 maximum on 9 May; the check below shows what that does to the test.
The fixed plant is DAMA's separately printed fixed fit, 0.00996 cpd/kg/keV with a period of one year and a maximum on June 2, written in each collaboration's own cosine: ANAIS's public macro uses cos[6.2832 × 0.00274 × (t + 62.17)], COSINE's notebook cos[2π (t − 152.5)/365.25]. A plant that matches the fit's own template must add exactly its amplitude to the fitted coefficient, and it does (+0.00996). That identity is arithmetic; the error and the threshold are what make the test informative.
Both plants and both fits are evaluated at bin centres. A real modulation averaged over ANAIS's 45-day bins would be 2.5% smaller at those centres, over COSINE's 15-day bins 0.3%; planting and fitting the same way keeps the comparison like for like.
The verdict, dated
OPEN as of 22 September 2026
Rebuilt from its own plotted points, DAMA's 2021 amplitude returns as 0.01011 ± 0.00074 cpd/kg/keV against the printed 0.01014 ± 0.00074. Planted at that size in a copy of ANAIS-112's public rates, it adds 3.95 standard errors to the unmodified fit, while the untouched ANAIS record sits 3.44 combined standard errors below DAMA's fixed-phase claim. What causes the rhythm is OPEN.
The scientific record, not this page, sets the category. DAMA's 2026 paper reports the rhythm persisting in its own data, now at 15.3 σ. ANAIS-112 measures 1.1 ± 2.5 cpd/ton/keV in the same 2-6 keV window, incompatible with DAMA at 3.5 σ (4.0 σ in 1-6 keV). COSINE-100 measures (0.0053 ± 0.0031) cpd/kg/keV in 2-6 keV; its sharpest tension with DAMA is in 1-3 keV, 3.57 σ. Their combined paper gives a simple six-year combination of (0.0027 ± 0.0021) cpd/kg/keV in 2-6 keV, excluding DAMA's signal at 3.53 σ. At the Identification of Dark Matter workshop on 1 June 2026, ANAIS-112 showed preliminary results from its full 8.4-year exposure: best-fit amplitudes compatible with zero, incompatible with DAMA at 4.0 σ in 1-6 keV and 5.6 σ in 2-6 keV, with a stated sensitivity of 5.1 σ (slide 22, marked preliminary). These are conference slides, not a paper, and nothing on this page is computed from them. OPEN does not mean equal support for the competing explanations. It means the observation stands, its dark matter interpretation is under strong independent pressure, and its cause has not been established.
Busoni and colleagues, in their June 2026 version, find that, in the most physically motivated cases of a common nuclear-recoil signal, the tension between DAMA and these same-target detectors exceeds 5 σ, and that lowering it needs significant tuning. Their fit uses the earlier DAMA record, not the 2026 exposure, and it does not rule on every possible origin of the rhythm.
The claimants' reply
In the 2026 paper DAMA reanalysed their final three continuous cycles with one common background across all three instead of a constant per cycle, and report amplitudes consistent with their standard analysis (χ²/d.o.f. = 8.4/20), slightly higher on average. They write that the observed DAMA annual modulation result cannot be mimicked by an artifact of the analysis method, and, of ANAIS-112 and COSINE-100, that some of their conclusions, including the intrinsically uncertain joint combinations, should be interpreted in light of the differences in detector performance, shielding and environmental conditions, electronics, data-taking procedures, and analysis methods, referring to earlier work for quenching factors (pp. 113-116). These are the authors' arguments. Their check does not by itself show that those differences account for the independent results.
What would change the verdict
An independent sodium iodide detector observing the same amplitude, energy dependence and calendar phase, with its detector response controlled, would support a common physical signal.
An identified instrumental or background cause that quantitatively reproduces DAMA's rate and passes DAMA's internal checks would establish an explanation of the rhythm.
A response-calibrated same-target exclusion strong enough to close a specified dark matter interpretation would close that interpretation, without explaining where DAMA's rhythm comes from.
