Repository navigation
v0.4.0
What's Changed
- Compute the batch-time split instead of reading it off a round by hand by @dannbleeker in #457
- Six more layouts drew a band nobody paid for by @dannbleeker in #458
- Give the horizontal legend's reservation and its draw one predicate by @dannbleeker in #459
- Make dashKind total so all three sinks agree a line is dashed by @dannbleeker in #460
- Fit a sunburst's outside labels to their neighbour, not just their own arc by @dannbleeker in #461
- Let a combo choose which value labels survive a frame too small for both by @dannbleeker in #462
- Give the task pane a way to be looked at by @dannbleeker in #463
- Ask the shared legend predicate from the mekko too, not just two of three sites by @dannbleeker in #464
- Close the upright residue: bound the totals, the line's name and the CAGR caption by @dannbleeker in #465
- Correct the showcase determinism claim: slide XML is stable, the bytes are not by @dannbleeker in #466
- Three more slots were burning the scratch slide, and the position theory was the correlate by @dannbleeker in #468
- Record that the burnt-slot prediction failed, and where the blocker actually moved to by @dannbleeker in #469
- Ask the tag question where this host will answer it, and record the other two as having no route by @dannbleeker in #470
- Check the upstream issues against GitHub instead of against our own notes by @dannbleeker in #471
- Make the tag question say which slide it wrote to and which it read from by @dannbleeker in #472
- Count every draw in the round, not just the four the scenario makes by @dannbleeker in #473
- Keep the rounds, write the protocol down, and check the upstream issues on a command by @dannbleeker in #474
- Judge the predictions and show what each scenario has said across rounds by @dannbleeker in #475
- Drive a round, and refuse to start one that cannot prove anything by @dannbleeker in #476
- Say when an answer rests on one attempt out of several by @dannbleeker in #477
- Measure the scratch-slide leak, and record why the obvious fix is not one by @dannbleeker in #478
- Record that the pass-boundary sweep is blocked by the accounting, not the timing by @dannbleeker in #479
- Give the clean-up a ledger, so it can count slides returned before it ran by @dannbleeker in #480
- Correct the record: the accounting was not the blocker, the host is by @dannbleeker in #481
- Ask the primitive directly: one known slide CAN be removed mid-run by @dannbleeker in #482
- Instrument the pass boundary: the deletes are innocent, the adds are not by @dannbleeker in #483
- A scratch slide that lands and cannot be named is still in the deck by @dannbleeker in #484
- A playwright ref is not a build stamp by @dannbleeker in #485
- Ask the host if it is awake before spending an hour finding out by @dannbleeker in #486
- The wedge is the host's session dying, and a reload clears it by @dannbleeker in #487
- Round 29: the scratch slides never landed, and the deck never grew by @dannbleeker in #488
- The calls in flight when PowerPoint crashes are reads, not writes by @dannbleeker in #489
- Watch for the crash mid-round instead of waiting out the thirty minutes by @dannbleeker in #490
- Let the driver clear the crash and start again by @dannbleeker in #491
- Ask whether the id round trip is what loses the config tag by @dannbleeker in #492
- The host answered: a shape is reachable only in the batch that created it by @dannbleeker in #493
- The write goes through the creation handle; the read does not by @dannbleeker in #494
- The settle pass has nothing to try, and the round says so in one field by @dannbleeker in #495
- The retry belongs in the drawing context, not the settle pass by @dannbleeker in #496
- Round 036: a fourth sample, and the night's last by @dannbleeker in #497
- The rule is resolution, not age — and the one-line fix does not work by @dannbleeker in #498
- Say which handle the failed tag write went through by @dannbleeker in #499
- Three unknowns are not agreement by @dannbleeker in #500
- Ask how long a creation handle lasts, because round 037 says that is the fix by @dannbleeker in #501
- Round 038: the from field says created, twice by @dannbleeker in #502
- Never file the same round twice by @dannbleeker in #503
- Keep the crash evidence before the tab that holds it is reloaded by @dannbleeker in #504
- Scope raising the rasterise arm count, rather than doing it late by @dannbleeker in #505
- A signed-out browser is not a closed pane by @dannbleeker in #506
