Signals & patterns
A signal is an event — a programme pauses, a net figure flips, a cliff approaches. A pattern is a measured regularity that survived a significance test. Both carry their evidence, because a signal you cannot check is a tip, not data.
Published patterns
A float that moves by less than its unlock is the ordinary case as often as it is an interesting one: tokens re-staked in the same week, or released to an address the schedule counts as circulating and the chain does not. The gap is measured; where it went is not. Measured over 4 quarters, 249 cliffs, from 8 Sep 2025 to 7 Sep 2026. The paired comparison gives q < 0.0001 after correcting across the whole family of 36 tests run that day; p < 0.0001 alone would not publish it.
Inputs and evidence
Window 2025-09-08 to 2026-09-07. p = 0.000, corrected to q = 0.000 across a family of 36 tests. Effect 0.736.
A programme whose budget was set once and is being spent down produces this shape on its own, and so does a fee burn following the chain's own activity. The trend is in the series; the reason for it is not. Measured over 180 days, 180 days, from 11 Mar 2026 to 7 Sep 2026. The Mann–Kendall gives q < 0.0001 after correcting across the whole family of 36 tests run that day; p < 0.0001 alone would not publish it.
Inputs and evidence
Window 2026-03-11 to 2026-09-07. p = 1.3e-15, corrected to q = 5.3e-15 across a family of 36 tests. Effect 0.401. 180 events read.
Sources: derived-chain (Chain events)
TRX
SHIB
CRO
UNIXRPRetired patterns
Exchange burn programmes are quarterly by design, so the timing is expected. What is measured here is the size of the concentration, not that it exists. Measured over 4 quarters, 3,321 burn events, from 26 Aug 2025 to 25 Aug 2026. The χ² vs uniform gives q < 0.0001 after correcting across the whole family of 52 tests run that day; p < 0.0001 alone would not publish it.
Retired 26 Aug 2026: the re-test could not run: the window no longer holds enough observations to support the test. Published 24 Aug 2026 and kept here with this note rather than removed — a finding that quietly disappears is indistinguishable from one that was never made.
Inputs and evidence
Window 2025-08-26 to 2026-08-25. p = 0.000, corrected to q = 0.000 across a family of 52 tests. Effect 0.890. 3321 events read.
Sources: derived-chain (Chain events), lcd (Chain state)
10EBA76A5C758BA32F6B0053626F1576C0AF6CFFCCA7FE61FC38BEC1FC42C7EC · 593FC06D26EABD9BB36AB65C23F73397E9E3078802ADD2AED455B40447FF5C1B · 5D8DBAE1A064AEDBBB315005D1EDF524BC412DBA1267159D1DC7F9B8D5D6BDCD · 2EDED561BDC731A731F4EFC2673C9D7E7012DD9B2631F3356333442B226CBA4E · 22D744B9E2BA302DC0CBC9F8118988304D8109CD230F29AE43C2BBA22D28C980
Exchange burn programmes are quarterly by design, so the timing is expected. What is measured here is the size of the concentration, not that it exists. Measured over 4 quarters, 708 burn events, from 26 Aug 2025 to 25 Aug 2026. The χ² vs uniform gives q < 0.0001 after correcting across the whole family of 52 tests run that day; p < 0.0001 alone would not publish it.
Retired 26 Aug 2026: the re-test could not run: the window no longer holds enough observations to support the test. Published 24 Aug 2026 and kept here with this note rather than removed — a finding that quietly disappears is indistinguishable from one that was never made.
Inputs and evidence
Window 2025-08-26 to 2026-08-25. p = 2.5e-76, corrected to q = 1.9e-75 across a family of 52 tests. Effect 0.496. 708 events read.