The DAMA project was concluded as planned and its experimental setups were dismantled in fall 2024, so new evidence must come from other detectors, from reanalysis of what DAMA recorded, or from DAMA's crystals run again by others: a June 2026 ANAIS-112 talk reported that INFN and the Gran Sasso laboratory have taken responsibility for the DAMA/LIBRA detectors and plan to restart the measurements within an international collaboration (slide 23). COSINUS, a cryogenic sodium iodide experiment at the same laboratory, reported in April 2026 that data taking with eight detector modules is planned to begin in 2026. Decided by: not yet. Time to decision: not measurable, because there has been no decision.
A second layer: the method, run on nothing
Where does your year begin?
DAMA's residuals are the rate minus its average over each annual cycle. Buttazzo, Panci, Rossi and Strumia pointed out that this turns a slow, steady drift into a sawtooth, and that a sawtooth's first harmonic looks like a yearly cosine: if cycles start around the beginning of September and the background rises, the harmonic peaks at the start of June. Try it on a record that contains no annual rhythm at all.
The constructed rate is a straight line sampled at the 156 digitised DAMA times, with their gaps, phases and weights. It is not DAMA's background: DAMA has not published the 2-6 keV total rate behind the residuals analysed here as a time series, and its 2026 paper plots per-detector rates, in 3-4 keV and 5-day bins, only for its last three cycles. Annual subtraction induces an annual component in this constructed record; that sentence is about the operation, not about DAMA's measurements.
Year starting on day 243 (1 September): annual subtraction induces +0.00973 cpd/kg/keV on the June 2 cosine; free phase 0.00973, maximum on day 154.8 (4 June). A true annual signal of 0.00996 keeps 99.3% of itself.
Constructed total rate: 2.00 cpd/kg/keV plus +0.031 per year, a change of +0.735 across the digitised record. Ideal sampling limit for this slope: 0.00987 at the most favourable boundary, +0.00987 at this one.
induced from a straight line
Solid: annual subtraction at the digitised DAMA cadence, for all 365 year starts. Dashed grey: the ideal-sampling limit. Dotted amber: DAMA's printed fixed amplitude, plus and minus. Shaded: the days on which DAMA/LIBRA's 15 published cycles began. With the continuous fit selected, the annual curves fade and the marker sits on the continuous result.
kept from a true annual signal
A true signal of 0.00996 cpd/kg/keV peaking on June 2, put through the same subtraction: the fraction of its fitted amplitude that survives.
At DAMA/LIBRA's own 15 published cycle dates (119 LIBRA points, cycles starting between 17 July and 12 November), the same slope induces +0.00874 cpd/kg/keV on the June 2 cosine with its maximum on 14 June, and a true annual signal keeps 99.3%. To induce all of DAMA/LIBRA's printed fixed-phase amplitude, 0.00941, at these dates would take a slope of +0.0334 cpd/kg/keV per year; the best sawtooth fit of Buttazzo and colleagues to DAMA/LIBRA (phase1 and the first six phase2 cycles), +0.0219 per year, induces +0.0062 here.
At the default slope, +0.031 cpd/kg/keV per year (the growth Buttazzo and colleagues simulated, π × 0.01 per year, rounded), the induced amplitude runs from −0.00909 (years starting 26 February) to +0.00973 (years starting 1 September), and 43 of the 365 possible year starts induce at least 0.00922, one printed standard error below DAMA's fixed amplitude. A true signal keeps between 89.4% and 99.5% of itself. A falling background flips every sign: what a rising one adds to a June maximum, a falling one takes away. These are this page's calculations on constructed rates.
The table holds all 365 year starts at the current slope: the induced signed and free-phase amplitudes, the day of the induced maximum, the ideal-sampling value, and the kept fraction of a true annual signal. It characterises a mathematical operation on constructed rates. It does not estimate how often DAMA's analysis would find a false signal, and it does not infer DAMA's background.
See the constructed residuals, and the exact operationEach point sits at a digitised DAMA time; the rate is constructed, not observed. The amber curve is the June 2 cosine fitted to these residuals. This plot always shows annual subtraction.
Start at 2.00 cpd/kg/keV and add the slope times the elapsed years. Within each experimental phase, points whose centres share floor[(day − start)/year] form one constructed year; the inverse-variance weighted mean of each year is subtracted, using DAMA's plotted errors as weights. The same cosine estimator as the anchor then fits the residuals with the period fixed at one year and the maximum at June 2, and a second fit frees the phase. The continuous comparison fits a constant, a straight line and the cosine to the unsplit rates, and returns zero for any straight line.