- A browser setup that survives the session that made it by @dannbleeker in #507
- Move the round-loop brief and journal into the repo by @dannbleeker in #508
- Round 039: the creation handle's lifetime, and it is not the constraint by @dannbleeker in #509
- A crash report that was empty because the report could not be read by @dannbleeker in #510
- Round 041: the first clean start in five, and one point does not make an effect by @dannbleeker in #511
- Why the ordering fix is a redesign, from having tried it by @dannbleeker in #512
- Ask the host what refuses the tag write, because the fake and Office.js disagree by @dannbleeker in #513
- Round 042 answers the ordering question, and the driver stops giving up on itself by @dannbleeker in #514
- The chart's tag goes on a handle nothing has resolved by @dannbleeker in #515
- Round 043: the tag-anchor fix is unvalidated, and one round cannot validate it by @dannbleeker in #516
- Make the tag question decidable, and ask whether the context was ever the lever by @dannbleeker in #517
- Round 044: the group question missed, and the miss says grouping is about age by @dannbleeker in #518
- Prepare the loop for a long unattended run by @dannbleeker in #519
- Grouping is what saves a config, not the tag handle by @dannbleeker in #520
- The re-read decays through a deck update, and the slide id was never the story by @dannbleeker in #521
- Move the decay measurement out of the probe and into production by @dannbleeker in #522
- A chart on a freshly added slide cannot be grouped, and that is why it loses its config by @dannbleeker in #523
- The rasterise question closes: it does not poison the next draw by @dannbleeker in #524
- Pair 6: the 2s crash has not recurred since the pre-flight shipped by @dannbleeker in #525
- Pair 7, and the documents were where the work was by @dannbleeker in #526
- Pair 8, and the doc drift this run created by @dannbleeker in #527
- CLAUDE.md was the last document pointing at the wrong problem by @dannbleeker in #528
- Guards for two of tonight's fixes, and the run stops on the sign-in by @dannbleeker in #529
- Stage 0: the ownership guarantee is not needed, and the queue is written down by @dannbleeker in #530
- Say what refreshShapes actually means by @dannbleeker in #531
- Put the tag anchor back on created[0] by @dannbleeker in #532
- Ask a settling slide twice before believing it by @dannbleeker in #533
- Make a round say WHY it lost the charts by @dannbleeker in #534
- Round 064: the retry works, and grouping does not save a config by @dannbleeker in #535
- Round 065: the pair replicates, chart for chart by @dannbleeker in #536
- Stop the pooled table asserting what the pair disproved by @dannbleeker in #537
- Name the questions that have never once answered by @dannbleeker in #538
- A single-batch chart cannot group here, and that is our bug by @dannbleeker in #539
- Re-read before grouping for every chart that will be grouped by @dannbleeker in #540
- Rounds 066+067: the prediction failed, and so had the instrument by @dannbleeker in #541
- Stop the batch-span note describing a gate that no longer exists by @dannbleeker in #542
- docs: correct the round loop brief's stale round number and pooled figures by @dannbleeker in #543
- docs: plan a fix before writing it, and raise observability on its own merit by @dannbleeker in #545
- Ask again when the re-read names none of the chart's shapes by @dannbleeker in #544
- Rounds 068+069: the pair did not replicate, and one field says why by @dannbleeker in #546
- Bind the tag target, so the settle has a handle that is not an id by @dannbleeker in #547
- Notice a browser that dies UNDER a running round by @dannbleeker in #548
- Rounds 070-072: same scale passes, for a reason it was not built for by @dannbleeker in #549
- Hold a build to its rounds, and check the drag round trip without a drag by @dannbleeker in #550
- Sweep a dirty deck rather than refusing the round over it by @dannbleeker in #551
- Let the driver finish the round it just ran by @dannbleeker in #552
- An absence of evidence is not a regression by @dannbleeker in #553
- Count the charts that cannot follow a drag by @dannbleeker in #554
- Write the origin tag through the binding by @dannbleeker in #555
- Rounds 075+076: the origin tag is fixed, and the binding is proven by @dannbleeker in #556
- Round 077: the first 4:3 round, and it is confounded by @dannbleeker in #557
- Make 4:3 safe to run nightly, and clear it of round 077 by @dannbleeker in #558
