https://mainthread.ai/changelog/ What shipped across David Jones's products, and what his research found. 2026-10-09T00:00:00Z David Jones https://mainthread.ai/changelog/#system-sentinel-1-14-0 2026-10-09T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-13-0 2026-10-09T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-12-0 2026-10-08T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-11-1 2026-10-08T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-11-0 2026-10-08T00:00:00Z https://mainthread.ai/changelog/#multithread-0-4-18 2026-10-01T00:00:00Z https://mainthread.ai/changelog/#multithread-0-4-17 2026-09-27T00:00:00Z https://mainthread.ai/changelog/#wordship-wordships-own-home 2026-09-26T00:00:00Z

Improved: Wordship now runs on its own servers. Your games, record and friends came along; sign in once more on each device. Improved: Deleting your account removes your sign-in too, in the app or on the web, even if you never saved a game.

https://mainthread.ai/changelog/#wordship-clearer-deletion 2026-09-26T00:00:00Z

Improved: The deletion page separates your Wordship play from the sign-in shared with other Mainthread apps, and says which one each action removes. New: If you created a sign-in here, you can request its removal as well, even if you never saved a game.

https://mainthread.ai/changelog/#multithread-0-4-16 2026-09-26T00:00:00Z https://mainthread.ai/changelog/#zoning-signal-0-31-0 2026-09-26T00:00:00Z

semantic_search now ranks every record three ways at once, by the words in its text, by its title and by meaning, and fuses the three lists.

https://mainthread.ai/changelog/#system-sentinel-1-10-1 2026-09-26T00:00:00Z https://mainthread.ai/changelog/#lichtenberg-2026-09-26-a 2026-09-26T00:00:00Z

Teaching the small model to handle reversed order on half of its familiar option pairs carried over to the other half in all four runs, and in none of the four to pairs it had not seen as options. Limit: It shows a bounded training effect on one task and does not establish a general rule or an internal pointer. The miss on new pairs may reflect their difficulty.

https://mainthread.ai/changelog/#adaptive-rrf-cycle-23 2026-09-26T00:00:00Z

Across 29 submission groups and 30 topics, averaging a team's rankings before adding one fixed reference run beats adding that run to a uniformly chosen single member, raising mean top-ten precision by 0.24 to 6.35 points under every completion of missing labels. Limit: This covers the current archive files on 30 reused topics (78 of 81 differ from their listed checksums), is not a new method, and only 13 of 29 teams are positive on their own.

https://mainthread.ai/changelog/#adaptive-rrf-cycle-22 2026-09-26T00:00:00Z

The first archived run's bytes did not match the publisher's listed checksum, so the frozen 81-submission comparison stopped before any evaluation. Limit: This is a data identity failure. It gives no evidence for or against averaging submitted outputs.

https://mainthread.ai/changelog/#adaptive-rrf-cycle-21 2026-09-26T00:00:00Z

Official metadata yields 29 eligible teams with 81 automatic submissions for a grouped test, and an exact synthetic case shows the earlier control design allows effects of either sign. Limit: Metadata and synthetic work only: no new ranking was downloaded or evaluated.

https://mainthread.ai/changelog/#wordship-steadier-turns-and-your-data 2026-09-25T00:00:00Z

Improved: The Deduce board no longer jumps after you press Enter, and your letters stay put until their colors turn. Improved: A finished solo game has one row of navigation, with Next word just below Share. New: Delete your Wordship play from your profile, or from the deletion page on the web. New: Guests can delete their play too, from the device they played on. New: A privacy page explains what Wordship keeps and why.

https://mainthread.ai/changelog/#system-sentinel-1-10-0 2026-09-25T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-30 2026-09-25T00:00:00Z https://mainthread.ai/changelog/#lichtenberg-2026-09-25-b 2026-09-25T00:00:00Z

Every version learned the task in its trained order. When the two options were listed in reverse, answers given in fixed positions stayed accurate, while answers written as option labels fell short of the accuracy bar, though they partly adapted. Limit: It is a result about what the task demands, and it leaves open both the internal mechanism and whether reversal can be learned in general.

https://mainthread.ai/changelog/#lichtenberg-2026-09-25-a 2026-09-25T00:00:00Z

Asked to select one of two displayed probability rows, the unchanged small network missed the target in every run, reacting about equally to the row it was asked about and the one it was not. Limit: The result holds for this model, task and training budget, and does not show the design is incapable of it.

https://mainthread.ai/changelog/#adaptive-rrf-cycle-20 2026-09-25T00:00:00Z

The variants' natural fusion scores about 1.6 points of top-ten precision below the mean of 256 predeclared shuffled controls under every completion of missing labels, excluding the proposed advantage. Limit: The controls are built from relevance labels, so their higher mean serves as a diagnostic and cannot be used as a retrieval policy.

https://mainthread.ai/changelog/#adaptive-rrf-cycle-19 2026-09-25T00:00:00Z

None of four later TREC-COVID rounds supplies all four original retrieval roles, so the transfer stopped before any rankings or judgments were acquired. Limit: This is a feasibility result. It is neither a negative retrieval result nor proof that transfer is impossible.

https://mainthread.ai/changelog/#adaptive-rrf-cycle-18 2026-09-25T00:00:00Z

With two documented variants in place of two repeated copies, the hybrid beats the neural-reranked source alone by 1.7 to 2 points of top-ten precision, but its comparison with the existing hybrid stays between minus 1 and plus 2 points. Limit: These are development results on the same 30 reused topics. No untouched query was tested.

https://mainthread.ai/changelog/#adaptive-rrf-cycle-17 2026-09-25T00:00:00Z

On 30 TREC-COVID topics, swapping one minority-source result into a two-source hybrid's top ten cannot raise top-ten precision, because its seven relevant admissions displaced seven relevant results. Limit: The planned later-label phase failed its preservation check and the follow-up was post hoc. The result covers one fixed policy on 30 reused topics and leaves open whether minority evidence is useful.

https://mainthread.ai/changelog/#adaptive-rrf-cycle-16 2026-09-25T00:00:00Z

On SciFact, 21 known-relevant documents appear only in the neural source's list, yet on every query ten lexically supported documents outrank every result only the neural source retained, so none reaches the fused top ten. Limit: This describes existing lists after the fact and tests neither reweighting nor reserved slots.

https://mainthread.ai/changelog/#bendr-1-52-0 2026-09-24T00:00:00Z

New: Copy a backup of all logged runs, including their names and notes. Paste it on another device to preview and add missing runs without removing older ones. Improved: How to measure now explains inside and outside corners, the close-leg gap and where to find the less common End reference. Improved: The measurement guide's tabs wrap on narrow screens and when you use larger text. Improved: Shared links and in-app support links use Bendr's web address, bendr.mainthread.ai.

https://mainthread.ai/changelog/#wordship-deduce-feels-familiar 2026-09-24T00:00:00Z

New: Press Play Deduce for a fresh word, then Next word when you finish. An unfinished word waits for you. New: Results show your record across Daily and fresh words: games played, win rate, best and average guesses, and how many guesses each win took. Your profile shows the same record. New: Guests can look back at their earlier Daily boards. Improved: The board is larger on phones, with a filled keyboard like the one you know. Tiles flip one after another, and the keyboard colors after the last one. Improved: Share is one button. Preview & options holds the preview and the choice of text or image. Improved: Answers come from familiar words, skip words you have already seen in Deduce, and leave out a few offensive ones. Improved: Screen readers announce each guess's clues once the tiles settle. Fixed: How to play closes when a round ends, so it no longer covers your result.

https://mainthread.ai/changelog/#wordship-salvo-on-phones-and-open-modes 2026-09-24T00:00:00Z

New: Deduce and Salvo each open with artwork that shows how every mode plays. New: Guests can play Deduce and Salvo practice games without an account. Improved: Practice starts on the first press, and opening a shared replay link no longer starts a game until you press Play. Improved: Salvo's timed word challenge, its keyboard and the latest clue fit on smaller phones. Improved: Coming back to a friend Salvo match shows the guesses behind each ship's score.

https://mainthread.ai/changelog/#system-sentinel-1-9-29 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-28 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-27 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-26 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-25 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-24 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-23 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-22 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-21 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-20 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-19 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-18 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-17 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-16 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-15 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-14 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-13 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-12 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-11 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-10 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-9 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-8 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-7 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-6 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-5 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-4 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-3 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-2 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-1 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-9-0 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-8-0 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-7-0 2026-09-24T00:00:00Z https://mainthread.ai/changelog/#wordship-salvo-for-everyone 2026-09-23T00:00:00Z

New: Play Daily Salvo, the fleet-hunting word game, without an account. New: Start or join a Salvo friend match with a short code, as a guest or signed in. Improved: Home puts Daily Deduce and friend play first. Start makes a room in one tap, and Join opens a code field right there. Improved: Deduce keeps the board and keyboard in view on smaller phones, with a compact line for your opponent's progress. Improved: Salvo asks for your word before showing the board, and while you wait you see your own fleet with misses marked. Improved: The defending friend watches the challenge live: their word, each guess and the countdown. Finished matches offer a new one. Fixed: A guess sent on a weak connection is checked against the game, never lost or counted twice.