For perfectly uniform sampling of whole years, the sawtooth's first harmonic has magnitude |b|/π, where b is the change per year, and its signed projection on the June 2 cosine is (b/π) sin[2π(start − 152.5)/year]; the dashed curve is that formula. The irregular DAMA cadence, its gaps and its unequal weights make the solid curve differ from it.
For DAMA's own cycle dates, each LIBRA point joins the cycle holding the larger part of its plotted time bin; 17 of the 119 plotted bins reach outside any single cycle, because DAMA's bins can span the short gaps between cycles. DAMA/NaI cycle dates are printed only to the month, so the NaI points are left out of that calculation.
The objection has its own counterevidence
Messina, Nardecchia and Piacentini compared the cosine with Buttazzo's sawtooth on DAMA's published 2-6 keV residuals and found the sawtooth disfavoured in every comparison except DAMA/NaI alone, with Bayes factors for the cosine from about 25 (LIBRA phase1 alone) to 108.9 (the whole record), under their priors and models. James, Rule, Barberio and colleagues built a toy model of DAMA/LIBRA with varying tritium and lead-210 activities and argue that a number of aspects of the DAMA signal are incompatible with an induced modulation arising from decays of background isotopes over the lifetime of the experiment. COSINE-100 applied a DAMA-like yearly subtraction to its own data in 2023 and found a significant modulation with almost the opposite phase, from a background that falls with time; that study also changed the event selection, and its data are available on request, so the released residuals used above do not reproduce it. DAMA answered Buttazzo and that COSINE study in 2022: each annual cycle starts before the expected December minimum and ends after the June maximum, so a decaying isotope under a constant background per cycle, they write, may only lead to underestimate the DM annual modulation amplitude, and COSINE's reversed phase is a negative amplitude at June 2, as a rate falling with time predicts. DAMA's 2026 common-background reanalysis, above, is the claimants' newest counterevidence. Buttazzo and colleagues themselves wrote that their most extreme case appears mildly disfavoured by a detailed study of the available data, but cannot be safely excluded, and their best sawtooth fit to DAMA/LIBRA needed a slope of (0.0219 ± 0.0026) cpd/kg/keV per year.
The claimants' full pipeline, run on nothing: not available. DAMA's raw detector rates, its per-detector subtraction and its noise structure cannot be recovered from plotted residuals, so this page cannot report how often DAMA's complete method would find an effect in data that contain none. The constructed experiment above is a deterministic property of one operation. The scientific verdict has not been decided by this mechanism.
On novelty: we searched the Wasteland corpus, arXiv-indexed web results and GitHub-indexed web results on 2026-09-22 and did not find an interactive page joining the 2021 DAMA residual reconstruction, injections into the public ANAIS and COSINE rate fits, and a paired boundary sweep of induced modulation and true-signal retention.
The check
Every number, and every free choice
Every rate fit on this page, DAMA's, ANAIS's, COSINE's and the constructed ones, goes through one weighted least-squares routine. Quoted errors are treated as independent standard deviations and the fit covariance is never rescaled to the scatter. The covariance between DAMA's plotted bins is not published and is not modelled.
The anchor's tolerance, and where it comes from
256 perturbations (seed 20260922) moved every marker, error-bar end, bin end and calibration tick within its half-unit drawing cell. The largest changes were 0.000041 (free amplitude), 0.000041 (fixed amplitude), 0.0000024 (errors), 0.057 (ratios) and 0.95 (χ²). 254 of 256 perturbed transcriptions (99.2%) still reproduce the anchor. The others miss on the fixed amplitude, by at most 0.000010 beyond the tolerance; unperturbed, the gap there is already −0.00003. Half the last printed digit plus the largest movement gives a budget of 0.000046, 0.0000074 and 0.107; the declared tolerances 0.00006, 0.00001 and 0.15 cover it.
The paragraph above is a verified snapshot (256 trials, seed 20260922); the button draws a new seed in this browser, so its result can differ.
The tolerance does not bound the one approximation that cannot be measured: that exposure was uniform inside each plotted bin. One DAMA/NaI point (NaI2008-020) has no horizontal stroke. Every other bin bar starts at the edge of the 16-unit marker, so a bin narrower than the marker leaves nothing to draw: this one is presumably at most about 28 days wide. Giving it the full 28 days moves the free amplitude by −5 × 10⁻⁹ and the fixed one by −2 × 10⁻⁸ cpd/kg/keV.
