What's changed in Masterji

The changelog you are reading this in was only ever visible inside the app, behind the panel in the corner. It now also lives at /changelog/, open to anyone with the link and no sign-in — every entry, dated, oldest to newest, including the ones from before this list was kept as files. Nothing about the in-app panel changed. Two things are better on the page: entries keep the paragraph breaks they were written with rather than being flattened into one, and the footer of the landing page finally points at both the changelog and the writing, which until now were reachable only if you knew the URL.

Every page that introduces Masterji led with the word coach — the tab title, the share card, the first line of the README, the headline on the landing page. The rhythm the product actually runs on, a task declared every morning and proof filed every evening, was in that paragraph too, third, and never the part in bold. So the sentence most people read described a coach with opinions and never mentioned the loop, and more than one reader came away describing a knowledgeable to-do list. The order is now the other way round: the daily rhythm first, the gate immediately after it, because a rhythm with nothing checking it is a habit tracker. Nothing about your goal, your phases, what counts as proof, or what Masterji will refuse has changed — this is the same product, described in the order it works in.

Masterji had two pages a stranger could find: the landing page and the walk-through. There is now a third, at /blog/, and two essays on it — one on why planning is the only part of building that cannot tell you no, and one answering how many customer interviews you actually owe before you start building, which is three, with three different people. Both are about the gate rather than about the method: the method is borrowed and credited, and there are better books on it than anything here would be. Nothing about your goal, your phases or your evenings changed.

The four-step bar at the top of the tour drew its unfinished steps in a colour meant for a raised surface, so on paper they came out lighter than the page while finished steps came out darker. The two ends of the meter sat on opposite sides of the page and a hairline outline hid it. Fixed: the three states now run in one direction, on either background.

The sign-in popup, the changelog, and the cold-start screen also got the enclosures the rest of the product already had — a panel now looks like a plate set into the page rather than a rectangle laid on it.

The switch between the register and the slate went out on the five screens that carry the full Masterji header, which quietly meant it was missing from the cohort board, a shared record, a shared board, the first-goal screen, the writing, and the please-wait screen you get when the server is cold. On those you got whatever your phone or laptop asked for and no way to say otherwise. It is on all of them now. Wherever you are, the word is at the top of the page.

Two of those matter more than the rest: a record or a board somebody shares with you is often the first Masterji screen you ever see, and it was the one place you could not choose how to read it.

The first essay at /blog/ described the evening a builder spent arguing with the coach instead of working, and said the fix was "still an open problem". That understated what the coach already does. Past three refusals it stops and has to answer which of two things is happening — the work isn't there, or the work is there and the two of you cannot understand each other — because those produce an identical stack of push-backs and only one is the builder's fault. That has been shipped for weeks. What genuinely is not built is ending the evening, and the essay now says which half is which. Nothing in the coach changed; the page was wrong about it.

The app was a chalkboard — dark slate, marigold chalk — for its whole life so far. It is now the other half of the same classroom, and the more literal one: a schoolteacher's register. Paper ground, ochre ink, ruled lines, and a red pen that appears in exactly one place, on the gate that refuses you.

Nothing about how the day works has changed. The same two cards, the same conversation, the same gate. What changed is what it is drawn on.

The token names did not move, so every screen followed at once rather than being repainted one at a time. The accent had to: marigold cannot hold 4.5:1 as body text on paper, and it is the colour of 40 pieces of text in this product, so it is ochre now.

Is it free, what happens if you miss a day, can you run two goals at once, what a cohort coordinator can actually see, and what happens when you disagree with the gate. Fifteen answers now live at /faq/, each one addressable at its own link so you can send somebody straight to it. Two of them quote the product rather than describing it — the proof bar for each phase, and the cohort board's columns — so they cannot quietly go stale. One question is deliberately not there yet: what happens to your conversations. That one needs a privacy page to be answered properly, and it does not have one.

Every button and chip in the app changed state the instant a pointer touched it — grey to marigold with nothing in between, which reads as the page redrawing rather than as the control noticing you. They now settle over about a seventh of a second, all of them at the same rate.