https://mainthread.ai/changelog/#system-sentinel-1-6-0 2026-09-23T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-5-0 2026-09-23T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-4-1 2026-09-23T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-4-0 2026-09-23T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-3-9 2026-09-23T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-3-8 2026-09-23T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-3-7 2026-09-23T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-3-6 2026-09-23T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-3-5 2026-09-23T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-3-4 2026-09-23T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-3-3 2026-09-23T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-3-2 2026-09-23T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-3-1 2026-09-23T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-3-0 2026-09-23T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-2-0 2026-09-23T00:00:00Z https://mainthread.ai/changelog/#wordship-play-before-sign-in 2026-09-22T00:00:00Z

New: Play Daily Deduce without an account. Your progress stays with this device. New: Start a Deduce friend room and share a short code. Your friend joins without an account. New: Keep what you played by saving it to a new account, right from the game you are in. New: Choose your look: Chart Room, Deep Signal or Lantern, each in light or dark. Improved: Daily Deduce opens straight onto the board, and a finished match shows the result first, then what to do next. Improved: After you finish a duel, watch your friend's hunt live. Fixed: Only the players in a game can see its answers and results. Fixed: Multi-round matches finish when their last round does, and live updates between friends work again.

https://mainthread.ai/changelog/#multithread-0-4-15 2026-09-22T00:00:00Z https://mainthread.ai/changelog/#multithread-0-4-14 2026-09-22T00:00:00Z https://mainthread.ai/changelog/#multithread-0-4-13 2026-09-22T00:00:00Z https://mainthread.ai/changelog/#multithread-0-4-12 2026-09-22T00:00:00Z https://mainthread.ai/changelog/#wordship-chart-room 2026-09-21T00:00:00Z

New: Wordship takes on Chart Room, a navigator's chart-paper look, in light and dark. Improved: A Salvo hit tells you which ship it struck.

https://mainthread.ai/changelog/#system-sentinel-1-1-0 2026-09-21T00:00:00Z https://mainthread.ai/changelog/#system-sentinel-1-0-1 2026-09-21T00:00:00Z https://mainthread.ai/changelog/#zoning-signal-0-30-0 2026-09-20T00:00:00Z

Every covered county now has a map: 25 jurisdictions across Lake, Orange, Osceola, Polk and Seminole, projected from Census TIGER boundaries, with the worst deviation from the source at 0.62 rendered pixels.

https://mainthread.ai/changelog/#system-sentinel-1-0-0 2026-09-20T00:00:00Z https://mainthread.ai/changelog/#zoning-signal-0-29-0 2026-09-17T00:00:00Z

A new MCP tool finds the routes between two entities, meetings or named patterns.

https://mainthread.ai/changelog/#bendr-1-51-0 2026-09-16T00:00:00Z

Fixed: The Start and Continues key stays at the bottom of the sketch whether or not the controls are showing their labels. It was sitting in the middle of the drawing. Improved: Insert before, Insert after and Remove now open from the bend's own card in the sketch, so the canvas menu is only about the view. Improved: That menu opens below its button in the same place every time, and tapping outside it visibly closes it. Improved: Compact icons is a switch that stays set between sketches, and it is also in Settings under Run display.

https://mainthread.ai/changelog/#bendr-1-50-0 2026-09-16T00:00:00Z

Improved: Follow shows the measured legs around each bend, with more room to understand its place in the run. Improved: Dimensions prioritizes adjoining centerline lengths with clear measurement lines and endpoints. Improved: A compact More menu opens beside its button. Zoom and Follow sit together near the bottom of the canvas. Fixed: Frosted controls have clearer edges. Follow includes an icon in both labeled and compact modes. Fixed: The drawing guide keeps its size when switching between Centerline, Vertex and Turn.

https://mainthread.ai/changelog/#bendr-1-49-0 2026-09-16T00:00:00Z

New: A strip at the bottom of the 3D viewer shows bend lengths, angles, measurement references and status, with arrows for moving through the run. New: Hold a viewer-strip bend to inspect or edit it, then return to your selected bend, camera view and place in the run. Improved: Frosted viewer controls float around the canvas edges, with an aligned toolbar and focused sheets for drawing guides and measurements. Improved: Temporary bend edits use the same controls and Undo history as the calculator. New: Bendr’s new brand identity appears in the app, Run Cards and website, keeping the familiar colors.

https://mainthread.ai/changelog/#bendr-1-48-0 2026-09-16T00:00:00Z

Improved: Camera controls and the rotating orientation guide stay inside the 3D canvas. Select a bend below it to highlight its section and edit it in place. Improved: Undo and Redo stay in the viewer header, with familiar insertion buttons between bends. Fixed: Required corner placement stays visible for both measurement ends. Conflict messages identify the readings that need attention. Fixed: Native sheet headers keep their title and close action aligned. Manual pages wait for their display size before loading and include readable source text. Improved: The release note uses native frosting and clears after a short pause. The marks card groups its guides and calculations into focused help sheets.

https://mainthread.ai/changelog/#bendr-1-47-0 2026-09-16T00:00:00Z

Fixed: Offset explanations open within the 3D viewer, with a clear path back to editing. Improved: Keyboard Done, degree-button positioning and ruler dragging make numeric entry easier. Improved: The 3D editor shares the calculator’s controls. Help headers and bend previews use frosted materials. Improved: Native manual links include the highlighted original pages. Last-leg entry and cut-length styling are tidier.

https://mainthread.ai/changelog/#zoning-signal-0-28-0 2026-09-16T00:00:00Z

People who appear in the record only as public-comment speakers, commenters or neighbouring owners are now "a resident", and their home addresses "[address withheld]", across every reading and meeting record.

https://mainthread.ai/changelog/#bendr-1-46-0 2026-09-15T00:00:00Z

Improved: The top Run Strip stays available while entering lengths and offsets. Swipe it away when you want more room. Improved: Frosted header and Run Strip materials now use native iOS rendering and respect reduced transparency. Improved: Offset measurement choices keep a steady layout, with a visible Measured across label and direct reference buttons. Fixed: Diagram labels use cleaner leaders. Bend direction controls remain available when the floating 3D button is hidden.

https://mainthread.ai/changelog/#bendr-1-45-0 2026-09-15T00:00:00Z

Fixed: Reference conflicts point to the choice that needs attention, while valid lengths keep their normal appearance. Improved: Corner placement comes before its diagram, with clearer labels and a steady preview while choosing the close leg. Improved: Offset measurements have a matching visual inspector that connects the measured distance to the calculated leg. Improved: Recalculation notices stay near the header and combine quick edits. Opening a measurement menu brings its choices into view.

https://mainthread.ai/changelog/#bendr-1-44-0 2026-09-15T00:00:00Z

Improved: Calculator, run display, appearance, startup and help preferences are grouped in one familiar Settings sheet.

https://mainthread.ai/changelog/#bendr-1-43-0 2026-09-15T00:00:00Z

Improved: The reference diagram now follows your selected endpoint, close leg and surface gap, with clear Center and Corner markers and a dimensioned endpoint allowance. Improved: Edit the gap directly, use familiar orange selections, and expand or hide other face references without losing your choice. Improved: Hold a run-strip bend to inspect its corner placement and diagram without leaving your current bend. Fixed: Reference editors stay open independently, unnecessary blank space is removed, and Explain keeps a steady frame as you explore its lessons. Fixed: Reference warnings identify the endpoint that needs attention and appear once per card. The 3D-button hide notice clears sooner.

https://mainthread.ai/changelog/#bendr-1-42-0 2026-09-15T00:00:00Z

Improved: Measurement choices use larger, full-width buttons and a steady layout, with immediate responses instead of resize animations. New: Tap the tube diagram or Explain this reading for illustrated help on centers, inside and outside corners, and tube ends. Fixed: Offset reading choices keep the length-reference editor open, preserving your place while you work.

https://mainthread.ai/changelog/#bendr-1-41-0 2026-09-15T00:00:00Z

New: Choose Center, Inside corner or Outside corner at each end of a measurement. Your original length stays in its field. New: See the measurement points and converted centerline length together. Tap the diagram or Explain this reading for a closer look without leaving the card. Improved: Corner placement, saved runs and shared Run Cards retain your readings. The direction control still shapes the 3D run.

https://mainthread.ai/changelog/#bendr-1-40-0 2026-09-14T00:00:00Z

New: Open Around a column from a bend’s measurement reference. Enter raw wall and column readings, choose the surface gap, and preview four bends before replacing your run. New: See how face readings become centerline lengths and stock marks. Column checks flag lost clearance and show endpoint shifts caused by rounded marks. Improved: Column readings stay with saved and shared runs. Manual leg edits stay yours when the bender changes, and Undo restores the run before a replacement.