Sources: derived-chain (Chain events), lcd (Chain state)
F0C976D0137F68F33EFA378405E673C0104133068B463055F1EC627BC132AF33 · 53D05953405D7A1768E52197368B342596514575586346DE1D4018603CC9AE89 · 86CB32E29257A6D118FF57B157D44CDFA320B789AB7BBE9AAFCFA89F2CB23155 · 69A285F8A10C9244453E086BA4C22D1B2D75163386237494DE456713A9CA99B7 · 6C9664090217B90C957289BB08B1F12E5A9C4AC163182AFC9B38664445458637
A programme running on a weekly schedule produces this shape on its own, so the ordinary reading is an operations calendar rather than market behaviour. Measured over 12 weeks, 52 burn events, from 2 Jun 2026 to 25 Aug 2026. The χ² vs uniform gives q < 0.0001 after correcting across the whole family of 52 tests run that day; p < 0.0001 alone would not publish it.
Retired 26 Aug 2026: the re-test could not run: the window no longer holds enough observations to support the test. Published 24 Aug 2026 and kept here with this note rather than removed — a finding that quietly disappears is indistinguishable from one that was never made.
Inputs and evidence
Window 2026-06-02 to 2026-08-25. p = 8.8e-37, corrected to q = 5.7e-36 across a family of 52 tests. Effect 0.765. 52 events read.
Sources: chain-rpc (Chain state)
0xc9bfcb4f66104cc6fc6af3f7a1dde57a1156031b14d4fcb4cbe9c8f3fac9583c · 0xe08e0e851deb68eab6d33c4bd891497b91ba1a7873db30aec700c8235c14773b · 0x4dc7611979d9db6933471e99c8073fba1ed87f56ea11cf5cf62fb8a31b57a260 · 0x5f53d889da71f22e904c2194636556ec83549d491f98ee110010781afa5bd7f9 · 0x93a01247d39dbead2cbd77e01644d369b488d1653c8328d7496d6c4f882af12e
Exchange burn programmes are quarterly by design, so the timing is expected. What is measured here is the size of the concentration, not that it exists. Measured over 4 quarters, 53 burn events, from 26 Aug 2025 to 25 Aug 2026. The χ² vs uniform gives q < 0.0001 after correcting across the whole family of 52 tests run that day; p < 0.0001 alone would not publish it.
Retired 26 Aug 2026: the re-test could not run: the window no longer holds enough observations to support the test. Published 24 Aug 2026 and kept here with this note rather than removed — a finding that quietly disappears is indistinguishable from one that was never made.
Inputs and evidence
Window 2025-08-26 to 2026-08-25. p = 1.8e-22, corrected to q = 1.1e-21 across a family of 52 tests. Effect 0.972. 53 events read.
Sources: derived-chain (Chain events), chain-rpc (Chain state)
0xc9bfcb4f66104cc6fc6af3f7a1dde57a1156031b14d4fcb4cbe9c8f3fac9583c · 0xe08e0e851deb68eab6d33c4bd891497b91ba1a7873db30aec700c8235c14773b · 0x4dc7611979d9db6933471e99c8073fba1ed87f56ea11cf5cf62fb8a31b57a260 · 0x5f53d889da71f22e904c2194636556ec83549d491f98ee110010781afa5bd7f9 · 0x93a01247d39dbead2cbd77e01644d369b488d1653c8328d7496d6c4f882af12e
Exchange burn programmes are quarterly by design, so the timing is expected. What is measured here is the size of the concentration, not that it exists. Measured over 4 quarters, 43 burn events, from 26 Aug 2025 to 25 Aug 2026. The χ² vs uniform gives q = 0.0001 after correcting across the whole family of 52 tests run that day; p < 0.0001 alone would not publish it.
Retired 26 Aug 2026: the re-test could not run: the window no longer holds enough observations to support the test. Published 24 Aug 2026 and kept here with this note rather than removed — a finding that quietly disappears is indistinguishable from one that was never made.
Inputs and evidence
Window 2025-08-26 to 2026-08-25. p = 5.6e-5, corrected to q = 1.3e-4 across a family of 52 tests. Effect 0.477. 43 events read.