Every other printed fit the digitised record can be held to
The combined paper's abstract prints 0.0021 for the six-year 2-6 keV error and one body paragraph prints 0.0019; the conflict is left as printed. Its 3.53 σ (3.5 σ in the PDF's rounding) is quoted, not recomputed from rounded inputs, and its main joint MCMC uses the three-year records, (0.0021 ± 0.0028), which the paper finds incompatible with DAMA's signal at 2.6 σ.
ANAIS's 18 printed nuisance parameters, 2-6 keV, against this page's reconstruction
Detector
Background index, printed
Here
Template fraction f, printed
Here
Detector 0
4.396 ± 0.007
4.396
0.97 ± 0.03
0.97
Detector 1
4.438 ± 0.007
4.437
0.98 ± 0.03
0.98
Detector 2
2.334 ± 0.005
2.334
0.97 ± 0.04
0.97
Detector 3
3.064 ± 0.006
3.064
0.94 ± 0.03
0.94
Detector 4
2.871 ± 0.005
2.871
0.95 ± 0.03
0.95
Detector 5
2.805 ± 0.005
2.805
0.80 ± 0.03
0.80
Detector 6
2.572 ± 0.005
2.572
0.99 ± 0.02
0.99
Detector 7
2.337 ± 0.005
2.337
1.05 ± 0.03
1.05
Detector 8
1.956 ± 0.004
1.956
0.99 ± 0.03
0.99
ANAIS-112 supplemental Table II. The background index is the mean fitted background rate, cpd/kg/keV; f is the share carried by the collaboration's simulated background template. A linear fit with nine constants, nine template coefficients and one cosine reproduces all of them, which is why this page calls it a reconstruction of the published fit rather than an approximation to it. It is still not a run of their RooFit χ² fit.
Every plant, in both controls, through the unmodified fit (default conventions)
Control and plant
Planted amplitude
Recovered
Change
Plant alone, standard errors
Doctored record, standard errors
Plant alone clears 3
Plant alone clears 5
ANAIS, floated 2021 waveform
0.01014
0.01087 ± 0.00249
+0.00982
3.95
4.37
yes
no
ANAIS, fixed 2021 waveform
0.00996
0.01101 ± 0.00249
+0.00996
4.01
4.43
yes
no
COSINE, floated 2021 waveform
0.01014
0.01497 ± 0.00314
+0.00977
3.12
4.77
yes
no
COSINE, fixed 2021 waveform
0.00996
0.01515 ± 0.00314
+0.00996
3.18
4.83
yes
no
Plant alone: the change divided by the refitted error, which is what runControl returns on a copy that holds only the plant. The doctored record's total also carries the released record's own estimate, so it never decides the test.
The floated plant under each reading of DAMA's calendar
Phase reading and year
ANAIS change
ANAIS plant alone
ANAIS clears 3
COSINE change
COSINE plant alone
COSINE clears 3
central-year maximum, 365-day year
+0.00982
3.95
yes
+0.00977
3.12
yes
central-year maximum, 365.25-day year
+0.00981
3.95
yes
+0.00976
3.11
yes
1995 origin, 365-day year
+0.00922
3.71
yes
+0.00915
2.92
no
1995 origin, 365.25-day year
+0.00921
3.70
yes
+0.00913
2.91
no
The fixed plant is untouched by these readings: it always peaks on June 2 in each collaboration's own convention.
Every free choice, and what it moves
Each is a live parameter of the engine; the verifier checks that each one moves what it names
Choice
Options
Default
Moves
Independent experiment
ANAIS, COSINE
ANAIS
control rate series, fitted amplitude, fitted error, injection result
Each request is sent through the gate that guards the real calculation. Each has an allowed twin that differs only in the offending input, and the verifier runs both.
What no input here can give:
A particle mass or an interaction cross-section: these are electron-equivalent rate fits.
An exclusion of every nuclear-recoil or dark matter interpretation: that needs detector response and quenching factors.
How often DAMA's analysis would produce a false signal: the constructed record is deterministic and DAMA's raw rates are unpublished.