https://mainthread.ai/changelog/#bendr-1-39-0 2026-09-14T00:00:00Z

Improved: A white track in light appearance, clearer dark chips, translucent arrows and corner status indicators make the run easier to read. Improved: Minimizing tucks the chips under the header. The restore handle stays reachable, and scrolling the page resumes following your bend. New: Hold a bend, then choose Edit bend to correct its values without leaving your place. Open the sketch when you need more context. Fixed: Recalculation highlights wait until you can see them. Quick navigation respects your latest action, and obsolete offset notes no longer leave empty space.

https://mainthread.ai/changelog/#bendr-1-38-0 2026-09-14T00:00:00Z

New: Enable Run strip in Settings to browse entered bends, step through the run, or hold a bend for a detail preview. New: Double-tap a strip arrow to find the next bend needing attention, or hold it to reach the first or final bend. Improved: Place the strip at the top or bottom and swipe toward that edge to stash it. Offset connections and brief recalculation outlines keep the run’s relationships visible.

https://mainthread.ai/changelog/#bendr-1-37-0 2026-09-14T00:00:00Z

Improved: On iPhone, the length field and fractions move into view together as the keyboard opens. Fixed: Dragging the tube preview or run sketch keeps the surrounding page still. Fixed: Settings uses a consistent background around its rounded corners in dark appearance.

https://mainthread.ai/changelog/#bendr-1-36-0 2026-09-14T00:00:00Z

Fixed: Choosing a fraction keeps the complete length when the native keyboard closes. New: In the native test builds, a downloaded update shows What’s new and a Restart button, and your whole-run entry is saved before restarting. Improved: Native header spacing, help-menu widths and fraction previews are more consistent. Drag the Settings handle down to close it. Fixed: Tube labels keep their inch symbols, and the offset result follows your light or dark appearance.

https://mainthread.ai/changelog/#bendr-1-35-0 2026-09-14T00:00:00Z

New: Starting direction is available during setup and above bend 1. Choose any of the six directions for the first straight section; the sketch and bend choices follow it. Improved: Starting direction stays with Undo, saved runs and Run Cards. Existing measurements and chosen bend directions are kept when you change it. Fixed: Closing a direction menu no longer pulls focus away from a field you have just started editing.

https://mainthread.ai/changelog/#bendr-1-34-0 2026-09-14T00:00:00Z

Improved: Run Cards support native link sharing, copying and deliberate import with Undo, and a shared card can open in the test build. Improved: Native editing sheets make room for the keyboard, and the app has its own Bendr icon and launch screen.

https://mainthread.ai/changelog/#bendr-1-33-0 2026-09-14T00:00:00Z

Improved: The sketch opens with bend dots and a small start/end legend. Turn on Dimensions for measurements and labels; Explain holds the extra guidance. Improved: Bend cards show plain direction names. The chooser keeps Current direction and New bend direction; detailed heading results and relative controls live under How it works.

https://mainthread.ai/changelog/#bendr-1-32-0 2026-09-14T00:00:00Z

New: End offset hides an unused offset box on a following bend. Use offset brings it back; your lengths, angles and marks stay as entered. New: Choose Up, Down, Left, Right, Forward or Back from each bend card. The same directions appear in your 3D sketch; hide the card controls in Settings if you prefer. Improved: Fraction buttons and the ruler follow your chosen increment in Settings. Choosing eighths hides the smaller sixteenths, without changing finer values already entered.

https://mainthread.ai/changelog/#multithread-0-4-11 2026-09-14T00:00:00Z https://mainthread.ai/changelog/#multithread-0-4-10 2026-09-14T00:00:00Z https://mainthread.ai/changelog/#multithread-0-4-9 2026-09-14T00:00:00Z https://mainthread.ai/changelog/#bendr-1-31-0 2026-09-13T00:00:00Z

Fixed: Removed the extra scrolling and blank space added when opening fractions. Length fields use the browser’s normal focus behavior; scroll the page to reach the keypad or ruler.

https://mainthread.ai/changelog/#bendr-1-30-0 2026-09-13T00:00:00Z

Fixed: Length entry prepares the inline fractions before opening keyboard focus, replacing the repeated scroll corrections that caused jumping between bends. Improved: Add bend uses the same immediate entry flow. The keypad and ruler keep their full-size controls, with room to scroll the last field into view.

https://mainthread.ai/changelog/#bendr-1-29-0 2026-09-13T00:00:00Z

New: The floating marks tracker is optional and starts off. Turn it on in Settings for longer runs; full marks remain below your bends. Improved: Routine notices fit their message in a smaller frosted glass surface and fade away automatically. Notices with important follow-up actions stay available. Fixed: Fraction entry responds to late browser scrolling when moving to another bend, while manual scrolling stays in your control.

https://mainthread.ai/changelog/#bendr-1-28-0 2026-09-13T00:00:00Z

Improved: The floating marks fade away while fraction entry is open and return when it closes. Fixed: Opening the phone keyboard scrolls the length and fraction choices into view together when they fit. Dragging the page keeps you in control.

https://mainthread.ai/changelog/#bendr-1-27-0 2026-09-13T00:00:00Z

Fixed: The fraction keypad and ruler now sit below the active input, keeping the measurement clear when the phone keyboard opens. Improved: Keep the equal-size fraction keys and scroll the page normally. No floating panel or separate scroll inside the choices.

https://mainthread.ai/changelog/#bendr-1-26-0 2026-09-13T00:00:00Z

Improved: Fraction buttons and the ruler open in a shared floating editor, without expanding the bend tile. New: An ordered fraction keypad gives every choice the same space. Whole clears just the fraction.

https://mainthread.ai/changelog/#bendr-1-25-0 2026-09-13T00:00:00Z

Improved: Settings groups appearance controls and their explanations, with matching switches for glass effects and the 3D run button. Improved: Run log shows its saved count together. Release notes start with the latest three updates; older ones appear as you scroll. Improved: Angle shortcuts have their own softly shaded group, with more room between turns and fractions.

https://mainthread.ai/changelog/#bendr-1-24-0 2026-09-13T00:00:00Z

New: Turn the sketch toward Up, Down, Left, Right, Forward or Back. The entered angle stays exact, with incoming and outgoing headings beside the control. Improved: Offset connectors stay diagonal, and the preview identifies a return to an earlier heading. Direction help explains the fixed reference and how the length reaches its bend. Improved: Open a bend’s sketch from its options menu. Direction choices stay with Undo, saved runs and shared cards; original relative plane controls remain available.

https://mainthread.ai/changelog/#bendr-1-23-0 2026-09-13T00:00:00Z

Fixed: Calculated offset lengths and their explanations follow your selected measurement increment throughout Bendr. Assigned lengths keep their entry precision, which is identified when it differs. Fixed: Very small offsets no longer blank the page or suggest zero spacing between marks. When the bend arcs need more room, the calculator explains why marking instructions are unavailable and keeps your inputs editable. Improved: Every logged run identifies its recorded measurement increment, including sixteenths. Changing your current setting leaves the saved figures intact.

https://mainthread.ai/changelog/#multithread-0-4-8 2026-09-13T00:00:00Z https://mainthread.ai/changelog/#multithread-0-4-7 2026-09-13T00:00:00Z https://mainthread.ai/changelog/#multithread-0-4-6 2026-09-13T00:00:00Z https://mainthread.ai/changelog/#multithread-0-4-5 2026-09-13T00:00:00Z https://mainthread.ai/changelog/#multithread-0-4-4 2026-09-13T00:00:00Z https://mainthread.ai/changelog/#multithread-0-4-3 2026-09-13T00:00:00Z https://mainthread.ai/changelog/#bendr-1-22-0 2026-09-12T00:00:00Z

Improved: Expand the floating marks where they are, without scrolling away from your entry. Numbered measurements stay readable, and the cut length has its own place. Improved: Cut lengths use a warmer, lighter panel with large, high-contrast digits on the calculator and Run Cards. New: The Logged confirmation offers View, Name and Share. View opens that saved run over the calculator, keeping your place. Improved: Run Cards have a smaller QR footer and clear feedback on their sharing and copying buttons. Logged runs also have a direct Share action. Fixed: Long offset explanations use the full width below the reading selector, so they no longer leave a gap above the input. Improved: Find your Run log in Settings or below the results. Logging the first run no longer adds a row above the inputs.

https://mainthread.ai/changelog/#bendr-1-21-0 2026-09-12T00:00:00Z

New: Review the current run with its entries, reading types and marks together. Tap a bend to edit, then return to the review. New: Create a branded Run Card, choose its details and appearance, then share an image, link or text. The link opens a fixed snapshot and leaves the recipient’s current run alone until they choose to open a copy. New: Give logged runs optional names and keep a short note about the piece or a particular bend. Improved: Floating marks and notices use frosted surfaces with soft edge highlights. Glass effects can be turned off in Settings. New: Keep the screen awake while reviewing, when the browser supports it. Closing the review releases it.

