{"source":"docs/research/12-predictions.md","mapping":"results/predictions-outcomes.json","about":"Each prediction in docs/research/12-predictions.md mapped to where its outcome is written. Built 2026-11-12 by reading the record; nothing in the notes was changed. outcome_ref is file:line of the outcome's first line. verdict: pass | fail | mixed | pending | not_run | unlinked (no outcome that can be tied to it with certainty). 'basis' says how the verdict was read. Append-only: later outcomes are added as new entries under 'updates', never by editing these.","built":"2026-11-12","counts":{"missed":13,"unlinked":9,"confirmed":10,"mixed":18,"not_run":1,"pending":6},"predictions":[{"id":"P1","title":"Liber I.1 beats Liber I at every matched step to 50,000","written":"2026-10-22","verdict":"fail","bucket":"missed","measured":"2026-10-22","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:26","underHeading":null,"note":null,"also":null},{"id":"P2","title":"Liber I.1 ends above 35% greedy top-1","written":"2026-10-22","verdict":"unlinked","bucket":"unlinked","measured":null,"basis":null,"outcomeRef":"docs/research/12-predictions.md:45","underHeading":null,"note":"the record still reads 'pending'; no outcome is tied to P2 anywhere","also":null},{"id":"P3","title":"Productum ends between 66% and 72% greedy top-1","written":"2026-10-22","verdict":"pass","bucket":"confirmed","measured":"2026-10-26","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:61","underHeading":null,"note":null,"also":null},{"id":"P4","title":"The full PaRoutes n1 run confirms the 200-target sample within noise","written":"2026-10-22","verdict":"pass","bucket":"confirmed","measured":"2026-10-23","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:80","underHeading":null,"note":null,"also":null},{"id":"P5","title":"Ordo as Texere's route score raises top-1 route quality and does not raise solved rate","written":"2026-10-22","verdict":"unlinked","bucket":"unlinked","measured":null,"basis":null,"outcomeRef":"docs/research/12-predictions.md:100","underHeading":null,"note":"the record still reads 'pending'","also":null},{"id":"P6","title":"Beam search roughly doubles Liber's top-1 to top-10 spread","written":"2026-10-22","verdict":"unlinked","bucket":"unlinked","measured":null,"basis":null,"outcomeRef":"docs/research/12-predictions.md:111","underHeading":null,"note":"the record still reads 'pending'; beam results in 05-liber.md are not scored against P6","also":null},{"id":"P7","title":"The mechanism model is near-perfect inside its 63 classes and near-useless outside","written":"2026-10-22","verdict":"unlinked","bucket":"unlinked","measured":null,"basis":null,"outcomeRef":"docs/research/12-predictions.md:128","underHeading":null,"note":"the record still reads 'pending'; docs/research/15-mechanism.md:159 says the in-class half was still to be measured","also":null},{"id":"P8","title":"Productum's top-3 is far above its greedy top-1, and TTA helps a little","written":"2026-10-26","verdict":"mixed","bucket":"mixed","measured":"2026-10-26","basis":"derived here, not scored in the record","outcomeRef":"docs/research/16-productum-improvement.md:94","underHeading":null,"note":"on all 1,000 reactions top-3 (71.6%) is 11.8 points above top-1 (59.8%), just under the claimed 12; the interim 226 were inside it (14.2). TTA was measured on the 226 only: +2.7, inside the claimed +1 to +4. The record's own P8 line was never updated past 'pending'.","also":["docs/research/12-predictions.md:140","docs/research/16-productum-improvement.md:86"]},{"id":"P9","title":"Atom-order augmentation removes Motus's positional shortcut","written":"2026-10-26","verdict":"pass","bucket":"confirmed","measured":"2026-10-27","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:154","underHeading":null,"note":"'confirmed on the falsifiers, with one wording miss'","also":null},{"id":"P10","title":"Our own mapper agrees with RXNMapper on most held-out patent reactions, templates far better than MCS","written":"2026-10-26","verdict":"mixed","bucket":"mixed","measured":"2026-10-26","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:186","underHeading":null,"note":"'mostly confirmed, one band missed'","also":null},{"id":"P11","title":"Motus I.2: rule-generated textbook classes, fine-tuned from Motus I.1","written":"2026-10-27","verdict":"pass","bucket":"confirmed","measured":"2026-10-27","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:220","underHeading":null,"note":null,"also":null},{"id":"P12","title":"**Written 2026-10-29 09:30, before the run.