A live-time exposure: a plotted time-bin width is not one.
Rates from raw counts: that needs the selection efficiency and the live time.
The apparatus
What this rests on
An underground detector in Italy counted slightly more events every year around June, and its team attributes the rhythm to dark matter. Rebuild their published result, then add the same rhythm to two other detectors' records to see whether those could have found it. This page cannot say what causes the rhythm or identify any particle.
Every figure on this page is computed from frozen local files, in your browser, by engine.js; the text above was written from the same calculation and a verifier recomputes every figure and fails on any disagreement, including a planted signal that goes missing, an epoch moved by months, a detector's first row deleted and a claimant point nudged off its place. A separate NumPy and SciPy implementation cross-checks the fits.
DAMA, 2021: the claim reconstructed here. R. Bernabei, P. Belli, V. Caracciolo, R. Cerulli, V. Merlo, F. Cappella, A. d'Angelo, A. Incicchitti, C. J. Dai, X. H. Ma, X. D. Sheng, F. Montecchia and Z. P. Ye (2021). The dark matter: DAMA/LIBRA and its perspectives. Proceedings manuscript for the Sixteenth Marcel Grossmann Meeting, pp. 1-20, arXiv:2110.04734v1. Abstract, Table 1 (PDF p. 5), Figure 2 (PDF p. 6), Table 2 (PDF p. 7). CC BY 4.0.
DAMA, 2008: the DAMA/NaI points. R. Bernabei et al., DAMA Collaboration (2008). First results from DAMA/LIBRA and the combined results with DAMA/NaI. European Physical Journal C 56, 333-355. DOI 10.1140/epjc/s10052-008-0662-y. arXiv:0804.2741v1, Figure 2 lower panel and its caption (PDF p. 8), text p. 9. Article under CC BY-NC 2.0: only numbers are transcribed here.
DAMA, 2003: how the fit treats a time bin. R. Bernabei et al. (2003). Dark Matter search. La Rivista del Nuovo Cimento 26(1), 1-73. arXiv:astro-ph/0307403v1, section 6.1, pp. 28-29.
DAMA, 2013: the LIBRA-phase1 cycle dates. R. Bernabei et al. (2013). Final model independent result of DAMA/LIBRA-phase1. European Physical Journal C 73, 2648. arXiv:1308.5109v2, Table 1.
DAMA, 2026: the continuing record and the reply. R. Bernabei, P. Belli, A. Bussolotti, F. Cappella, V. Caracciolo, R. Cerulli, C. J. Dai, A. d'Angelo, A. Incicchitti, A. Leoncini, X. H. Ma, A. Mattei, V. Merlo, F. Montecchia, X. D. Sheng and Z. P. Ye (2026). Model-independent results from DAMA/LIBRA-phase2-empowered. Nuclear Physics and Atomic Energy 27(2), 105-117. DOI 10.15407/jnpae2026.02.105. Published online 25 June 2026. Abstract p. 105, Table 2 p. 110, section 7 pp. 113-115, conclusions pp. 115-116. CC BY-NC 4.0.
ANAIS-112, 2025: six years, same target. J. Amaré et al., ANAIS-112 Collaboration (2025). Towards a Robust Model-Independent Test of the DAMA/LIBRA Dark Matter Signal: ANAIS-112 Results with Six Years of Data. Physical Review Letters 135, 051001. DOI 10.1103/ntnl-zrn9. arXiv:2502.01542v2 (revised 30 July 2025), Table I and Table II (PDF p. 4); supplemental Table II (PDF p. 12).
ANAIS-112 public rates. ANAIS Collaboration (2025). ANAIS112-6years. Dark Matter Data Center, GitLab project 12572: event_data 2-6 keV (45-day bins, nine detectors), simulated_background 2-6 keV (15-day bins), a112modFit.C. No redistribution licence is stated.
COSINE-100, 2025: the full dataset. N. Carlin et al., COSINE-100 Collaboration (2025). COSINE-100 full dataset challenges the annual modulation signal of DAMA/LIBRA. Science Advances 11, eadv6503. DOI 10.1126/sciadv.adv6503. arXiv:2409.13226v2, Table I.
COSINE-100 public residuals. COSINE-100 Collaboration. PublicData, COSINE100-6YearsModulation: EventRate_erc2-6.csv, PseudoUnderDama_erc2-6.csv, README and visualizer notebooks. CC BY 4.0.