https://mainthread.ai/changelog/#bendr-1-20-0 2026-09-12T00:00:00Z

Improved: The floating marks bar appears softly once a mark is available and you are working in the bend entries. Empty runs have no bar or reserved gap, and typing stays in place. Improved: Hold the 3D button over the trashcan to see it collapse inside. Move away to bring it back, or let go to hide it. Undo restores it.

https://mainthread.ai/changelog/#bendr-1-19-0 2026-09-12T00:00:00Z

Fixed: Manual pages remain visible on Safari when its browser lacks a newer text-reading feature. Source links still open the relevant page in Bendr’s reader. Fixed: The first mark and decimal-entry hint have reserved space, so they do not push your input down as you type. Improved: Offset diagrams show where straight centerlines meet at virtual corners, with aligned dimensions and bend angles. Improved: The Hide target stays fully in view, becomes clearer as you approach, and sits in front of the 3D button so you can see where to release. New: The Offset calculator has an entry-options menu for Ruler or Buttons. Your choice applies throughout Bendr on this device.

https://mainthread.ai/changelog/#bendr-1-18-0 2026-09-12T00:00:00Z

Fixed: The manual reader now handles Safari’s offline-page routing correctly. A separate exact-page reader link stays available if an embedded page cannot load. Improved: The ruler has a centered half-inch mark and a white bracket that slides smoothly before settling on an exact sixteenth. Entry previews reach the edges with a soft fade. Improved: The 3D button gently pulls toward an outlined Hide target, then shrinks into it. Undo in the header can bring it back. Improved: Tube options illustrate diameter, bend radius and wall thickness. Hiding the selected size keeps your measurements and asks you to choose again; Undo restores it. New: Explain in each offset box shows the offset, diagonal leg and measurement types with your values or a labeled example. Improved: Wall thickness keeps a compact field in the same place while editing.

https://mainthread.ai/changelog/#bendr-1-17-0 2026-09-12T00:00:00Z

New: Manufacturer links open the relevant manual page with the section highlighted, and the manufacturer’s video can play in the app, starting at the relevant step. Improved: Size-change notices stay until dismissed and never shift your entries. View details explains the change; Last size change in tool options brings it back. Improved: Drag the 3D button to Hide, with Undo to bring it back. Its dock avoids the marks and notices, and stays put when the keyboard closes. Fixed: Wall thickness is editable before starting a run. The value becomes a field in the same place, with no extra editing row. Improved: Options stay centered while closing. Visual previews show the Buttons and Ruler entry choices.

https://mainthread.ai/changelog/#bendr-1-16-0 2026-09-12T00:00:00Z

New: An orange ruler selects exact sixteenths. Choose Ruler or Buttons in any bend’s options; the preference applies across the app. Improved: Bend options hold Delete. Tool options hold shown sizes and manufacturer links. Illustrated guides explain measuring and marking when you need them. Improved: The tube faces the size choices, with matching typography and orange dimension lines. Restored drafts have a clear review notice. New: Pinch, pan and twist the 3D run with two fingers. Drag its floating button to a corner or slide it to an edge; tap the handle to bring it back. Fixed: Fraction buttons in the 3D viewer now enter the value on their label. Previously each selected one sixteenth too little; the main-screen buttons were correct.

https://mainthread.ai/changelog/#bendr-1-15-1 2026-09-12T00:00:00Z

New: Tap a dimension, dot or tube segment to inspect it. Crowded labels become anchored dots; zooming in reveals more. New: Previous and Next walk through your run. Focus brings a feature closer and follows along; Fit returns to the whole piece. New: Attributes lists every drawn landmark, including overlapping ones. Other side turns the camera without changing your bend directions.

https://mainthread.ai/changelog/#bendr-1-15-0 2026-09-12T00:00:00Z

New: Open 3D run to inspect pinned lengths and bend angles. Edit the same run here, including offsets, fractions, new bends and Undo. New: Explain highlights the centerline, virtual vertex and turn angle on your own run. Sketch directions change only the drawn shape. Improved: The tube sample carries its diameter and wall dimensions. Tilt it, release, and it springs home. Reduced motion is respected.

https://mainthread.ai/changelog/#bendr-1-14-0 2026-09-12T00:00:00Z

New: Drag the tube to look around it. Reset view brings it back; arrow keys work too. Fixed: The tube is hollow: both openings and the inside wall stay aligned as the size or viewing angle changes.

https://mainthread.ai/changelog/#bendr-1-13-1 2026-09-12T00:00:00Z

Fixed: Size choices stay in place from the first tap, with larger circles to compare the diameters. Improved: Wall warnings show on Start a run, Whole run and Offset. Typing a wall no longer animates the bore from zero.

https://mainthread.ai/changelog/#bendr-1-13-0 2026-09-12T00:00:00Z

Improved: Tube size has a visual selector. The cross-section changes with the diameter and wall thickness you enter. Improved: Tap the wall value to edit it. Unresolved wall readings stay visible when the editor is closed.

https://mainthread.ai/changelog/#bendr-1-12-1 2026-09-12T00:00:00Z

Fixed: The logo files are far smaller, so a fresh open on a weak signal is not waiting on them.

https://mainthread.ai/changelog/#bendr-1-12-0 2026-09-12T00:00:00Z

Improved: A new look in charcoal and off-white, light and dark, where one orange means the mark: the figures you cut to, the selected choice and the bend. New: The Bendr logo, in the title bar and on the start screen.

https://mainthread.ai/changelog/#bendr-1-11-0 2026-09-12T00:00:00Z

New: Once the app has opened on your phone, it opens and works out marks with no signal. New: A line at the bottom says when it is running the copy on the phone, and with a signal it always fetches the newest build first.

https://mainthread.ai/changelog/#bendr-1-10-0 2026-09-12T00:00:00Z

Improved: The title bar is translucent: the run shows through it, blurred, as you scroll. Light and dark both.

https://mainthread.ai/changelog/#bendr-1-9-0 2026-09-12T00:00:00Z

New: Settings has Appearance: System, Light or Dark. System follows the phone.

https://mainthread.ai/changelog/#bendr-1-8-0 2026-09-12T00:00:00Z

New: In Settings, choose whether marks read to the nearest 1/16, 1/8, 1/4 or 1/2. The cut is the last mark plus the last leg rounded up. New: Show the math still gives the exact, unrounded figures, and what you type is not changed. New: Choose the page the app opens on: last used, Whole run, or Offset. New: A stock wall thickness to fill in whenever a run starts. Improved: A logged run remembers how fine its marks were, so opening it later compares like with like.

https://mainthread.ai/changelog/#bendr-1-7-1 2026-09-12T00:00:00Z

Fixed: The Undo and Redo arrows are the familiar curved ones, and each is a full-size button again.

https://mainthread.ai/changelog/#bendr-1-7-0 2026-09-12T00:00:00Z

New: Settings, top right. Release notes live here, so when the version goes up you can see what changed. New: After an update, a note at the bottom of the screen says so and opens these notes. Improved: Undo and Redo are arrows only, since they are rarely needed; Start new keeps its word. The link to the build before this one moved into Settings.

https://mainthread.ai/changelog/#bendr-1-6-0 2026-09-12T00:00:00Z

Fixed: Start new, opening a logged run and starting a fresh run over an old one are each one Undo away, even after the phone reloads the page. Improved: The run set aside by Start new no longer needs its own card and warnings; Undo brings it back. Fixed: The version menu opens where you are looking, however far down the run you have scrolled.

https://mainthread.ai/changelog/#bendr-1-5-1 2026-09-12T00:00:00Z

Improved: The caret by the version is one you can actually see, and the version is as easy to tap as the buttons beside it.

https://mainthread.ai/changelog/#bendr-1-5-0 2026-09-12T00:00:00Z

Fixed: Start new used to clear everything and delete the saved run. It now drops the tube size only; the measurements stay. Fixed: Undo, Redo and Start new stay in the title bar on every screen and grey out instead of disappearing. Improved: Log cards open without closing each other, so the card you tap does not jump. Improved: A caret by the version number, so it reads as a menu.

https://mainthread.ai/changelog/#bendr-1-4-0 2026-09-12T00:00:00Z

New: Pick the tube size first, then how many bends. Nothing is worked out until the size is chosen, and it is not carried over from the last run. New: The bend count from your notes lays out that many boxes, so an empty one at the end is a bend you missed. Improved: Each leg has one list of readings in your words: Ref end, Center, End and Far end. Improved: Choosing a turn no longer jumps to the next bend. Improved: Length and turn come first in each box, there is more space between inputs, and the "even" chip is gone.

https://mainthread.ai/changelog/#multithread-0-4-2 2026-09-12T00:00:00Z https://mainthread.ai/changelog/#multithread-0-4-1 2026-09-12T00:00:00Z https://mainthread.ai/changelog/#multithread-0-4-0 2026-09-12T00:00:00Z https://mainthread.ai/changelog/#multithread-0-3-0 2026-09-12T00:00:00Z https://mainthread.ai/changelog/#multithread-0-2-0 2026-09-12T00:00:00Z https://mainthread.ai/changelog/#adaptive-rrf-cycle-15 2026-09-12T00:00:00Z

