Repository navigation
v0.3.1
What's Changed
- fix: four things four real PowerPoint-web runs exposed by @dannbleeker in #233
- feat(demo): record what a run believes it tagged, and what the readback saw by @dannbleeker in #234
- feat(tools): audit a real deck from its bytes, not from the add-in's report by @dannbleeker in #235
- fix: stop the repair guessing at what the host would not show it by @dannbleeker in #236
- refactor: remove the smoke-test option by @dannbleeker in #237
- fix: tag the shape the host still recognises, and trust the write over the read by @dannbleeker in #238
- feat: join a run log to the deck it produced, in one command by @dannbleeker in #239
- feat: run both insert paths on one click by @dannbleeker in #240
- test: share the fake host, and stop it healing stale proxies by @dannbleeker in #241
- feat: a host self-test for the five paths nothing has ever exercised by @dannbleeker in #242
- fix: keep an ordinary insert re-editable on a host that refuses stale proxies by @dannbleeker in #243
- fix: four defects in the tooling shipped over the last four PRs by @dannbleeker in #244
- fix: guard three more config-keyed tables against Object.prototype by @dannbleeker in #245
- fix: a typed formula could take the pane's preview down by @dannbleeker in #246
- fix: stop the repair pass deleting another run's slide by @dannbleeker in #247
- fix: two unbounded loops a datasheet cell could start by @dannbleeker in #248
- fix: no chart can put NaN into a .pptx by @dannbleeker in #249
- docs+feat: re-check the Office.js approach, and self-test the everyday path it exposed by @dannbleeker in #250
- docs: make the Phase 2 test run standing, and order it by risk by @dannbleeker in #251
- fix: eleven defects from the real-PowerPoint sessions, and read the slide size by @dannbleeker in #252
- feat: place a second chart beside the first, not only below it by @dannbleeker in #253
- docs: default every session in this repo to caveman-style replies by @dannbleeker in #254
- docs: bring the standing test run up to what today's fixes changed by @dannbleeker in #255
- fix: never read an Office.js proxy the host did not answer for by @dannbleeker in #256
- fix: sweep every unguarded proxy read, and record which phase an error escaped by @dannbleeker in #257
- docs: correct the selftest's claim that a shape cannot be selected by @dannbleeker in #258
- feat: three self-test scenarios the selection API unlocked, and a shorter manual run by @dannbleeker in #259
- fix: a slide delete must be verified, not assumed by @dannbleeker in #260
- fix: a run can be stopped after it stops drawing, and cannot wait forever by @dannbleeker in #261
- feat: show every step live in the pane, so a crash leaves evidence by @dannbleeker in #262
- fix: newest step at the TOP, so a crash leaves it where it can be read by @dannbleeker in #263
- gate the wedged selection scenario as a host limitation, and record it by @dannbleeker in #264
- keep the log of a run that never ends by @dannbleeker in #265
- ask the host which selection call wedges it, instead of guessing by @dannbleeker in #266
- a hung sync says nothing about whether the work happened by @dannbleeker in #267
- test the selection path with a real click, and say what a slide swap costs by @dannbleeker in #268
- a bug hunt: twelve crashes and losses, and the sweeps that found them by @dannbleeker in #269
- ask the host what it actually does, so the fake can be checked against it by @dannbleeker in #270
- Believe the deck, not the proxy — and stop the selection watcher piling on by @dannbleeker in #271
- Two things the real host taught us: a read that did not happen is not a fact, and silence needs a deadline by @dannbleeker in #272
- Stop claiming success over work the host did not do by @dannbleeker in #273
- A slide handle the host will only name once, and eight questions nobody asked by @dannbleeker in #274
- A chart the drawing context could not tag is not finished by @dannbleeker in #275
- Only an id crosses a sync, never a handle by @dannbleeker in #276
- Say why the rasteriser came back with nothing by @dannbleeker in #277
- A scenario chooses its own words; it does not choose whether a deadline fired by @dannbleeker in #278
- One click, one file, and a fake that cannot drift from the host in silence by @dannbleeker in #279
