feat(LTX25-RESIDENCY-W0-LIVE): a render now says what it is doing while it does it (#1413) - #1443

Merged

localai-bot merged 31 commits into

mainfrom

row/LTX25-RESIDENCY-W0-LIVE

Aug 20, 2026

Commits on Aug 19, 2026

test(LTX25-RESIDENCY-W0): the red — a render through the ABI writes no phase table at all (#1010)
mudler
committed
d12db0d
feat(LTX25-RESIDENCY-W0): a render now says where it spent its wall, and what it held while it did (#1010)
mudler
committed
e6f58c2
fix(LTX25-RESIDENCY-W0): a LOAD starts the timeline, so a second engine does not inherit the first one's residue (#1010)
mudler
committed
040a4ee
docs(LTX25-RESIDENCY-W0): the phase table, its two lanes and what the device column does not yet say (#1010)
mudler
committed
ae9a7f2
merge: origin/main into row/LTX25-RESIDENCY-W0 — one unrelated spec-decode wave
mudler
committed
43693a4
merge: origin/main into row/LTX25-RESIDENCY-W0 — a spec-only lease-clock row
mudler
committed
4465955
fix(LTX25-RESIDENCY-W0): the device column answers -1 on CUDA too, and #1126 is why (#1010)
mudler
committed
4fe6b47
record(LTX25-RESIDENCY-W0): the artifact, re-taken on the merged tree, and the ranking that flipped between two identical runs (#1010)
mudler
committed
d32530b
fix(LTX25-RESIDENCY-W0): hand the sampler thread out under the lock instead of probing it outside one (#1010)
mudler
committed
4f3c243
spec(LTX25-RESIDENCY-W0-LIVE): the phase table reports on the runs that finish, and this campaign has none (#1413)
mudler
committed
2ee6183
test(LTX25-RESIDENCY-W0-LIVE): the red — a render through the ABI prints nothing at all while it runs (#1413)
mudler
committed
1057dc7
feat(LTX25-RESIDENCY-W0-LIVE): a render now says what it is doing while it does it (#1413)
mudler
committed
bca9b29

Navigation Menu

feat(LTX25-RESIDENCY-W0-LIVE): a render now says what it is doing while it does it (#1413) - #1443

feat(LTX25-RESIDENCY-W0-LIVE): a render now says what it is doing while it does it (#1413)#1443
localai-bot merged 31 commits into
mainfrom
row/LTX25-RESIDENCY-W0-LIVE

Commits on Aug 19, 2026

test(LTX25-RESIDENCY-W0): the red — a render through the ABI writes no phase table at all (#1010)

feat(LTX25-RESIDENCY-W0): a render now says where it spent its wall, and what it held while it did (#1010)

fix(LTX25-RESIDENCY-W0): a LOAD starts the timeline, so a second engine does not inherit the first one's residue (#1010)

docs(LTX25-RESIDENCY-W0): the phase table, its two lanes and what the device column does not yet say (#1010)

merge: origin/main into row/LTX25-RESIDENCY-W0 — one unrelated spec-decode wave

merge: origin/main into row/LTX25-RESIDENCY-W0 — a spec-only lease-clock row

fix(LTX25-RESIDENCY-W0): the device column answers -1 on CUDA too, and #1126 is why (#1010)

record(LTX25-RESIDENCY-W0): the artifact, re-taken on the merged tree, and the ranking that flipped between two identical runs (#1010)

fix(LTX25-RESIDENCY-W0): hand the sampler thread out under the lock instead of probing it outside one (#1010)

spec(LTX25-RESIDENCY-W0-LIVE): the phase table reports on the runs that finish, and this campaign has none (#1413)

test(LTX25-RESIDENCY-W0-LIVE): the red — a render through the ABI prints nothing at all while it runs (#1413)

feat(LTX25-RESIDENCY-W0-LIVE): a render now says what it is doing while it does it (#1413)

Commits on Aug 20, 2026

wip(LTX25-RESIDENCY-W0): checkpoint the in-progress W0 gate repair before disk exhaustion

merge: origin/main into row/LTX25-RESIDENCY-W0 for the W0 gate repair

test(LTX25-RESIDENCY-W0): the seconds differential was a coin flip, so tie the name to the work STRUCTURALLY (#1010)

record(LTX25-RESIDENCY-W0): the artifact carries its own provenance now, and the review's findings are written down (#1010)

record(LTX25-RESIDENCY-W0): say plainly that the two sampler-lifetime repairs carry no mutation (#1010)

fix(LTX25-RESIDENCY-W0): the console lane was a passenger of the file write, and the success-path limit was written nowhere (#1010)

merge: origin/main into row/LTX25-RESIDENCY-W0 after the fresh-review repairs

fix(LTX25-RESIDENCY-W0): containment covered ONE of the three carrying phases, and the largest could hand 82% of itself to a neighbour undetected (#1010)

merge: the concurrent containment repair into this branch's review repairs

record(LTX25-RESIDENCY-W0): the artifact predates the anchors it is now evidence for, and F10 was called unobservable by a spec that was wrong (#1010)

merge(LTX25-RESIDENCY-W0): take origin/main 63d87805c under the W0 anchors

merge: the W0 record repairs that landed while these review repairs built

merge: the concurrent console-lane repair, and its merge of these anchors

fix(LTX25-RESIDENCY-W0-LIVE): the tick placement was argued four times and checked nowhere, and the cost claim was false in the small (#1413)

merge: the repaired base into W0-live, and the trace in USAGE is now an OBSERVED one

merge: the base branch's own merge topology, with no tree change

merge: take origin/main under W0-live, and keep the tick above the forward #1092 rewrote

record(LTX25-RESIDENCY-W0-LIVE): W0's own SUM gate is a ratio whose floor sits inside its own measurement, and it is red on main (#1439)

merge: take origin/main aa0996d0d, which moved while this lane was gating