Adding SPLADE++ to four lexical lists raised mean binary nDCG@10 from 0.667 to 0.687 across 300 queries, but SPLADE++ alone (0.704) and a simple two-source hybrid (0.705) scored higher. Limit: One collection with conventional binary scoring supports no new fusion method, generalization or significance claim.

https://mainthread.ai/changelog/#bendr-1-3-0 2026-09-11T00:00:00Z

New: Log this run keeps what you typed, how each leg was read, and the marks and cut you were shown, so a piece that does not fit can be traced back. New: Open a logged run to read it, or open it back into the calculator; it is reworked with today’s arithmetic and any difference is named.

https://mainthread.ai/changelog/#lichtenberg-2026-09-11-d 2026-09-11T00:00:00Z

Rereading the saved predictions, the two models' choices sat further from each graph's actual route weights than a uniform guess would. Limit: It is a look back at results already seen, with no new model runs, and identifies no internal mechanism.

https://mainthread.ai/changelog/#lichtenberg-2026-09-11-c 2026-09-11T00:00:00Z

Across 512 paired prompts, swapping the requested goal shifted the models' route choices by about a hundred-thousandth of what an exact reference shifts, and their routing stayed slightly worse than uniform. Limit: The tiny responses that remain do not show useful goal use, and cannot say whether they are noise or come from any internal mechanism.

https://mainthread.ai/changelog/#lichtenberg-2026-09-11-b 2026-09-11T00:00:00Z

Analysis showed that copying the requested ending and following the list of legal moves reaches 97.73 percent success while ignoring the graph's route weights, and a paired audit was designed around that shortcut. Limit: Designed and not yet run. The shortcut is a worked control on reduced models, and it was not fitted to the trained ones.

https://mainthread.ai/changelog/#lichtenberg-2026-09-11-a 2026-09-11T00:00:00Z

Two small neural networks trained on the graphs reached the requested ending with 97.5 and 94.6 percent success, computed exactly over every graph and goal tested, yet their route choices were further from the ideal than unguided walking. Limit: Reaching the goal often does not show the models learned goal-guided routes, and interpretation was deferred.

https://mainthread.ai/changelog/#adaptive-rrf-cycle-14 2026-09-11T00:00:00Z

The cycle fixed a new scientific-abstract collection, five systems and the metric bridge before inspecting any effectiveness result. Limit: Preparation only: no retrieval effectiveness was measured.

https://mainthread.ai/changelog/#adaptive-rrf-cycle-13 2026-09-11T00:00:00Z

Under a fixed top-ten diagnostic metric, adding SPLADE++ to four-source rank fusion raises the mean on both query sets under every allowed completion of missing judgments, though 9 queries favor the original fusion. Limit: This is a finite result on heavily reused development queries. It shows no general superiority and gives no basis for routing queries.

https://mainthread.ai/changelog/#adaptive-rrf-cycle-12 2026-09-11T00:00:00Z

An inspection of existing retrieval data and one real provenance question found that direct lookup and hash checking already answer it, so no new solver or multi-agent experiment was run. Limit: This is a retrospective inspection by investigators who knew the project history. No preregistered experiment was run.

https://mainthread.ai/changelog/#adaptive-rrf-cycle-11 2026-09-11T00:00:00Z

A single language-model verifier gave 4 of 4 correct answers and 16 of 16 correct validity judgments whether valid support was the minority or the majority, leaving no gap here for a multi-agent stage to repair. Limit: Direct program solving could produce every correct output, so the result does not show the verifier relied on the evidence.

https://mainthread.ai/changelog/#adaptive-rrf-cycle-10 2026-09-11T00:00:00Z

Across six calls, a language model returned all 18 trace judgments correctly, a pattern no named position, constant or endpoint-only shortcut reproduces. Limit: The panel reuses six traces from one arithmetic program, so it shows correct outputs on that program and does not measure general verification ability.

https://mainthread.ai/changelog/#adaptive-rrf-cycle-09 2026-09-11T00:00:00Z

A language model correctly accepted one valid execution trace and rejected two invalid ones, before and after copies of a bad trace were added. Limit: A simpler positional policy also scores perfectly here, so the result does not show step-by-step checking.

https://mainthread.ai/changelog/#lichtenberg-2026-09-10-k 2026-09-10T00:00:00Z

A bounded test of whether a small neural network trained on these graphs learns to steer toward a requested goal was designed, and a separate check with no training showed its passes stay within an ordinary processor's limits. Limit: Nothing was trained, so no model had learned anything yet, and all six predictions about internal mechanisms remained untested.

https://mainthread.ai/changelog/#lichtenberg-2026-09-10-j 2026-09-10T00:00:00Z

Rechecking the saved values, all 216 pairs still called for different first moves when the final move was forced to be the same, so a later chance to revise does not explain the difference. Limit: It holds for one problem family, horizon and reset price. Six single decisions changed when the next state could be seen, but that did not pass the controlled test.

https://mainthread.ai/changelog/#lichtenberg-2026-09-10-i 2026-09-10T00:00:00Z

Across all 27,225 reachable states of a small two-episode problem, 216 pairs share a compressed summary and the same best single move yet need opposite first moves once reinforcement carries into a second episode. Limit: These histories were prescribed, so they say nothing about how often this happens.

https://mainthread.ai/changelog/#lichtenberg-2026-09-10-h 2026-09-10T00:00:00Z

An exact calculation of the range of goal chances consistent with a compressed memory settled the right action in 508 of 512 cases at one set of prices. Limit: It uses one fixed price profile on one graph and supports no claim about average efficiency or the general value of extra information.

https://mainthread.ai/changelog/#lichtenberg-2026-09-10-g 2026-09-10T00:00:00Z

On a small diamond-shaped graph, searches that first learned the wrong ending recovered in 231 of 256 runs when their route memory was kept, against 255 of 256 when the preference for that ending was erased. Limit: It covers one small graph with fixed settings, and which intervention looks best depends on how recovery is measured.

https://mainthread.ai/changelog/#lichtenberg-2026-09-10-f 2026-09-10T00:00:00Z

Sixty-four fresh language-model calls that varied the suspected causes produced no omissions, and all 256 retained values were correct. Limit: Every task returned to its starting state, so copying the input would also score perfectly, and the earlier failure stays unexplained.

https://mainthread.ai/changelog/#lichtenberg-2026-09-10-e 2026-09-10T00:00:00Z

Over 48 language-model sessions every answer the record supported came back correct, but both conditions dropped the same known earlier value on identical input, and access to an archive repaired it at the next step. Limit: It is one failure pattern on one pair of identical inputs, and gives no error rate.

https://mainthread.ai/changelog/#lichtenberg-2026-09-10-d 2026-09-10T00:00:00Z

Extracting records from 11 fixed histories, a language model reached the 55 percent ceiling that a two-record lookup allows under both prompts, so the added guidance showed no gain. Limit: The task hit its ceiling, so it cannot show whether guidance helps, and it shows narrow model behavior, with no mechanism identified.

https://mainthread.ai/changelog/#lichtenberg-2026-09-10-c 2026-09-10T00:00:00Z

An exact cost model comparing 20 fixed policies over 160 traces showed the cheapest one depends on the prices assumed: look information up when an archive is cheap, keep it when the archive is expensive. Limit: The costs are declared assumptions, and nothing here tests a learned summary or a neural network.

https://mainthread.ai/changelog/#lichtenberg-2026-09-10-b 2026-09-10T00:00:00Z

As reinforcement grew stronger, searches whose route memory started on the wrong answer recovered in 247, then 59, 3 and 0 of 256 runs. A sharp final answer appeared even without memory, so sharpness alone does not show memory at work. Limit: It covers this exact model only, and did not test for a sudden tipping point or for trained networks.

https://mainthread.ai/changelog/#lichtenberg-2026-09-10-a 2026-09-10T00:00:00Z

Rereading the saved runs showed the combined system held on to a found route far more reliably than the comparison with a degraded guide (0.999 against 0.409) but found it more slowly, so the claim of faster discovery was withdrawn. Limit: No new experiment was run, and those numbers came from an early controller that used the true goal to decide what to reinforce, changed before wrong goals were tested.

https://mainthread.ai/changelog/#adaptive-rrf-cycle-08 2026-09-10T00:00:00Z

Eight matched programs passed their construction checks, but an answer imbalance meant an always-false guess could mimic a difficulty effect, so the solver batch was parked. Limit: No model responses were collected in this cycle.

https://mainthread.ai/changelog/#adaptive-rrf-cycle-07 2026-09-10T00:00:00Z

Two prompt variants answered all eight small programs correctly, so the gate requiring observable errors and disagreement failed and the setup was parked. Limit: This is a negative feasibility result for one small task and says nothing about solver accuracy in general.