- Editing a chart on a slide this session added threw the whole update away by @dannbleeker in #280
- Two fixes from a real run, and honest labels on what guards them by @dannbleeker in #281
- Window getItem behind a fault, ask what it does, and assert the thing nobody was asserting by @dannbleeker in #282
- Phase 3: the five questions this host would not let us ask by @dannbleeker in #283
- Phase 4: ask what makes a long run slow down, instead of guessing by @dannbleeker in #284
- The round collects the deck and the screenshots itself by @dannbleeker in #285
- The ladder joins the routine battery, and a button puts the deck back by @dannbleeker in #286
- Some questions ask their own follow-up by @dannbleeker in #287
- Build around three published PowerPoint-web bugs by @dannbleeker in #288
- Ask the host about three more tracker bugs, validate the manifests in CI — and fix what that found by @dannbleeker in #289
- Sweep the office-js tracker weekly, and gate on charts being visible by @dannbleeker in #290
- Write down the upstream-reporting candidate, and correct two backlog entries by @dannbleeker in #291
- Consolidate project memory after a session of real-host work by @dannbleeker in #292
- Three standing quality checks, and an honest note on a fourth by @dannbleeker in #293
- Skip static mutants, so the weekly mutation run can finish by @dannbleeker in #294
- Demote the golden-image gate: rejected as a hash, thin as a structure by @dannbleeker in #295
- Pin what collide.ts actually does: 53% to 63% mutation score by @dannbleeker in #296
- Put the live step list at the top of the Testing panel by @dannbleeker in #297
- Fix the no-op rescale, and make the probe say what it left behind by @dannbleeker in #298
- Stop retiring a run's record before its file has been saved by @dannbleeker in #299
- Stop cancelling Pages deployments that GitHub is still working on by @dannbleeker in #300
- Mark a chart that drew but could not be named, so the settle gets its turn by @dannbleeker in #301
- Stop grouping through handles the host has already refused by @dannbleeker in #302
- Revert the settle after adding a slide: it cost 18 of 19 probe answers by @dannbleeker in #303
- Settle a chart's config through a collection read when the host refuses its id by @dannbleeker in #304
- Retire two stale claims in the project memory, and hand off the round in flight by @dannbleeker in #305
- Give an ungrouped chart's parts list the same recovery its config tag has by @dannbleeker in #306
- Record what the extended statements answered, and keep both ends of them by @dannbleeker in #307
- Make the settle say when it came up empty, and record a root-cause fix that failed by @dannbleeker in #308
- Name the properties the grouping re-read needs, on Microsoft's own guidance by @dannbleeker in #309
- Stop the repair pass reporting numbers that answer a different question by @dannbleeker in #310
- Stop asking a dead host, record two unstable answers, and group the Testing panel by @dannbleeker in #311
- Stop reporting an assumed shape count as a measured one by @dannbleeker in #312
- Correct a claim made this morning: fast is the broken mode by @dannbleeker in #313
- Let the three group questions be asked on the host that refuses id read-backs by @dannbleeker in #314
- Ask whether a group will COUNT the children it refuses to list by @dannbleeker in #315
- Stop calling every timeout the known selection limitation by @dannbleeker in #316
- Record the two slide-proxy answers as alternating, not flipped once by @dannbleeker in #317
- Pair a late answer with the item that actually gave up on it by @dannbleeker in #318
- Stop the picture scenario passing on a round-trip with no picture in it by @dannbleeker in #319
- Make the runbook name controls that exist in the pane by @dannbleeker in #320
- Record the 2026-08-08 answer sheet, and stop the gate reading silence as an answer by @dannbleeker in #321
- Fix two things the first real round of these checks exposed by @dannbleeker in #322
- Stop the visibility scenario rasterising a slide it just added by @dannbleeker in #323
- Stop reporting a drawn chart as never drawn by @dannbleeker in #324
- The visibility scenario passed, so put it back in the routine round by @dannbleeker in #325
- Name the degradation experiment's host calls, so its crash is evidence by @dannbleeker in #326