** Liber I at its final checkpoint (400,000 steps), beam 10 on","written":null,"verdict":"unlinked","bucket":"unlinked","measured":null,"basis":null,"outcomeRef":null,"underHeading":null,"note":"no outcome recorded; docs/research/05-liber.md:102 says the run was moved to the server","also":null},{"id":"P13","title":"Texere with Liber at the first disconnection","written":"2026-10-29","verdict":"mixed","bucket":"mixed","measured":"2026-11-10","basis":"count of claims held vs missed (no single headline)","outcomeRef":"docs/research/12-predictions.md:263","underHeading":null,"note":"run on a different setup than written (stated in the outcome)","also":null},{"id":"P14","title":"The forward check as a metric","written":"2026-10-29","verdict":"unlinked","bucket":"unlinked","measured":null,"basis":null,"outcomeRef":null,"underHeading":null,"note":"no outcome recorded","also":null},{"id":"P15","title":"Probatio I","written":"2026-10-29","verdict":"mixed","bucket":"mixed","measured":"2026-10-29","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:320","underHeading":null,"note":null,"also":null},{"id":"P16","title":"Condicio I","written":"2026-10-30","verdict":"fail","bucket":"missed","measured":"2026-10-31","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:460","underHeading":"P21 (2026-10-30 15:00, before the run) Scriptum I text mode","note":"claims 2-3 never measured","also":null},{"id":"P17","title":"Imago I pilot","written":"2026-10-30","verdict":"pass","bucket":"confirmed","measured":"2026-10-30","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:408","underHeading":"P20 (2026-10-30 13:00, before the run) Imago I: hard photos and every notation together","note":null,"also":null},{"id":"P18","title":"Scriptum I label-mode pilot","written":"2026-10-30","verdict":"unlinked","bucket":"unlinked","measured":null,"basis":null,"outcomeRef":null,"underHeading":null,"note":"no outcome recorded","also":null},{"id":"P19","title":"Imago I on bad phone photos","written":"2026-10-30","verdict":"not_run","bucket":"not_run","measured":"2026-10-30","basis":null,"outcomeRef":"docs/research/12-predictions.md:389","underHeading":null,"note":"merged into P20; its claims carried into P20 unchanged","also":null},{"id":"P20","title":"Imago I: hard photos and every notation together","written":"2026-10-30","verdict":"mixed","bucket":"mixed","measured":"2026-10-31","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:444","underHeading":"P21 (2026-10-30 15:00, before the run) Scriptum I text mode","note":null,"also":null},{"id":"P21","title":"Scriptum I text mode","written":"2026-10-30","verdict":"pass","bucket":"confirmed","measured":"2026-10-30","basis":"count of claims held vs missed (no single headline)","outcomeRef":"docs/research/12-predictions.md:434","underHeading":null,"note":"claim 1 confirmed; claim 2 retracted as not measurable this way","also":null},{"id":"P22","title":"Condicio v2, conditions as five role slots","written":"2026-10-31","verdict":"mixed","bucket":"mixed","measured":"2026-10-31","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:496","underHeading":null,"note":"4 of 6","also":null},{"id":"P23","title":"telling Productum's right predictions from its wrong ones","written":"2026-10-31","verdict":"mixed","bucket":"mixed","measured":"2026-10-31","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:543","underHeading":null,"note":null,"also":null},{"id":"P24","title":"telling Productum's right predictions from its wrong ones","written":"2026-10-31","verdict":"mixed","bucket":"mixed","measured":"2026-10-31","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:552","underHeading":null,"note":null,"also":null},{"id":"P25","title":"telling Productum's right predictions from its wrong ones","written":"2026-10-31","verdict":"fail","bucket":"missed","measured":"2026-11-02","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:587","underHeading":"P27: Imago with pen ink in training (written 2026-11-02 17:40, before training)","note":null,"also":null},{"id":"P26","title":"Condicio v3, the same five slots read through our own encoders","written":"2026-11-02","verdict":"fail","bucket":"missed","measured":"2026-11-02","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:598","underHeading":"P27: Imago with pen ink in training (written 2026-11-02 17:40, before training)","note":null,"also":null},{"id":"P27","title":"Imago with pen ink in training","written":"2026-11-02","verdict":"pending","bucket":"pending","measured":"2026-11-02","basis":null,"outcomeRef":"docs/research/12-predictions.md:640","underHeading":"P28: Imago I (imago-mixed-2, as served) on REAL patent drawings, USPTO-30K (written 2026-1","note":"partial outcome only: claim 3 confirmed; claims 1, 2 and 4 not run","also":null},{"id":"P28","title":"Imago I (imago-mixed-2, as served) on REAL patent drawings, USPTO-30K","written":"2026-11-02","verdict":"mixed","bucket":"mixed","measured":"2026-11-02","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:624","underHeading":null,"note":null,"also":null},{"id":"P29","title":"Imago in the patent-figure style","written":"2026-11-02","verdict":"fail","bucket":"missed","measured":"2026-11-03","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:667","underHeading":"P30: Scriptum label mode retrained on tablet-style ink (written 2026-11-03 00:15, before t","note":"'MISSED on real patent drawings, 1 of 4'","also":null},{"id":"P30","title":"Scriptum label mode retrained on tablet-style ink","written":"2026-11-03","verdict":"pass","bucket":"confirmed","measured":"2026-11-03","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:682","underHeading":null,"note":null,"also":["docs/research/12-predictions.md:683"]},{"id":"P31","title":"Probatio I as Texere's step filter","written":"2026-11-03","verdict":"mixed","bucket":"mixed","measured":"2026-11-03","basis":"count of claims held vs missed (no single headline)","outcomeRef":"docs/research/12-predictions.md:707","underHeading":null,"note":"2 wrong, 1 right, 1 not re-measured","also":null},{"id":"P32","title":"Imago continued from mixed-4 with more patent-style drawings","written":"2026-11-03","verdict":"mixed","bucket":"mixed","measured":"2026-11-05","basis":"count of claims held vs missed (no single headline)","outcomeRef":"docs/research/12-predictions.md:848","underHeading":"P34: enantiomers merged when nothing chose between them (written 2026-11-05 18:30, before ","note":"claim 1 wrong, claim 2 right","also":null},{"id":"P33","title":"a purchasable molecule that is nearly the whole target is not a starting material","written":"2026-11-05","verdict":"mixed","bucket":"mixed","measured":"2026-11-05","basis":"count of claims held vs missed (no single headline)","outcomeRef":"docs/research/12-predictions.md:816","underHeading":"P34: enantiomers merged when nothing chose between them (written 2026-11-05 18:30, before ","note":"tied to P33 by its raw file (results/texere/n1-p33-advanced.json) and its subject (the advanced-intermediate rule). Claim 1 right, claim 2 'passes as written and fails in spirit', claim 3 not answered","also":null},{"id":"P34","title":"enantiomers merged when nothing chose between them","written":"2026-11-05","verdict":"pass","bucket":"confirmed","measured":"2026-11-05","basis":"count of claims held vs missed (no single headline)","outcomeRef":"docs/research/12-predictions.md:794","underHeading":null,"note":"3 of 3 right","also":null},{"id":"P35","title":"aim at stereochemistry, not at another drawing style","written":"2026-11-05","verdict":"mixed","bucket":"mixed","measured":"2026-11-07","basis":"count of claims held vs missed (no single headline)","outcomeRef":"docs/research/12-predictions.md:937","underHeading":null,"note":"2 wrong, 2 right","also":null},{"id":"P36","title":"real patent drawings, with the benchmark held out","written":"2026-11-08","verdict":"mixed","bucket":"mixed","measured":"2026-11-09","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:1003","underHeading":null,"note":"two of four","also":["docs/research/12-predictions.md:1004"]},{"id":"P37","title":"more real drawings, longer (cernere-real-2)","written":"2026-11-09","verdict":"mixed","bucket":"mixed","measured":"2026-11-09","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:1043","underHeading":null,"note":"three of four","also":["docs/research/12-predictions.md:1044"]},{"id":"P38","title":"making forward faster without retraining","written":"2026-11-09","verdict":"mixed","bucket":"mixed","measured":"2026-11-09","basis":"count of claims held vs missed (no single headline)","outcomeRef":"docs/research/12-predictions.md:1136","underHeading":"P39: Resonantia I, 13C: the bar and the test protocol (written 2026-11-09 night, before an","note":"3 held, 2 missed","also":null},{"id":"P39","title":"Resonantia I, 13C: the bar and the test protocol","written":"2026-11-09","verdict":"mixed","bucket":"mixed","measured":"2026-11-10","basis":"count of claims held vs missed (no single headline)","outcomeRef":"docs/research/12-predictions.md:1202","underHeading":"P40: Cernere II on the 1,933 held-out USPTO-30K images (written 2026-11-10, before any cod","note":"ship rule met; claims: held, held, partly, one not measured","also":["docs/research/12-predictions.md:1108"]},{"id":"P40","title":"Cernere II on the 1,933 held-out USPTO-30K images","written":"2026-11-10","verdict":"pending","bucket":"pending","measured":null,"basis":null,"outcomeRef":null,"underHeading":null,"note":"cernere2-1 still training (2026-11-12); addendum at 12-predictions.md:1658","also":null},{"id":"P41","title":"Resonantia, wider and longer (resonantia-2)","written":"2026-11-10","verdict":"pending","bucket":"pending","measured":null,"basis":null,"outcomeRef":null,"underHeading":null,"note":"resonantia-2 paused at 205k of 300k steps","also":null},{"id":"P42","title":"NMR to structure, by ranking known candidates","written":"2026-11-10","verdict":"unlinked","bucket":"unlinked","measured":null,"basis":null,"outcomeRef":null,"underHeading":null,"note":"no outcome recorded","also":null},{"id":"P43","title":"Resonantia-H, 1H