https://mainthread.ai/changelog/#adaptive-rrf-cycle-06 2026-09-10T00:00:00Z

Three valid responses closely matched the noise-aware answers, but the provider refused the fourth call and the fixed workflow stopped with four inputs unsent. Limit: With five of eight planned inputs lacking a valid response, the primary comparison is null.

https://mainthread.ai/changelog/#adaptive-rrf-cycle-05-replication 2026-09-10T00:00:00Z

With the collector fixed, a language model returned probabilities within 0.00000004 of the exact Bayesian answers on all 20 fixed requests, including the difference between copied and independent agreement. Limit: This covers one small, fully specified synthetic model. It does not measure real-world calibration or general reasoning.

https://mainthread.ai/changelog/#adaptive-rrf-cycle-05 2026-09-10T00:00:00Z

The collector misread empty agent counters as activity and stopped the batch after its first response, leaving the planned test incomplete. Limit: The stop came from a measurement bug in the collector, and no model reasoning failure was involved.

https://mainthread.ai/changelog/#adaptive-rrf-cycle-04 2026-09-10T00:00:00Z

In an exact synthetic model, knowing which reports share an origin improves probability estimates, while always protecting the dissenting minority makes them worse. Limit: These are properties of a deliberately constructed model. No language-model gain was measured and no new aggregation method is claimed.

https://mainthread.ai/changelog/#adaptive-rrf-cycle-03 2026-09-10T00:00:00Z

Neither year's link between corroboration and relevance cleared the preset ten-point threshold across resampling, so the protocol stopped this branch and moved to a model with known truth. Limit: Stopping was a resource decision and leaves open whether specialist sources are useful.

https://mainthread.ai/changelog/#adaptive-rrf-cycle-02 2026-09-10T00:00:00Z

SPLADE++ supplies many more judged examples than the lexical sources of top-ten documents no other source ranks highly, but for those documents, support from other lists cannot be separated from whether the old lexical candidates contained them at all. Limit: This cycle counted candidates and judgments only; no ranking metric or fusion result was computed.

https://mainthread.ai/changelog/#adaptive-rrf-cycle-01 2026-09-10T00:00:00Z

After correcting the rank offset and tie handling, a tuned k still helps one six-ranker setup, but it lowers the original four-ranker setup and both transfers between the two query sets lose. Limit: Results come from 43 and 54 reused queries, generated lexical rankers and partial relevance judgments.

https://mainthread.ai/changelog/#bendr-1-2-0 2026-09-07T00:00:00Z

New: A leg or an offset can be measured to a wall at either end, and half the tube comes off per wall. Only where that bend is 90°.

https://mainthread.ai/changelog/#bendr-1-1-1 2026-09-07T00:00:00Z

Improved: The circle between bends is unfilled, its plus centred, and it sits clear of the cards.

https://mainthread.ai/changelog/#bendr-1-1-0 2026-09-07T00:00:00Z

New: A circle on the seam between two bends inserts one there. Improved: The note after Undo or Redo sits under the marks strip instead of over it. New: The version number shows under the title, so a screenshot says which build it is.

https://mainthread.ai/changelog/#bendr-1-0 2026-09-05T00:00:00Z

New: Enter legs and turns to get the cut length and every mark measured from the reference end, using the adjustment table in Swagelok’s published hand bender manual. New: Sixteenth-inch entry, an Offset page, and your entry saved on the phone.

https://mainthread.ai/changelog/#ascend-2026-08-30-3 2026-08-30T00:00:00Z

Sign in with less repetition, choose an optional first-workout guide, and retry exercise loading without losing your place. New: When your training history is empty, Home offers a one-time invitation into the existing five-step walkthrough; dismiss it, or replay it anytime from Settings. Improved: The required 16+ confirmation still happens before sign-in begins; the email code sheet no longer asks for the same confirmation again. Fixed: When no saved exercises are available and the catalog request fails, the picker explains the connection problem and offers Retry without closing the sheet or clearing search and muscle filters.

https://mainthread.ai/changelog/#ascend-2026-08-30-2 2026-08-30T00:00:00Z

Design and share a completed workout's Lift Card while its invitation link, QR, code, and friend-send wait for a connection. New: Ascend builds the Lift Card from the completed workout already on your device, so unavailable invitation setup no longer closes the designer or blocks outward sharing. Improved: When the connection returns, Retry adds the link to perform the workout, the QR, the code, and Send workout in Ascend; until then, the shared card leaves out its invite footer. Improved: When a detailed set list will not fit, Ascend asks for Best set or Counts instead of cropping completed rows. Fixed: If the active account changes while friend-send is open or in flight, Ascend expires that share instead of reusing its workout or showing a stale result.

https://mainthread.ai/changelog/#ascend-2026-08-30-1 2026-08-30T00:00:00Z

Decline dismisses a request; Block removes the connection and shared workout access, then keeps both people out of each other's Ascend social surfaces. Fixed: Not now and Decline remove only the pending request, so either person can reconnect later. Fixed: Open social screens stop showing the blocked person immediately while fresh data reloads. Improved: Block removes the friendship and both directions of shared workout access, then keeps both people out of each other's requests, profiles, activity, standings, live status, friend workouts, and inbox.

https://mainthread.ai/changelog/#zoning-signal-0-27-0 2026-08-15T00:00:00Z

Orlando, Kissimmee, Mount Dora, Eustis, Tavares and Montverde join the corpus: 83 harvested documents, six place dossiers, 25 cities in all.

https://mainthread.ai/changelog/#ascend-2026-08-04-1 2026-08-04T00:00:00Z

On a plate-loaded exercise, type a working weight and see a live plate summary before you leave the keypad. New: On barbell and configured plate-loaded exercises, the keypad turns each working weight into a live plate summary without taking you away from the set row. New: Open the plate breakdown to choose barbell or machine, set the bar when one is involved, keep only the plate sizes your gym has, and decide how Ascend should balance and count the load. Improved: Tap the target to type an exact weight, tap plus or minus for a small change, or hold either control to move quickly through a larger adjustment. Improved: The compact breakdown distinguishes exact from closest loads, states whether counts are per side or total, and keeps every visible plate term readable before you lift.

https://mainthread.ai/changelog/#ascend-2026-08-01-1 2026-08-01T00:00:00Z

Set values now save as you type, whether or not the set is marked complete. Improved: Weight, reps, reps left, duration, and distance are saved on your device as drafts with each valid entry, with no Next or checkmark needed to keep them. Improved: An unchecked row stays a draft; only the checkmark logs the set and lets it affect volume, records, rest, recovery, and workout timing. Fixed: Leaving, backgrounding, or restarting Ascend no longer resets valid values that were still sitting in an unchecked set row.

https://mainthread.ai/changelog/#ascend-2026-07-31-3 2026-07-31T00:00:00Z

Recover forgotten workouts with a considerate reminder, an honest end time, and room to log work you missed. New: When an active workout appears forgotten, Ascend gives you a gentle, cardio-aware reminder; opening it never finishes the workout, and you review the end first. New: If you trained but forgot to log the last work, enter catch-up mode and add it against the time training actually ended instead of hours of later idle time. Improved: Review the last completed set, choose a custom end when needed, and let Finish wait for pending work and server confirmation before the workout closes. Improved: Still training, Not now, and reminder taps preserve your place, survive an app restart, and stay scoped to the exact workout and activity moment that created them. Fixed: Returning to a forgotten workout no longer forces you to accept an inflated duration or manually reconstruct its end from memory.

https://mainthread.ai/changelog/#ascend-2026-07-31-2 2026-07-31T00:00:00Z

See accurate real-world comparisons everywhere your workout volume appears. New: The comparison library now spans regulated gym equipment, everyday objects, named vehicles, animals, spacecraft, and landmarks, each with its source recorded. New: Your immediate workout recap and web printout now carry the same comparison already available in completed-workout stories and Lift Cards. Improved: Every comparison is computed in kilograms, uses named configurations and whole animals, and keeps its rounding error bounded. Improved: History, League Activity, the finish recap, Lift Cards, text shares, the volume HUD, and print all render one deterministic shared result. Fixed: A 12,000 lb workout now reads as about 2.7 named base Ford F-150 pickups instead of one vague, drastically overweight truck.

https://mainthread.ai/changelog/#ascend-2026-07-31-1 2026-07-31T00:00:00Z

Open completed workouts as rich stories of volume, records, and every set. New: Completed workouts now lead with centered volume, a fun real-world comparison, tiled stats, and a responsive Training Log for strength and cardio. New: The Progress Ledger replaces generic record prose, while exact record labels sit with the sets that earned them whenever that finish-time detail is available. Improved: Your workouts and a friend's now share the same rich recap, with only the actions and private details that belong to each viewer. Improved: Friend identity and the close control stay visible over patterned and colored Display backgrounds. Fixed: Later workouts can no longer erase records genuinely earned in an earlier recap, and incomplete shared data stays unknown instead of being guessed.