- Two slides.add() in quick succession kill this host, and the pane says when it dies by @dannbleeker in #327
- Stop charging a slideless question for a slide, and a stalled host for a bug by @dannbleeker in #328
- Give every scenario that shares a slide its own place on it by @dannbleeker in #329
- Say which way the host got in the way, and name the scenarios by @dannbleeker in #330
- The degradation experiment measured something, and it is not the context by @dannbleeker in #331
- Record the 40b5e44 answer sheet, and a second hint about what slows a run by @dannbleeker in #332
- The scenario's score is a measurement of when the host flips, not a defect by @dannbleeker in #333
- The slowdown is per-slide, and the second sample is what says so by @dannbleeker in #334
- Place every scenario's shapes from the box they must clear, not a guessed one by @dannbleeker in #335
- Make the diagnostics say what happened, instead of leaving it to be inferred by @dannbleeker in #336
- Repair the three list-valued decorations at the boundary, like every other input by @dannbleeker in #337
- Read the hostile sweep's key list from the types, and fix what it then found by @dannbleeker in #338
- Name the call before a stall, and stop asking a question rounds cannot answer by @dannbleeker in #339
- Retract the predecessor account, and give the stall record its baseline by @dannbleeker in #340
- The idle gap before a stall is not the variable — measured, not argued by @dannbleeker in #341
- Round 8: the settle can rescue a fresh chart, and the degradation has three stages by @dannbleeker in #342
- Round 9: give the stall's predecessor a baseline, before reading anything into it by @dannbleeker in #343
- The predecessor's name survived its baseline — so build the control that isolates it by @dannbleeker in #344
- The rasterise lead is dead too — and the control that found it was rigged by @dannbleeker in #345
- Bring the pending section current, so a fresh session is not misled by it by @dannbleeker in #346
- Give CLAUDE.md a source map and the Windows gate, and drop a paragraph it says twice by @dannbleeker in #347
- Ask whether a binding can name a shape this host will not hand back by @dannbleeker in #348
- Stop measuring the clock when the question is the arithmetic by @dannbleeker in #349
- Read a round that has no deck, instead of printing usage at it by @dannbleeker in #350
- Take Python out of the local gate, so the whole suite runs on Windows by @dannbleeker in #351
- Let a round photograph every slide it adds, so an empty one can be proved by @dannbleeker in #352
- Ask again at the end of the run, instead of losing half the sheet to a bad minute by @dannbleeker in #353
- Give the binding question a control, so its refusal is an answer by @dannbleeker in #354
- Say what the round left in the deck, and name the button that clears it by @dannbleeker in #355
- Stop the second pass bailing on one refused slide, and stop it doing so in silence by @dannbleeker in #356
- Say that the sheet's yield is a sample, before someone reads it as a result by @dannbleeker in #357
- Put the round button first, and default the second witness to on by @dannbleeker in #358
- Ask the binding question sixth, where questions like it get answered by @dannbleeker in #359
- Stop handing the next question a slide the last one wrecked by @dannbleeker in #360
- Let the pane tell you it is older than the site by @dannbleeker in #362
- Make the weekly office-js sweep worth opening by @dannbleeker in #363
- Claim the slide we appended, and bound how long we wait for it by @dannbleeker in #364
- Add the counterbalanced arms up across rounds, so they can eventually answer by @dannbleeker in #365
- Record the binding question as unanswerable, and stop chasing causes by @dannbleeker in #366
- Draw a chart the visibility scenario can finish, and finish the clock fix by @dannbleeker in #367
- Three tool CLIs that exited 0 without running, and a diff that would not name its host by @dannbleeker in #368
- The flake sweep could never file a report, and nothing has ever checked the install path by @dannbleeker in #369
- Four checks that could not fire, or fired into the void by @dannbleeker in #370
- Bump to 0.3.1 for a release that actually ships the manifest fix by @dannbleeker in #371
Full Changelog: v0.3.0...v0.3.1