Compact v0.3.0C1HF2 / FORK1: next epoch 1,863 · 10,796 committed candidates · 4,576 basins · 110 qualified F2 basins · 14,904 committed leases.
Since the epoch-1807 fork: +56 epochs · +324 candidates · +107 basins · +5 F2 basins across Al₂SiO₅, SiO₂, Mg₂SiO₄, MgO and Ca₂SiO₄. New ML-best screening energies were also recorded for Mg₂SiO₄, CaAl₂O₄ and Ca₂SiO₄.
Interpretation: the search is still producing new qualified regions and better ML screening results. DCC/L2 materially changes the trajectory, but this snapshot does not establish superiority over matched comparators. Atlas has 0 promoted rules in the current lineage and matched transfer evidence remains open.
Claim boundary: ML screening only; no stable-material, novelty, DFT, hull or usefulness claim is promoted. Cost-efficiency claims are also held back until a coordinator recovery-accounting anomaly is reconciled. 22 Sep evidence JSON →
Operativna posodobitev · 12:46: neposredni Work 8zCockpit status kaže native W8 proces ALIVE v RUN_38 persistent INSPECT resume. To je dokaz živosti procesa, ne nov znanstveni checkpoint; javni števci zato ostajajo na zgornjem preverjenem posnetku pri epohi 1.863.Operational update · 12:46: direct Work 8zCockpit status shows the native W8 process ALIVE in RUN_38 persistent INSPECT resume. This proves process liveness, not a new scientific checkpoint; the public counters therefore remain on the verified snapshot above at epoch 1,863.
Arena, ki se iz iskanja materialov uči, kako iskati naprej. MDL×DCC izbira prihodnje delo; Atlas naučene vzorce preizkuša na novih sestavah.An arena that learns from materials search how to search next. MDL×DCC selects future work; the Atlas tests learned patterns on other compositions.
Glavna veja je B6/H1: hitra zanka iskanja je povezana s počasnejšo zanko učenja Atlasa. MDL (Minimum Description Length) ocenjuje, kateri modeli in vzorci gospodarno pojasnjujejo opažanja. DCC (Dynamic Complexity Controller) vpliva na izbiro nadaljnjih akcij. MACE ocenjuje energije in sile kandidatnih struktur.The mainline is B6/H1: a fast search loop is coupled to a slower Atlas learning loop. MDL (Minimum Description Length) evaluates which models and patterns explain observations economically. DCC (Dynamic Complexity Controller) influences subsequent action selection. MACE evaluates candidate structures through energies and forces.
Odloča o nadaljevanju posamezne relaksacije: še dela, vrne kandidata v obdelavo ali zaključi porabo.Decides how to continue an individual relaxation: do more work, return a candidate for further processing, or end the allocation.
Primerja napovedi za dovoljene akcije in razporeja delo med sestave, generatorje ter reprezentacije.Compares predictions for legal actions and allocates work across compositions, generators and representations.
Atlas oblikuje pravila, jih rangira in odpira za preizkus. Njegove odločitve lahko spremenijo prihodnji nabor izvedljivih akcij.The Atlas forms and ranks rules, making them available for probing. Its decisions can change the future set of executable actions.
Opažanje → naučeno pravilo → dejanska nova akcija → primerjalni preizkus prenosa → obračun celotnega stroška → promocija ali ohranjena zavrnitev s pogojem ponovnega odprtja.Observation → learned rule → actual new action → matched transfer validation → full cost accounting → promotion or a retained rejection with a reopening condition.
Slabši rezultat zmanjša prednost. Arhiv ohrani razlog, dokaz in možnost ponovnega odprtja. Omejen aktivni Atlas lahko tako ostaja živ tudi po zapolnitvi.A weaker result lowers priority. The archive retains the reason, evidence and opportunity to reopen. A bounded active Atlas can keep learning after it fills up.
Napake pri obnovi, nepravilni sestavi ali prenosu pravil postanejo trajni regresijski testi in preizkusi z znanim pravilnim izidom.Resume errors, composition violations and rule-transfer failures become durable regression tests and canaries with known expected outcomes.
| VejaArm | VlogaRole |
|---|---|
| B2 | Običajni stroškovno občutljiv UCB comparator.Ordinary cost-aware UCB comparator. |
| B5 | MDL-M + DCC-S |
| B6/H0 | B6 z izključeno H1 Atlas zanko; kontrola učinka učenja.B6 without the H1 Atlas loop; a control for the learning effect. |
| B6/H1 | Glavna native večnivojska raziskovalna veja.The main native multilevel research arm. |
| H2 | ssMDL×DCC kot omejen eksperiment.ssMDL×DCC as a bounded experiment. |
Napoved je zabeležena pred izidom, logične epohe so deterministične, učenje se knjiži natanko enkrat. Primerjave morajo imeti iste informacije in primerljive proračune. Ta posnetek sam ne dokazuje prednosti pred B2 ali H0.Predictions are recorded before outcomes, logical epochs are deterministic, and learning is committed exactly once. Comparisons require the same information and comparable budgets. This snapshot alone does not establish superiority over B2 or H0.
FORK1 je podedoval stanje pri epohi 1807: 10.472 kandidatov, 4.469 basinov in 105 F2. Naslednjih 56 epoh je dodalo 324 kandidatov, 107 novih basinov in 5 novih F2 basinov.FORK1 inherited the epoch-1807 state: 10,472 candidates, 4,469 basins and 105 F2. The next 56 epochs added 324 candidates, 107 new basins and 5 new F2 basins.