https://mainthread.ai/changelog/#ascend-2026-07-30-1 2026-07-30T00:00:00Z

See a truthful end time and open detailed controls over your workout without stopping the countdown. New: Detailed rest controls now open in a layered panel above the active workout, keeping the next set and previous performance within reach. Improved: Rest alerts now tell you the exact local end time instead of a remaining-seconds estimate that can age before it reaches the phone. Improved: See the live countdown, add or subtract 15 seconds, and toggle cue sounds without clearing the timer. Fixed: Back to workout, a drag, Escape, Android Back, or Close now preserves the same rest; only End rest clears it. Fixed: Replacing or skipping rest now revokes stale alerts that no longer belong.

https://mainthread.ai/changelog/#ascend-2026-07-29-1 2026-07-29T00:00:00Z

Weight, set type, muscle focus, and duration now live where you expect them, while the active row stays visible as you type. New: Tap LB or KG to choose total or per-hand load and set workout-specific sled weight, with clear guidance about what changes past and current volume. New: Tap the set number to choose any of seven set types, including Myo-rep and Rest-pause mini-sets, or to open muscle focus without leaving the exercise. Improved: CONFIRM LB, CONFIRM REPS, and LOG SET replace the old ambiguous action and redundant LEFT shortcut. Improved: For duration, 5 now means 5:00; :30 and 1:30 make seconds explicit, with MIN : SEC guidance beside the input. Improved: Lower set fields move above the logger dock and stay put while you advance, dismiss and reopen, or switch rows quickly. Fixed: A completed or discarded workout no longer returns through Resume while an older sync is catching up.

https://mainthread.ai/changelog/#ascend-2026-07-27-2 2026-07-27T00:00:00Z

Browse and add exercises through one clear library, with muscle and training context before you choose. New: Picker cards, custom exercises included, now show primary and secondary muscle focus, how many completed workouts used the exercise, and when it was last used. New: Hold an exercise to open its History, Charts, Records, and more above the picker, then return to the same search and selection. Improved: The Start library, exercise details, and active-workout picker now share your Display backdrop, cards, search, tabs, and protected controls. Fixed: Start's main action no longer changes appearance after the console opens, and workout option controls remain visible over camo and colored backgrounds. Fixed: With search open, one ADD closes the keyboard and picker and inserts once; Android Back closes the top layer instead of escaping the workout.

https://mainthread.ai/changelog/#ascend-2026-07-27-1 2026-07-27T00:00:00Z

Keep records and improvements visible in a live Progress Ledger, beside a calmer rest controller that belongs to the workout. New: Records and meaningful set improvements now collect in a live Progress Ledger, so the work you are building does not disappear with a brief signal. Improved: Rest remains glanceable at the bottom of your workout, with clearer timing controls and more room for the session itself. Improved: In Rest Cues, starting rhythms return to compact cards with a trailing peek, and the carousel runs edge to edge without a dark enclosing trough. Fixed: Start now opens from one place, so returning from Settings cannot leave a second console or an opaque footer behind. Fixed: Finishing or discarding a workout now clears its active rest, including scheduled cues that should no longer reach you.

https://mainthread.ai/changelog/#ascend-2026-07-25-5 2026-07-25T00:00:00Z

Open a calmer Start console that carries your selected color, texture, and panel treatment from the rest of Ascend. Improved: The Start console now uses your selected color or Digital Camo canvas, and both its action keys and saved workouts honor your Cards & panels treatment. Improved: Saved workouts stay visible while Ascend refreshes them, a failed refresh offers Retry, and New routine remains available in every state. Improved: Start now has a visible close control, responsive Activity controls for smaller phones, and links only to destinations that work. Fixed: The direct empty-workout and saved-routine buttons now wait for the live-session check, defer to Resume when one exists, and offer an explicit offline fallback when that check fails. Fixed: Closing Start cancels a pending start, and rapid close and reopen or nested sheets no longer leave the console catching taps.

https://mainthread.ai/changelog/#ascend-2026-07-25-4 2026-07-25T00:00:00Z

Carry your chosen backdrop, panel style, and one calm navigation rhythm through every Settings destination. Improved: Display, Rest cues, What's New, Using Ascend, and About & Legal now share the same title scale, content breathing room, and navigation grammar. Improved: Rest Cues now carries your selected color or Digital Camo backdrop, while its main panels honor your Cards & panels treatment. Improved: Bare glyphs replace boxed Settings icons, and Using Ascend now keeps the same comfortable inset as What's New. Fixed: Frosted Glass and locked-screen cue descriptions now name the visual and sound behavior that each surface can actually deliver.

https://mainthread.ai/changelog/#ascend-2026-07-25-3 2026-07-25T00:00:00Z

Shape light and dark looks, color and texture, and cards and panels through one fluent visual designer. Improved: Your live Home-instrument preview now leads into three stable destinations: Light & dark, Color & texture, and Cards & panels. Improved: Choose Clean or Digital Camo first, then shape its color, block size, and intensity in the same focused workspace. Improved: Skin ramps, camo fields, and surface treatments now use generous preview cards so each choice reads before you make it. Fixed: Editors open in one reusable pushed panel, and a rapid tap made while it closes is queued instead of disappearing.

https://mainthread.ai/changelog/#ascend-2026-07-25-2 2026-07-25T00:00:00Z

Read each effort value directly beneath its Reps Left label while load and rep values keep their clean left edge. Improved: Previous and Current now sit directly beneath each exercise title, while two exercises share each Training Log row so more of the workout remains visible. Improved: The Previous column now includes the reps-left value logged for each lifting set, and leaves unlogged values absent rather than guessing them. Fixed: Comparison fit checks now model the same two-column rows the card renders, including uneven and odd exercise counts, before sharing is enabled.

https://mainthread.ai/changelog/#ascend-2026-07-23-5 2026-07-23T00:00:00Z

Matched Story and Post headers, adaptive camo contrast, and dependable designer panels make every Lift Card easier to shape and share. Improved: Story and Post now share the same left-anchored Ascend lockup and right-aligned workout identity, with a cleaner rhythm into Volume Moved. Improved: Bright or complex camo places key copy on restrained HUD surfaces only when contrast needs help; quiet looks stay open and clean. Fixed: Full Preview, Look, and Content now reopen reliably after Back or a camo edit, with no need to switch formats to reset the designer. Fixed: Designer content now continues behind the home-indicator area while keeping the final controls safely reachable by scrolling.

https://mainthread.ai/changelog/#ascend-2026-07-23-4 2026-07-23T00:00:00Z

Preview the complete card, compose any digital-camo color, and share Story or Post layouts that keep every detail clear. New: Tap the working preview to inspect the complete Story or Post before it leaves your hands, then return without losing a choice. New: Compose any digital-camo color with Fine, Medium, or Chunky blocks and Subtle or Bold intensity for this card alone. Improved: Story uses more of its tall canvas while respecting social overlays, and volume plus workout stats now sit in a stronger centered rhythm. Improved: Share Lift Card opens the image share menu; Send Workout opens an Ascend friend inbox layer and returns to your design when you close it. Fixed: Switching between Story and Post keeps volume and exercise names visible, while long training logs remain inside their card region.

https://mainthread.ai/changelog/#ascend-2026-07-23-3 2026-07-23T00:00:00Z

Turn a completed workout into a branded Lift Card for Stories or posts, choosing the look and exactly how much detail feels right to share. New: Compose a Story or portrait post from one live preview, then share the exact card you designed. New: Start with Minimal, Highlights, or Full, then decide whether to include volume, its fun comparison, workout stats, exercises, and weights and reps. Improved: Every card carries the Ascend Lifting League lockup and a compact scannable workout invitation built for social safe zones. Improved: Use your current Display or choose Onyx, Paper, Cyan, or Violet for the card alone, and the app itself stays exactly as you configured it. Fixed: Image and text-sharing fallbacks now honor the same disclosure choices, and an unavailable personal record stays unknown instead of becoming a false zero.

https://mainthread.ai/changelog/#ascend-2026-07-23-2 2026-07-23T00:00:00Z

One themed workout share now carries the recap, QR, link, code, and custom exercises, and a friend can perform it or request beta access from that same invitation. New: Every workout recap now carries the same scannable QR, link, and six-character code, so the person who sees it can bring that workout into their own training. New: Friend and workout links now explain what was shared, preserve it through sign-in, and let a new lifter request closed-beta access by email. Improved: The separate Send and Share controls are now one clear flow for the recap image, workout code, link, and direct friend send. Improved: Shared workout recaps now follow your active light or dark palette and digital-camo composition instead of using one fixed export style.

https://mainthread.ai/changelog/#ascend-2026-07-22-2 2026-07-22T00:00:00Z

A calmer Settings control center keeps your profile, training, privacy, support, and account controls clear without showing everything at once. New: Profile, training, privacy, learning, account, and diagnostics now begin in one calm, grouped home instead of one long wall of controls. New: Your identity now has a focused home where your display name and optional @handle can be reviewed and saved as one intentional change. Improved: Training, privacy, data rights, and diagnostics open as focused destinations, keeping the first screen short without removing a single capability. Improved: Bodyweight and handle prompts now open the exact Training or Profile setting they refer to.