- The first full cycle: 4:3 is fine, and the divergence check was not by @dannbleeker in #559
- Cut the cost of mining a round, and run the night without a person between the rounds by @dannbleeker in #560
- An update no longer gives up on a chart whose id the host refused by @dannbleeker in #561
- Stop the round driver hanging, and stop it dying on a surprise by @dannbleeker in #562
- Measure whether an update strands the rest of an ungrouped chart by @dannbleeker in #563
- Two bugs in how rounds are filed and judged by @dannbleeker in #564
- Six defects in run handling, two of them introduced this afternoon by @dannbleeker in #565
- Rounds 082-083, and the orphan instrument rebuilt after its first reading by @dannbleeker in #566
- Stop retrying a pane there is no add-in to open by @dannbleeker in #567
- Correct two docs that describe a reading the instrument no longer takes by @dannbleeker in #568
- Round 084, and a growth reading the host had not settled on by @dannbleeker in #569
- Put the add-in back after a browser death, and round 085 by @dannbleeker in #570
- Count the slides a single add lands, not the add by @dannbleeker in #571
- Round 086, and the stranding question finally put by @dannbleeker in #572
- Four things round 086 taught, three of them about the reporting by @dannbleeker in #573
- Round 087: the settle held, and the stranding count reaches three by @dannbleeker in #574
- Write down what rounds 082-087 taught, and correct what went stale by @dannbleeker in #575
- Bug hunt: ten charts that drew something other than their data by @dannbleeker in #576
- Bug hunt, round two: nine more charts that drew something other than their data by @dannbleeker in #577
- Take the decoration sweep to zero, and stop three more kinds drawing off the canvas by @dannbleeker in #578
- Record what the frame gate does not sweep, and the 42 it hides by @dannbleeker in #579
- Write alt text into the skill's generated .pptx by @dannbleeker in #580
- Sweep the whole frame x font grid, and close the 42 overflows it hides by @dannbleeker in #581
- Sweep text against text at the fonts either side of the default by @dannbleeker in #582
- Store the style in the DECK, so a shared deck keeps its branding by @dannbleeker in #583
- Take the overlap gate to every font, and stop the axis climbing into the title by @dannbleeker in #584
- Document what the widened gates guarantee, for users rather than for CI by @dannbleeker in #585
- Group the part of a chart the host will name, rather than none of it by @dannbleeker in #586
- Sweep the decorations, and fix the five ways they left the chart by @dannbleeker in #587
- Sweep the options and the data, and fix the nine ways they left the chart by @dannbleeker in #588
- Stake #586 in the ledger, and stop the judge answering from an older round by @dannbleeker in #589
- Let a dated prediction be judged, and write the round prompt down by @dannbleeker in #590
- Round 088, and the prediction two bugs were hiding between them by @dannbleeker in #591
- Round 089: a failed read is not an absence, and the pair takes back a reading by @dannbleeker in #592
- Close the stranding blind spot, and stop the fake refusing what Office returns by @dannbleeker in #593
- A greyed-out ribbon is not a missing add-in by @dannbleeker in #594
- Round 090: the deck-style read never answers, and it is 2 for 2 by @dannbleeker in #595
- The number nobody could read was the answer by @dannbleeker in #596
- Round 091: the feature nobody could click, and a fix that hid its own subject by @dannbleeker in #597
- The first real pair, and it says the verdict cannot see grouping by @dannbleeker in #598
- Three green lights wired to nothing, and an instrument that never reached a round by @dannbleeker in #599
- The call that hangs is the one asking an EMPTY namespace for its only item by @dannbleeker in #600
- A first-ever failure the pair dissolved, and a probe that asserted its answer by @dannbleeker in #601
- Spend the bad first call — six rounds say the second one works by @dannbleeker in #602
- The rounds take back my retry by @dannbleeker in #603
- Spend the bad call BEFORE the read, not inside it by @dannbleeker in #604
- Silence is not an answer — make the read say when it worked by @dannbleeker in #605
- The deck-style read is cured, and the round says so by @dannbleeker in #606
- The visibility gate's blindness has exactly one cause, and it is not instability by @dannbleeker in #607