The bigger gap was the press. Nothing acknowledged being pressed at all, which matters most on the control where the wait is longest: filing the evening's proof on a phone, where the tap and the server answering are not the same moment. Buttons now travel a pixel under your finger.

Two things for anyone who does not use a mouse. A keyboard now reaches a "skip to content" link before the header, so the pane switcher, the changelog and sign-out are no longer between you and the day's controls on every page. And if your device is set to reduce motion, the app now honours it — the gate bar fills without animating, the tour jumps between slides instead of gliding, and nothing else moves either.

Two screens you were never meant to see, and until now both of them were the framework's rather than ours: a mistyped or out-of-date address gave you black text on a white page, and a screen that failed part way through drawing gave you one line about a client-side exception. Neither had a link on it, so the only way out of either was the back button.

Both are Masterji's now. A dead address says so plainly and puts the home page and the tour in front of you, and it names where a shared record or cohort board actually lives in case that is the link you were holding. A screen that breaks says we broke it, offers to draw it again without reloading the app, and tells you that nothing you had already filed is at risk — a proof, a check-in and a declared task are written on the server the moment they are accepted.

There is a third piece at /blog/, on the pile of half-finished projects everybody has. It is about the difference the record already draws between an idea that died after real contact and one dropped before anyone outside your head got a vote — INVALIDATED against UNTESTED, two contact proofs apart. The essay says out loud what the archive has always done quietly: a closed idea with conversations behind it is filed next to the wins, and a member between ideas shows on the cohort board dimmed rather than struck through. Nothing about closing a goal changed; it was never blocked and still isn't.

Masterji has a second ground. The word next to the wordmark switches between the register he marks in and the slate he works at, and it remembers which one you picked; leave it alone and he follows whatever your phone or laptop already asks for. The slate is not a new look — it is the one this app opened with, and marigold comes back with it, because the reason marigold left was that it could not be read on paper. On the slate it can.

The paper got a darker hand at the same time, and that is the half you will notice without pressing anything. Most of the small print in this product — the day counts, the phase line, the note under the composer, the timestamps — was drawn in one grey, and that grey was set just barely light enough to pass the accessibility floor while being used at twelve and a half pixels. It is darker now, along with the ochre, the sage and the red, and every one of them clears the stricter bar rather than the minimum one. The boxes you type into got a firmer edge for the same reason: theirs was faint enough that the field and the card behind it were nearly the same colour.

Nothing moved, nothing was renamed, and no screen changed shape.

The cohort board has existed for a while and nothing public explained it. There is now a page at /for-coordinators/ written for whoever runs an E-Cell, an incubator batch or a college programme: what the board shows, which is four server counts and nothing anybody typed, and why counts rather than a reporting template. It is honest about the two things that are not built — cohorts are still set up by hand, and there are no logos to show. Nothing about the board, the join code or the watch link changed.

Masterji's repository was public, and two places invited you to go and read it: the landing page linked the gate conditions in gates.py, and the tour linked the playbook corpus. The repository is private from today, so both links would have answered 404 — a proof link that fails argues against the sentence it sits in. They are gone, and the claim they backed has changed with them: the gate conditions and the playbooks are shared on request rather than published. What has not changed is what those files say or what the coach does with them. The method is still borrowed, still credited by name, still distilled rather than reproduced, and every refusal still traces to a condition and a test.

Masterji lived at masterji.mscsoftwares.in, a subdomain of a portfolio site, which said nothing about what it is. It now answers at helpindiebuild.com, which does. Nothing about your account, your goal or your streak moved — it is the same app at a name worth typing. The old address still works and sends you here, so a bookmark or a link you shared earlier is not broken.

Masterji moved to helpindiebuild.com and nothing told search engines about it: robots.txt and sitemap.xml both answered 404, no page said which of the two addresses was the real one, and every link anyone shared rendered as a grey box with no picture. All three are fixed. There is now a share card with the म on it, the pages say what they are in the vocabulary assistants read, and the cold-start screen no longer sits where a search result for the product's own name should be. Nothing about the app changed — the same goal, the same gates, the same evenings.