shifts","written":"2026-11-10","verdict":"mixed","bucket":"mixed","measured":"2026-11-11","basis":"count of claims held vs missed (no single headline)","outcomeRef":"docs/research/12-predictions.md:1299","underHeading":null,"note":"1 wrong, 2 right; not served yet","also":null},{"id":"P44","title":"Proprietas I, pKa","written":"2026-11-10","verdict":"fail","bucket":"missed","measured":"2026-11-10","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:1418","underHeading":"P45: Productum textbook fine-tune (written 2026-11-10 17:40, before the generator exists o","note":"not served","also":null},{"id":"P45","title":"Productum textbook fine-tune","written":"2026-11-10","verdict":"pass","bucket":"confirmed","measured":"2026-11-12","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:1908","underHeading":"P55: mech-10, proton transfers back (written 2026-11-12 04:05, before its data or training","note":"served as Productum 1.1","also":null},{"id":"P46","title":"Motus Build-mechanism fix, guided search plus corrected SN2/E2 data","written":"2026-11-10","verdict":"fail","bucket":"missed","measured":"2026-11-10","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:1492","underHeading":"P47: Motus batch (a), carbonyl chemistry, run mech-5 (written 2026-11-10 21:00, before the","note":null,"also":["docs/research/12-predictions.md:1559"]},{"id":"P47","title":"Motus batch (a), carbonyl chemistry, run mech-5","written":"2026-11-10","verdict":"fail","bucket":"missed","measured":"2026-11-11","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:1563","underHeading":"P49: Motus batch e, radical halogenation with fishhook arrows, run mech-7 (written 2026-11","note":null,"also":null},{"id":"P48","title":"Motus batches b, c, d (EAS, alkene additions, pericyclic), run mech-6","written":"2026-11-10","verdict":"fail","bucket":"missed","measured":"2026-11-11","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:1567","underHeading":"P49: Motus batch e, radical halogenation with fishhook arrows, run mech-7 (written 2026-11","note":null,"also":null},{"id":"P49","title":"Motus batch e, radical halogenation with fishhook arrows, run mech-7","written":"2026-11-10","verdict":"fail","bucket":"missed","measured":"2026-11-11","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:1610","underHeading":"P50: Motus mech-8, corrected EAS plus radicals and hydroboration in one run (written 2026-","note":null,"also":null},{"id":"P50","title":"Motus mech-8, corrected EAS plus radicals and hydroboration in one run","written":"2026-11-11","verdict":"fail","bucket":"missed","measured":"2026-11-11","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:1622","underHeading":null,"note":null,"also":null},{"id":"P51","title":"Motus mech-9, every EAS writing kept with its arrow re-derived","written":"2026-11-11","verdict":"fail","bucket":"missed","measured":"2026-11-12","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:1792","underHeading":"P53: stereo-aware Resonantia, 13C and 1H (written 2026-11-11 22:05, before any code or dat","note":"P54 later showed the batch-b miss was an input bug (mech-9 passes on corrected inputs)","also":["docs/research/12-predictions.md:1836"]},{"id":"P52","title":"Resonantia (13C) and Resonantia-H (1H) against ChemDraw's ChemNMR","written":"2026-11-11","verdict":"unlinked","bucket":"unlinked","measured":null,"basis":null,"outcomeRef":null,"underHeading":null,"note":"no outcome recorded","also":null},{"id":"P53","title":"stereo-aware Resonantia, 13C and 1H","written":"2026-11-11","verdict":"pending","bucket":"pending","measured":null,"basis":null,"outcomeRef":null,"underHeading":null,"note":"stereo data still being prepared (2026-11-12)","also":null},{"id":"P54","title":"tb200 batch b with Lewis-acid catalysts numbered","written":"2026-11-12","verdict":"pass","bucket":"confirmed","measured":"2026-11-12","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:1836","underHeading":null,"note":"mech-9 passes the bars but was not served: it failed the service's proton-transfer test (line 1869)","also":["docs/research/12-predictions.md:1869"]},{"id":"P55","title":"mech-10, proton transfers back","written":"2026-11-12","verdict":"fail","bucket":"missed","measured":"2026-11-12","basis":"the record's own headline","outcomeRef":"docs/research/12-predictions.md:1955","underHeading":"P56: productum-tb-2, hydride reductions (written 2026-11-12 08:00, before training)","note":null,"also":null},{"id":"P56","title":"productum-tb-2, hydride reductions","written":"2026-11-12","verdict":"pending","bucket":"pending","measured":null,"basis":null,"outcomeRef":null,"underHeading":null,"note":"productum-tb-2 not trained yet","also":null},{"id":"P57","title":"mech-11, proton transfers at a gentler learning rate","written":"2026-11-12","verdict":"pending","bucket":"pending","measured":null,"basis":null,"outcomeRef":null,"underHeading":null,"note":"mech-11 not trained yet","also":null}],"updates":[]}