https://mainthread.ai/changelog/#ascend-2026-07-22-1 2026-07-22T00:00:00Z

Pull to refresh every League view, with standings that make the current week and a fresh start clear. New: Pull down on Activity, Standings, or Friends to fetch the latest League state, with a light haptic the moment your pull commits. Improved: Standings now name the current Monday–Sunday range, so every total has a clear timeframe. Fixed: When a new week opens at zero, Standings holds back rankings and explains the fresh start instead of placing everyone at #1.

https://mainthread.ai/changelog/#ascend-2026-07-21-1 2026-07-21T00:00:00Z

Display now starts with one complete preview and reveals each layer of customization only when you want it. New: See Training, Calendar, or Readiness inside the complete appearance, color, texture, and surface composition. New: Appearance, Color, Texture, and Surface now open as focused editors instead of one long wall of controls. Improved: Named digital camo colors now carry the app's Action accent while Recovery adapts to stay unmistakably distinct. Improved: Block, intensity, frost, and depth controls appear only when the selected material or surface can use them. Fixed: Training now states the visual rule plainly: more blocks means more sets, and brighter means greater effort.

https://mainthread.ai/changelog/#ascend-2026-07-20-4 2026-07-20T00:00:00Z

Ascend now explains what changed, why it matters, and exactly which version is on your device. New: The update-ready notice now carries the next release's title, purpose, and highlights. New: A permanent, offline release feed in Settings keeps every meaningful improvement easy to find. Improved: Version, native build, runtime, and product release now have distinct, understandable labels. Improved: New-release education stays dismissible and never interrupts a set or forces an update.

https://mainthread.ai/changelog/#ascend-2026-07-20-3 2026-07-20T00:00:00Z

A new Calendar instrument turns the rhythm of training and recovery into a navigable part of the HUD. New: Choose Calendar beside Training and Readiness, then move month by month through your history. New: Tap a training, recovery, or open day to reveal its details and the action that belongs there. Improved: HUD stats and volume comparisons now follow the selected month instead of pretending it is 30 days. Improved: One-row and two-row stat clusters both scroll fully above the floating navigation when a day opens.

https://mainthread.ai/changelog/#ascend-2026-07-20-2 2026-07-20T00:00:00Z

Finding, inviting, and understanding training partners now lives in one clearer social home. Improved: Friends moved into League, beside the activity and standings that give those relationships meaning. Improved: Codes, QR scanning, handles, and your own invite are layered into one keyboard-safe flow. Improved: A high-contrast active selector slides between destinations and stays clear over patterned backgrounds.

https://mainthread.ai/changelog/#ascend-2026-07-19-1 2026-07-19T00:00:00Z

Rest timing now feels intentional: a rising three-count, a decisive finish, and cues composed around how you train. New: A rising count and a final go cue replace generic timer sounds with a recognizable signature. New: Choose the moments, reach, sound, and vibration that fit a felt-sense workout or a strict cadence. Improved: Rest-over alerts and live timing can keep working when the phone is no longer in your hand.

https://mainthread.ai/changelog/#lichtenberg-2026-07-11-a 2026-07-11T00:00:00Z

The first runs on one family of constructed graphs were summarized at the time as four predictions confirmed, one measuring instrument failed, and a mixed verdict on how a wrong goal signal leads the search astray. Limit: That summary was later superseded: the faster-discovery prediction was unsupported, and the early controller had used the true goal when deciding what to reinforce.

https://mainthread.ai/changelog/#zoning-signal-0-26-0 2026-06-04T00:00:00Z

The corridor's binding constraint moves from roads to water.

https://mainthread.ai/changelog/#zoning-signal-0-16-0 2026-05-23T00:00:00Z

Ten corpus releases in two days.

https://mainthread.ai/changelog/#zoning-signal-0-15-0 2026-05-20T00:00:00Z

The agent surface now logs every tool call, accepts feedback from agents through a new MCP tool, and publishes what it reads at /agent-observatory.

https://mainthread.ai/changelog/#zoning-signal-0-14-0 2026-05-10T00:00:00Z

The lens values agents use to request a stakeholder's view had drifted from the canonical roster, and this release restores it: synthesis plus seven stakeholder lenses, including broker and attorney.

https://mainthread.ai/changelog/#zoning-signal-0-13-0 2026-05-09T00:00:00Z

The corpus doubles in size while the architecture deepens.

https://mainthread.ai/changelog/#zoning-signal-0-12-1 2026-05-09T00:00:00Z

Every page now ends with a row linking the five machine surfaces, such as /llms.txt and the MCP manifest, so browsing agents that fetch only linked URLs can reach them.

https://mainthread.ai/changelog/#zoning-signal-0-12-0 2026-05-09T00:00:00Z

The tool surface catches up to the content surface.

https://mainthread.ai/changelog/#zoning-signal-0-11-0 2026-05-09T00:00:00Z

The contract layer between content and AI agents catches up to the prose layer.

https://mainthread.ai/changelog/#zoning-signal-0-10-0 2026-05-08T00:00:00Z

Every named entity in the South Lake County corpus, from board members, applicants and attorneys to agencies, legislation and projects, gets its own page at /entities/{slug}, linked both ways to every meeting, brief, watch and pattern that mentions it.

https://mainthread.ai/changelog/#zoning-signal-0-9-0 2026-05-08T00:00:00Z

Named patterns become first-class artifacts.

https://mainthread.ai/changelog/#zoning-signal-0-8-0 2026-05-08T00:00:00Z

The Watch now closes its loop.

https://mainthread.ai/changelog/#zoning-signal-0-7-0 2026-05-08T00:00:00Z

The observatory now publishes what it is tracking.

https://mainthread.ai/changelog/#zoning-signal-0-6-0 2026-05-07T00:00:00Z

The full standardized meeting record from January 2024 through May 2026 is now readable: 92 meetings of the Clermont, Leesburg, Minneola and Groveland planning and zoning boards, each at its own URL with its items, a public-comment summary, key signals and a source trail.

https://mainthread.ai/changelog/#zoning-signal-0-5-0 2026-04-28T00:00:00Z

Subscriptions now drive a personalized feed.

https://mainthread.ai/changelog/#zoning-signal-0-4-0 2026-04-28T00:00:00Z

The observatory now recognizes returning readers.

https://mainthread.ai/changelog/#zoning-signal-0-3-0 2026-04-28T00:00:00Z

Two new ways into the civic intelligence graph: semantic search for readers and MCP discovery for AI agents.

https://mainthread.ai/changelog/#zoning-signal-0-2-0 2026-04-28T00:00:00Z

Four index pages and the core navigation system that structures how the observatory is traversed.

https://mainthread.ai/changelog/#zoning-signal-0-1-5 2026-04-28T00:00:00Z

The launch surface shipped with one place dossier and one corridor.

https://mainthread.ai/changelog/#wordship-salvo-playtest-fixes 2026-04-27T00:00:00Z

New: Hits, sinks and a finished fleet each have their own motion, which stays still when your device asks for reduced motion. New: Share a Salvo result as text or an image, with a link that invites friends to play too. Improved: Daily and Practice Salvo play the new hunt, then deduce, and a found ship shows its letters on the board. Improved: Daily puzzles use familiar, well-known words. Improved: Tapping a match invite joins you to it, and the defender sees their own word while their friend guesses. Fixed: Ships place with one tap, practice words are always available, and the challenge timer no longer runs out right away.

https://mainthread.ai/changelog/#wordship-salvo-v2-hunt-then-deduce 2026-04-27T00:00:00Z

New: Friend matches play Salvo in three phases: hunt ships by coordinate, solve each located ship's word in a timed six-guess challenge, and score by how close you came. Improved: Every fleet word has five letters, the same as Deduce. Improved: Phones and desktops each get a layout made for them. New: Share a finished multi-round Deduce match as text or an image. Fixed: Watching a friend's Deduce round shows every letter as it is typed.

https://mainthread.ai/changelog/#wordship-stats-streaks-platform-reframe 2026-04-27T00:00:00Z

New: Daily Deduce tracks your current and longest streak, and your profile shows your full record. New: See your head-to-head record with any friend. New: The board-and-word game has its own name and home: Salvo. Improved: Daily puzzles are published a week ahead, so today's is always ready. Fixed: Pages no longer scroll sideways on phones.

https://mainthread.ai/changelog/#wordship-wordship-public-launch 2026-04-27T00:00:00Z

New: Daily Deduce: one word a day, six guesses, the same word for everyone. New: Practice with random words as often as you like. New: Challenge a friend to one, three or five rounds, and watch their board live when you finish first. New: Share a result with a link that lets a friend play the same word without seeing it.

https://mainthread.ai/changelog/#zoning-signal-0-1-0 2026-04-27T00:00:00Z

The Zoning Signal Observatory is live.