If your E-Cell gave you a join code, there is now a "Cohort" link in the top corner to enter it — and once you have joined, the same link is how you get back to the board. It used to live at an address you had to remember or bookmark; now it sits next to "How it works" and "What's new", where the rest of your account lives.

The "Cohort" link now sits in the footer of the opening screen as well as the dashboard. If your E-Cell handed you a join code before you had picked an idea, you no longer have to commit to one first to find the place to enter it — it is reachable from the moment you land.

The model behind Masterji's replies moved up a generation. What he can do has not changed — he still cannot advance your phase or bank a proof on his own; the database decides that, and every verdict still has a deterministic fallback if the provider is down. You may notice his wording reads a little differently.

The person running a cohort — the E-Cell or IIC faculty member — can now watch its board from a read-only link, without enrolling as a builder in their own cohort. The link shows exactly what a member sees: first names and the counts the gate accepted, never an email and never a word anybody wrote. It has no button that changes anything, because the page it opens has no way to write. Staff mint the link in the admin and can revoke it — the old URL then stops working — the same way a join code is rotated.

The guided tour still showed the cards first and the chat underneath, which stopped being the product some releases ago: the draft is written in the conversation, the presses stand under the words they came from, and on a phone the app opens on the chat. Every slide now reads in that order — his message before the goal card, the conversation before Today, the inline proof card before the evening form. The gate keeps its own slide, and keeps its cards. One thing on it was also wrong: the 2 of 3 refusal printed the errand you get at 0 of 3, telling a builder with two conversations banked to go and have their first. It now says what the server says at that count — the third one, and the ask that costs them something.

On a wide screen, the column that holds your commit box and the conversation beside it are now two panels in the same frame, standing the same height — instead of a card next to a loose stack that stopped wherever it ran out. The "What IDEA will ask you for" panel has an even shape now, and its heading reads as a sentence: IDEA kept its capitals because it is the phase's name, and the rest of the line stopped shouting. Nothing moved on the phone, and nothing about what the screen does changed.

Once you're building, the day you'll launch is something you argue down from "someday" to an actual Friday in the conversation — and then you used to leave the chat, open the launch box, and pick that same day and room again by hand. Now Masterji writes down the day and the room you named, they appear above the launch box in your words, and one tap fills both controls. He only writes down a day you said or agreed to yourself: there is still no default date and no suggested one, because a day the app picked is not a day you committed to. Nothing is on the record until you press Set — the launch trail is append-only and every move on it is one you made, so a drafted day never writes a row. Set still checks everything it always did: not a day that has been, inside the next three months, and a real room off the ladder. The box works on its own exactly as before — type the day yourself, ignore the draft, or move the date later.

Talking your goal into a sharper sentence happens in the conversation, and until now the sentence then had to be typed a second time: leave the chat, find the reword control on the goal card, and write it out again from memory. Now Masterji writes down the sharper wording you arrived at, it appears under the goal in your words, and one tap opens the reword box already holding it. It is your sharpening he writes down, not his preferred phrasing of your idea — the goal is the thing every proof you file is evidence for, so the words stay yours. Nothing is renamed until you press Save wording, and the same window applies as always: the wording is yours to change until the first proof banks against it, and after that it stays as it is. Draft one before that evening and press it after, and you will get the same refusal you would have got typing it — the suggestion never outruns the record. The reword control still works on its own: type it yourself, or ignore the draft.

A phase unlocks in the conversation, and the next question — what will this one have produced? — used to be one you answered in the chat and then typed again into the box under the phase. Answer it once now: Masterji writes your line down in the turn after the unlock, it appears above the box with your words in it, and one tap puts it in the field. He asks at most once and never chases it — a phase advances on proofs whatever this says, and leaving it blank costs you nothing. Nothing is saved until you press Save, and the line the coach quotes back for the next three weeks is only ever the one you pressed. The box still works on its own: type it yourself, reword it later, or ignore the draft entirely.

At Traction you tell Masterji the number — "two people paid today" — and then scroll down and type 2 into the box under Today. He now writes it down with the rest of tonight's proof: the figure shows up under his draft, and the button that fills the proof box fills the number box with it. Only ever the number you actually said, never one he worked out or carried over from yesterday — if you did not give him a figure, the box stays empty, which is the right answer. And it stays a box: edit it, clear it, or ignore all of this and type the number yourself as before. Nothing is recorded until you press Submit proof, and a number that comes out wrong is dropped on its own — it can never cost you the proof it arrived with.