ANAIS-112 and COSINE-100 combined, 2025. N. Carlin et al., COSINE-100 and ANAIS-112 Collaborations (2025). Combined Annual Modulation Dark Matter Search with COSINE-100 and ANAIS-112. Physical Review Letters 135, 121002. DOI 10.1103/9j7w-qp1c. arXiv:2503.19559v2, revised 22 September 2025. Abstract; the joint MCMC uses three-year records.
Busoni and colleagues, 2026 version. G. Busoni, J. M. Cornell, W. Handley, F. Kahlhoefer, A. Kvellestad, M. Pitts, L. Street, A. C. Vincent and M. White. DAMA/LIBRA and dark matter: decisive tension or contrived cancellation. arXiv:2510.05216v2, revised 24 June 2026. Accepted for publication; the final venue was not verified here.
Buttazzo, Panci, Rossi and Strumia, 2020. D. Buttazzo, P. Panci, N. Rossi and A. Strumia (2020). Annual modulations from secular variations: relaxing DAMA? Journal of High Energy Physics 2020(04), 137. DOI 10.1007/JHEP04(2020)137. arXiv:2002.00459. CC BY 4.0.
Messina, Nardecchia and Piacentini, 2020. A. Messina, M. Nardecchia and S. Piacentini (2020). Annual modulations from secular variations: not relaxing DAMA? Journal of Cosmology and Astroparticle Physics 04 (2020) 037. DOI 10.1088/1475-7516/2020/04/037. arXiv:2003.03340v2.
COSINE-100, 2023: an induced modulation. G. Adhikari et al., COSINE-100 Collaboration (2023). An induced annual modulation signature in COSINE-100 data by DAMA/LIBRA's analysis method. Scientific Reports 13, 4676. DOI 10.1038/s41598-023-31688-4.
DAMA, 2022: the reply to Buttazzo and to COSINE. R. Bernabei, P. Belli, F. Cappella, V. Caracciolo, R. Cerulli, C. J. Dai, A. d'Angelo, A. Incicchitti, A. Leoncini, X. H. Ma, V. Merlo, F. Montecchia, X. D. Sheng and Z. P. Ye (2023). Dark Matter: DAMA/LIBRA and its perspectives. SciPost Physics Proceedings 12, 025. DOI 10.21468/SciPostPhysProc.12.025. arXiv:2209.00882v1 (2 September 2022), section 3, pp. 6-7. CC BY 4.0.
James, Rule, Barberio and colleagues, 2025. R. S. James, K. Rule, E. Barberio and 36 co-authors (2025). The DAMA/LIBRA signal: an induced modulation effect? arXiv:2408.08697v2, revised 28 March 2025. A journal version was not verified here.
ANAIS-112 at IDM 2026: preliminary, full exposure. I. Coarasa Casas, on behalf of the ANAIS team (2026). ANAIS-112: Towards Completing the Model-Independent Test of the DAMA/LIBRA Signal. Oral presentation, 16th International Workshop on the Identification of Dark Matter (IDM 2026), Zaragoza, 1 June 2026. Slides IDM2026_ANAIS_IvanCoarasa_vdef.pdf: slide 22 (PDF p. 28), marked preliminary, for the full-exposure results; slide 23 (PDF p. 29) for the DAMA/LIBRA detectors. Conference slides, not a refereed paper.
COSINUS, 2026. The COSINUS Collaboration (2026). COSINUS: a model-independent challenge to the DAMA/LIBRA dark matter claim using cryogenic NaI detectors. Communications Physics 9, 213. DOI 10.1038/s42005-026-02620-9. Published 23 April 2026.
Source terms were read on 22 September 2026. DAMA's 2021 manuscript and the COSINE-100 repository are CC BY 4.0. DAMA's 2008 article is CC BY-NC 2.0 and its 2026 article CC BY-NC 4.0: only transcribed numbers and short quotations appear here. The ANAIS-112 repository states no redistribution licence; this page ships a newly formatted numeric extract of its rates and templates, credits the collaboration, and asserts no licence over their data. The NOTICE file gives every term and every change.
Loading the frozen local records. Until they load, every figure shown is the verified snapshot.