- Time a successful rasterise, and write down what actually blocks 4:3 by @dannbleeker in #608
- Let a leg set its own slide size, when the owner asks for it by @dannbleeker in #609
- Two guards for traps that bit twice, and two meters nobody read by @dannbleeker in #610
- A stall that names its own call site by @dannbleeker in #611
- Keep the read the deck backs, and trace the one nobody recorded by @dannbleeker in #612
- The cold read never stopped failing — it fails eleven times a round by @dannbleeker in #613
- The cold read replicates, and the retry never fails at the SHORT one by @dannbleeker in #614
- Answer the question BACKLOG left open: the retry hides it by @dannbleeker in #615
- Sweep the sibling: no baseline is not a baseline of zero by @dannbleeker in #616
- The 4:3 deck was never 4:3, and the check could not have told anyone by @dannbleeker in #617
- The first 4:3 pair, and a fix of mine that broke a diagnosis by @dannbleeker in #618
- The deck is innocent, and the archive lied about the profile by @dannbleeker in #619
- Correct two claims of mine, and make the archive able to catch the third by @dannbleeker in #620
- Let the round browser's page follow its window by @dannbleeker in #621
find's miss message contains the query, and I believed it twice by @dannbleeker in #622- The second round of a pair is worse, and always has been by @dannbleeker in #623
- One build three times: the observer is the variable by @dannbleeker in #624
- The idle pair refutes "the observer is the variable" by @dannbleeker in #625
- It was the pane's age all along, and the metric that hid it was mine by @dannbleeker in #626
- A clean deck is not a clean pane — reload it between rounds by @dannbleeker in #627
- A reload raises a modal, and a modal blocks everything by @dannbleeker in #628
- The fix works: a second round that scored a first round's numbers by @dannbleeker in #629
- The 4:3 profile effect was never real by @dannbleeker in #630
- The visibility gate's blindness has one cause, and it is 40 for 40 by @dannbleeker in #631
- Pull the between-rounds reload: it costs the sideload about 1 round in 4 by @dannbleeker in #632
- sideloadAddIn: reload before walking, or a stuck menu defeats every attempt by @dannbleeker in #633
- Wait for the document, not the clock, before the sideload walk by @dannbleeker in #634
- The add-in was never missing: a narrow window hid it, four times by @dannbleeker in #635
- The blind gate is fixed, and pane age is dose-dependent by @dannbleeker in #636
- The selection scenario never failed, and I said it had by @dannbleeker in #637
- Pool the fallbacks nothing was reading, and see drift a median hides by @dannbleeker in #638
- The redraw drift was never a drift, and the feature has never run by @dannbleeker in #639
- The parts list: the archive cannot answer, so ship the instrument by @dannbleeker in #640
- The instrument answered on its first run: it is grouping by @dannbleeker in #641
- Teach the in-place update to read a group's members by @dannbleeker in #642
- The anchor is the group's first member, not the tagged shape by @dannbleeker in #643
- The mapping is solved; the wall moved by @dannbleeker in #644
- Stamp the scene fingerprint into the generated deck by @dannbleeker in #645
- Write node 0 to the group's anchor, not to the group by @dannbleeker in #646
- Record what the group-member ordering actually rests on by @dannbleeker in #647
- Round 146: the in-place update ran, 3 of 13 by @dannbleeker in #648
- The gate counted two of the in-place update's three outcomes by @dannbleeker in #649
- Read the error, not the payload it was handed by @dannbleeker in #650
- A successful recovery erases its own evidence by @dannbleeker in #651
- Retire two questions production answered before the probe could by @dannbleeker in #652
- The share limit was never measured; the archive measures it by @dannbleeker in #653
- Revert the share limit to 0.5 — 0.8 crashes PowerPoint by @dannbleeker in #655
- A crashed round left nothing, and it did not have to by @dannbleeker in #656
- The share limit at 0.6 — an eighth of the dose that killed the host by @dannbleeker in #657
- Round 151: 0.6 holds, and the ladder has no more rungs by @dannbleeker in #658
- Chunk the in-place writes — bound the cost the host actually measures by @dannbleeker in #659
- Round 152: chunking costs 108ms on a 164-second scenario by @dannbleeker in #660
- Raise the limit to 0.8 again, with the batch bounded this time by @dannbleeker in #661