Sources: derived-chain (Chain events), chain-rpc (Chain state)
0x3313394800733dfbcd8df091aa581c232f9c89ba085f806cf21c8d80c4d54cfb · 0x2a9f14d9849df5e9d27de89e5c3b189e408439f29c4dd37d8fe487a9924c8ed3 · 0xc01b651dfa5617c2804c1b024309d3cbd29e26969f9b2af0c516f789b463fe62 · 0x3afe7bfff84cd130054754fe15e72f305818375a8efacdf4009272caca5bda89 · 0x93d90f64627bb47a05010d280ba8ed4502b1086af2ef143e6865189f76bb5619
One operator's scheduler produces this shape on its own. A clustered hour is a cron entry before it is anything else, and the chain cannot tell the two apart. Measured over 12 weeks, 1,328 burn events, from 1 Jun 2026 to 24 Aug 2026. The χ² vs uniform gives q < 0.0001 after correcting across the whole family of 51 tests run that day; p < 0.0001 alone would not publish it.
Retired 25 Aug 2026: the re-test could not run: the window no longer holds enough observations to support the test. Published 24 Aug 2026 and kept here with this note rather than removed — a finding that quietly disappears is indistinguishable from one that was never made.
Inputs and evidence
Window 2026-06-01 to 2026-08-24. p = 0.000, corrected to q = 0.000 across a family of 51 tests. Effect 0.565. 1328 events read.
Sources: chain-rpc (Chain state)
0xc9bfcb4f66104cc6fc6af3f7a1dde57a1156031b14d4fcb4cbe9c8f3fac9583c · 0xc9bfcb4f66104cc6fc6af3f7a1dde57a1156031b14d4fcb4cbe9c8f3fac9583c · 0xc9bfcb4f66104cc6fc6af3f7a1dde57a1156031b14d4fcb4cbe9c8f3fac9583c · 0xc9bfcb4f66104cc6fc6af3f7a1dde57a1156031b14d4fcb4cbe9c8f3fac9583c · 0xc9bfcb4f66104cc6fc6af3f7a1dde57a1156031b14d4fcb4cbe9c8f3fac9583c
Exchange burn programmes are quarterly by design, so the timing is expected. What is measured here is the size of the concentration, not that it exists. Measured over 4 quarters, 61 burn events, from 25 Aug 2025 to 24 Aug 2026. The χ² vs uniform gives q < 0.0001 after correcting across the whole family of 51 tests run that day; p < 0.0001 alone would not publish it.
Retired 25 Aug 2026: the re-test could not run: the window no longer holds enough observations to support the test. Published 24 Aug 2026 and kept here with this note rather than removed — a finding that quietly disappears is indistinguishable from one that was never made.
Inputs and evidence
Window 2025-08-25 to 2026-08-24. p = 3.2e-27, corrected to q = 1.5e-26 across a family of 51 tests. Effect 1.000. 61 events read.
Sources: etherscan-v2 (Chain events)
0x166624ce9f62f7fe5e61d977d8d8aa99ed332791cfcd398e53b445b05b63c9c2 · 0x364f83c70cf35f6697214e711303ba42cc4421f3634d66dbfcb43dabd1782c2e · 0x09c31ac93c32806908a66d7ddb86ef9c95bd7dc5b135c579aa980a377b5b6cfc · 0xf314101528bb6f954d31dde3758b325ad538426416d128e9e9c384852909f49f · 0xb50fc9845373e7708c0bcf093316d0ab43d1fbcf300397078b0c8d55ca8ad07f
The registered family
Registered 2026-08-18.
Registered 2026-08-18.
Registered 2026-08-18.
Registered 2026-08-18.
Registered 2026-08-18.
Registered 2026-08-20.
Registered 2026-08-22.
Registered 2026-08-22.
Registered 2026-08-22.
Signals for LUNC
Showing only what fired for Terra Classic. Show every token.
The 12-month projection for Terra Classic (LUNC) is now -0.0% of the measured float. The direction changed, which is the event that most often invalidates the reason somebody holds a token — the figure moving within one direction does not.
The 12-month projection for Terra Classic (LUNC) is now +0.0% of the measured float. The direction changed, which is the event that most often invalidates the reason somebody holds a token — the figure moving within one direction does not.