Od 14.904 opaženih odločitev je DCC-S spremenil izbrano akcijo 9.118-krat (61,2%). Statistics path se je od surovega rankinga razlikoval 12.671-krat (~85%). L2/Atlas je izbral 330 akcij in 411-krat spremenil prihodnji legal action space.Across 14,904 observed decisions, DCC-S changed the selected action 9,118 times (61.2%). The statistics path differed from raw ranking 12,671 times (~85%). L2/Atlas selected 330 actions and changed the future legal action space 411 times.
Trenutni FORK1 Atlas beleži 377 ustvarjenih pravil, 328 zavrnitev, 231 reopen dogodkov in 592 arhivskih dogodkov. Aktivnih probe pravil je 16, promoviranih pa 0.The current FORK1 Atlas records 377 rules created, 328 rejections, 231 reopen events and 592 archive events. There are 16 active probe rules and 0 promoted rules.
Novi segment ima cap približno 1.035.236 CPU-core-s in zabeleženih približno 231.832 CPU-core-s porabe, vendar je ~228 tisoč core-s pripisanih kategoriji coordinator_preflight_replay_io_unclassified. Zato trenutni AUC/CPU ni pošten dokaz učinkovitosti algoritma.The new segment has a cap of about 1,035,236 CPU-core-s and about 231,832 CPU-core-s recorded as spent, but roughly 228k core-s are attributed to coordinator_preflight_replay_io_unclassified. Current AUC/CPU therefore is not a fair algorithm-efficiency measure.
Search signal je pozitiven: po forku nastajajo novi F2 basini in novi ML-best rezultati. Vendar še nimamo matched dokaza, da B6/H1 MDL×DCC/Atlas premaga močan običajni comparator.The search signal is positive: new F2 basins and new ML-best results appear after the fork. But there is not yet matched evidence that B6/H1 MDL×DCC/Atlas outperforms a strong ordinary comparator.
FORK1 ni ostal le pri novih basin ID-jih. V novem segmentu so se pojavili tudi novi najboljši ML-screening energijski rezultati za več kemij. To je signal za nadaljnjo validacijo, ne trditev o stabilnosti.FORK1 did not merely add basin IDs. The new segment also produced new best ML-screening energy results for several chemistries. This is a signal for further validation, not a stability claim.
Nov ML-best rezultat v FORK1 in eden od kemijskih prostorov z novim F2 basinom.A new ML-best result in FORK1 and one of the chemistries with a new F2 basin.
Nov ML-best rezultat; isti novi segment je ustvaril tudi nov F2 basin za Ca₂SiO₄.A new ML-best result; the same new segment also produced a new F2 basin for Ca₂SiO₄.
Nov najboljši opaženi ML-screening rezultat v tej sestavi v trenutnem lineage-u.A new best observed ML-screening result for this composition in the current lineage.
Al₂SiO₅ · SiO₂ · Mg₂SiO₄ · MgO · Ca₂SiO₄
Najbolj zanimive screening kandidate je treba zamrzniti kot validation set in jih preveriti z močnejšim neodvisnim modelom oziroma DFT. Ločeno mora arena izvesti matched fresh campaigns proti comparatorju z enakim informacijskim in računskim proračunom.The most interesting screening candidates should be frozen as a validation set and checked with a stronger independent model or DFT. Separately, the arena must run matched fresh campaigns against a comparator with the same information and compute budget.
Arena danes hrani lasten Atlas in primerja strukture znotraj runa. Pripravljeni so vmesniki za referenčne zbirke, vendar ta run še nima priklopljene napolnjene zunanje knjižnice. MACE je model energije in sil; ni katalog za potrjevanje novosti.The arena currently keeps its own Atlas and matches structures within the run. Reference-data interfaces are prepared, but this run has no populated external library connected. MACE is an energy and force model; it is not a novelty-verification catalogue.
Trenutni javni posnetek temelji na compact FORK1 FAST-RICH Live Extractu z naslednjo epohalno mejo 1863. Extract ima PARTIAL_HISTORY pokritost: dovolj za trenutno stanje, agregate, Atlas/policy/controller evidence in bounded recent history, ne pa za polni replay vseh zgodovinskih dokumentov.The current public snapshot is based on the compact FORK1 FAST-RICH Live Extract with next-epoch boundary 1863. Coverage is PARTIAL_HISTORY: enough for current state, aggregates, Atlas/policy/controller evidence and bounded recent history, but not a full replay of every historical document.
8zMaterials_HF2_FORK1_LIVE_EXTRACT_20260922_124109_975911.zip
SHA-256: a3f616d5e6fa51b782b5ae0e83ea0d65cae05c23a57c4d685f9fd0959bd5b09d
SHA3-256: 2d1fd4f43d04a66ceb77f4deaa2aa3c9dbf45c6f21e69f20c0f7b2c065ccfdbf
5,41 MB outer bundle · ~14,29 MB nekompresiranega rich payload-a · run ID compact-fork-332e1f36e631fa8b9ffc980d · next_epoch 1863 · strict_original_budget_continuation=False.5.41 MB outer bundle · ~14.29 MB uncompressed rich payload · run ID compact-fork-332e1f36e631fa8b9ffc980d · next_epoch 1863 · strict_original_budget_continuation=False.
Snapshot 11.–13. septembra ostaja pomemben za zgodovino razvoja in zgodnje transfer poskuse, vendar ni več current status. Števci 39 F2, 260 pravil in zgodnji matched-transfer primer se zato ne smejo brati kot trenutno stanje FORK1.The 11–13 September snapshot remains important as development history and for early transfer experiments, but it is no longer the current status. Counts such as 39 F2, 260 rules and the early matched-transfer example should therefore not be read as current FORK1 state.