- Round 153: the crash was per-SYNC size, and the feature is at its ceiling by @dannbleeker in #662
- Name what the recovery recovered from, and pool it by @dannbleeker in #663
- Report what a round records and no tool reads by @dannbleeker in #664
- Round 154 confirms 153, and the recovery label works on the host by @dannbleeker in #665
- Start a round from a fresh session, on request by @dannbleeker in #666
- Round 155: the first round in the archive to start first time by @dannbleeker in #667
- Drop the automation flag Chrome prints a warning about by @dannbleeker in #668
- Revert the automation-flag change — the add-in would not sideload by @dannbleeker in #669
- recover() opens the deck, not just the browser by @dannbleeker in #670
- Front a deck tab that is already open — my own regression by @dannbleeker in #671
- The ribbon threshold sat below a window that was genuinely cramped by @dannbleeker in #672
- Close a sign-in popup that outlived its flow by @dannbleeker in #673
- A read that could not RUN is not an absence by @dannbleeker in #674
- Prune the profile caches in the round loop, keeping the sign-in by @dannbleeker in #675
- Ask whether the driver can USE a browser, not whether one exists by @dannbleeker in #676
- Judge a prediction on every round that asked, not on the newest one by @dannbleeker in #677
- Stop two reports demanding work that was already done by @dannbleeker in #678
- Claim the aged-handle resolve, and correct what rests on it by @dannbleeker in #679
- Print the grouping rate that is current, not only the one that cannot be read by @dannbleeker in #680
- A median cannot see a step, so print the sequence by @dannbleeker in #681
- The pair inherits an aged pane, and the denominator that halved by @dannbleeker in #682
- Name what the decline SAW, not only what it concluded by @dannbleeker in #683
- Stake the fresh-pane claim, and warn when a stake back-dates itself by @dannbleeker in #684
- One denominator for the scratch handback, and record which arm a round ran in by @dannbleeker in #685
- Split the arms on the pane, not on the flag — and the arm was never missing by @dannbleeker in #686
- Pool the id churn, and correct a host claim written before the archive existed by @dannbleeker in #687
- Read the deck inventory the gate has printed all along by @dannbleeker in #688
- Read the parts-list counters, and retire two claims the fix outlived by @dannbleeker in #689
- Record what the slide-size ladder actually costs, without acting on it by @dannbleeker in #690
- Anchor two guards that pass on nothing by @dannbleeker in #691
- Make npm run cycle obey the pair rule, and correct six stale doc claims by @dannbleeker in #692
- Count the group that THREW — it left the numerator and the denominator together by @dannbleeker in #693
- Correct my own over-claim: the throw/loose-shapes correspondence is one-way by @dannbleeker in #694
- Record what the positional guess picked, and correct the throw's population by @dannbleeker in #695
- Refuse a positional guess that names no shape of ours by @dannbleeker in #696
- Widen to the width the ribbon actually needs by @dannbleeker in #698
- Journal the guess that answered, and the test the fix still needs by @dannbleeker in #697
- Record the sign-in wall that stopped round 181, and how to recognise it by @dannbleeker in #699
- Suppress the flag banner with --test-type, and keep the flag by @dannbleeker in #700
- Ask the host what creationId actually is, before building on it by @dannbleeker in #701
- Correct the record: shape.id is not volatile, and keep the parts list by @dannbleeker in #702
- Record which tag route WORKED, so the failure counts get a denominator by @dannbleeker in #703
- Verify the product sweep by hand — five of eight claims did not survive by @dannbleeker in #704
- Say plainly that the parts-list fix is inert, and why the next step is a trap by @dannbleeker in #705
- Ask whether the collection can be walked by index where it refuses to be listed by @dannbleeker in #706
- Record two closed leads, and stop on a real sign-in prompt by @dannbleeker in #707
- Tell a sign-in ERROR from a sign-in PROMPT, and recover from the first by @dannbleeker in #708
- Ask the settling slide a second time, and correct the argument against it by @dannbleeker in #709
- Say which settle ask a pass was, so the second one is not invisible by @dannbleeker in #710
- Prove the second settle ask rescues a chart the first cannot by @dannbleeker in #711
Full Changelog: v0.3.2...v0.4.0