The ratio divided $2,021,354 of burns by $2.85 of issuance — everything LUNC added over twelve months. Nine days in ten the figure is absent because nothing was added at all; on the tenth a rounding error became a rate, and signal 74 published it at high severity as "LUNC burns now cover everything added". The arithmetic was right and the sentence was about the $2.85. A denominator of 9.8e-7% of float is below the materiality gate this methodology already applies everywhere else, and the ratio is withheld rather than restated: what share of nothing was covered has no answer. The headline is unmoved because it never used this ratio — net float is issuance and unlocks less burns over float, and none of those changed. Any signal that fired on the previous figure is superseded by this one. The correction is dated and stays on the record: publishing our own errors is the only version of this that a reader can rely on.
The release was smaller than the day's unrelated destruction, so the subtraction took the whole of it and wrote nothing: a reader saw no event rather than a wrong figure. The zero below is structural, not rounded — LUNC's figure is measured from two chain readings, so the ledger does not enter it; measured with the methodology, -0.071200 before and after. Any signal that fired on the previous figure is superseded by this one. The correction is dated and stays on the record: publishing our own errors is the only version of this that a reader can rely on.
Retired 26 Aug 2026: the re-test could not run: the window no longer holds enough observations to support the test. Published 24 Aug 2026 and kept here with this note rather than removed — a finding that quietly disappears is indistinguishable from one that was never made.
Retired 26 Aug 2026: the re-test could not run: the window no longer holds enough observations to support the test. Published 24 Aug 2026 and kept here with this note rather than removed — a finding that quietly disappears is indistinguishable from one that was never made.
Retired 24 Aug 2026: the corrected q rose to 0.05 on re-test, above the 0.10 gate. Published 18 Aug 2026 and kept here with this note rather than removed — a finding that quietly disappears is indistinguishable from one that was never made.
Exchange burn programmes are quarterly by design, so the timing is expected. What is measured here is the size of the concentration, not that it exists. Measured over 4 quarters, 3,877 burn events, from 25 Aug 2025 to 24 Aug 2026. The χ² vs uniform gives q < 0.0001 after correcting across the whole family of 51 tests run that day; p < 0.0001 alone would not publish it.
A programme running on a weekly schedule produces this shape on its own, so the ordinary reading is an operations calendar rather than market behaviour. Measured over 12 weeks, 3,602 burn events, from 1 Jun 2026 to 24 Aug 2026. The χ² vs uniform gives q < 0.0001 after correcting across the whole family of 51 tests run that day; p < 0.0001 alone would not publish it.
The coverage ratio for Terra Classic (LUNC) is 668234.10 — burns divided by issuance plus scheduled unlocks over the window. It crossed 1.00, which is the threshold at which a programme changes from slowing dilution to reversing it. The ratio moving inside one side of that line is not this event.
The test flags a position in the series and nothing else. A change of operator, a treasury decision and a contract migration all produce this shape, and this series cannot separate them. Measured over 180 days, 180 days, from 19 Feb 2026 to 18 Aug 2026. The Pettitt + CUSUM gives q < 0.0001 after correcting across the whole family of 28 tests run that day; p < 0.0001 alone would not publish it.
Subscribe to signals
All 7 are built and evaluated for every reader; 3 of them are deliverable to your free plan by email. Telegram, webhooks and an SSE stream are not built yet and are not sold.
Testing a wide family of hypotheses across dozens of tokens produces false findings by arithmetic, not by malice. Ask thirty questions at the 5% level and one or two should look significant with nothing behind them. Results are corrected across the whole test family — Benjamini–Hochberg, controlling the false discovery rate — and re-tested weekly rather than fixed at first sight.
Last run 2026-09-07: 36 tests across 56 tokens, corrected as one family of 36. Method v2.16.
At q < 0.10, roughly one published finding in ten is expected to be a coincidence. That is the advertised rate, not a claim of certainty.
No causal language. Copy is linted for "because", "causes", "leads to" and any price implication. We report when burns happen, never what it means for the price.
Mundane explanation first. Most weekday clustering is a cron job, not market behaviour, and that is stated up front rather than under the interesting reading.
Corrections propagate. If a figure behind a fired signal is corrected, the correction goes out on the channel that carried the original. Silence after an error is worse than the error.
How every figure is computed · Methodology