The switch in the header offered a language Masterji did not reliably speak. In the room before your goal he answered in Hinglish; in the daily chat, where you actually spend the week, he answered in English whichever side you had pressed. A control that changes nothing is worse than no control, so it has gone, and Masterji is English everywhere. Nothing about your goal, your record, your streak or the gate changes, and the header has a little more room on a phone. If Hinglish comes back it comes back properly — in both rooms, not in one.

The dashboard used to open on the Today card, with the conversation one tap away. On a phone it now opens on the conversation — since drafts complete and commit right in the chat, that is where a day usually starts. Today has not moved: same pill, one tap, and its badges still tell you when a draft, notes, or an open day is waiting there. On a computer nothing changed; both halves are on screen at once, as before.

When Masterji writes today's task or tonight's proof out of your conversation, the draft now appears in the chat with its button on it: Declare it under a drafted task, Submit proof — link and screenshot included — under tonight's. On a phone that ends the errand the draft used to create, where the thing you had just finished landed in the pane you were not looking at.

Nothing else moved. It is the same press writing the same row, so nothing is ever recorded until you make it; Today still carries the meter, the streak, the record and the refusals, and its own controls still work exactly as they did. A draft that is not finished still shows what it is missing, and filing it anyway is still your call. The card is only ever the offer that is live right now — the moment you file it, or he rewrites it, or the day turns over, it stops being a button, so nothing in yesterday's scrollback can file anything into today.

When Masterji read this morning's task and called it too vague or off-phase, the only button under him said "File tonight's proof" — a complaint you could do nothing about until tomorrow. Two things change. There is now a "reword today's task" control on the card, so the wording of today's work is yours to sharpen at any point before you file the proof. And where he objected, he now writes the sharper version of your own task underneath, one press away from the box. It stays your task and your call: take his wording, write your own, or leave it exactly as it is. Nothing is refused, and nothing is rewritten without you pressing Declare it.

The room before your goal used to stop the moment you had one candidate worth committing to — and IDEA's four questions were left for your first morning, which meant re-answering things you had already spent the conversation on. Now those four questions stand on the left of the screen from your first turn in the room, and fill in as you answer them: the problem, the one place those people already are, why you think they're there, and how you'd get one conversation this week. Masterji works through the open ones with you, and what lands carries over so your first morning starts from what you already told him. The room is twenty turns instead of fifteen, because that is a longer conversation than just choosing. It still refuses nothing: Commit works from the first turn and at nought of four exactly as it does at four of four — while the four are open the questions are simply the loudest thing on the screen.

Since 14 August, any turn where Masterji did something rather than only talking broke after he had already answered. You saw "Masterji lost the thread — try again." That covered writing tonight's proof up from your conversation, checking the gate when he thought you had earned the next phase, opening the close box, and everything the workshop does — parking a candidate, suggesting a title, totting up how much of IDEA's bar you had already reached. The plain talking turns were fine, which is what made it look random. Nothing you filed was lost and no proof, streak or phase was affected; the turn simply ended in an error instead of doing the thing. It now does the thing.

The problem statement playbook taught four parts — who, the trigger, the workaround, why the workaround fails — and then told you that a paragraph missing any of them was an idea rather than a problem statement. The gate has never asked for the trigger: IDEA counts who has the problem, what they do about it today, and why that's bad. So a builder who read the playbook properly could write everything Masterji actually asks for and still conclude they were not ready to file. The playbook now names those three as the standard. The trigger stays where it was, taught as method rather than as a requirement — the moment the problem shows up, and how often, is still what tells you whether the problem is worth acting on, and both of the accepted examples have one. Nothing about the gate changed; it was already the more honest of the two.

Masterji's field guide to over-engineering listed six smells with Kubernetes and permissions matrices at the top — things you reach for after a few years inside a software company, not things a first product dies of. The list is reordered to the order a first project actually meets them: tool polishing and the rewrite at the top, infrastructure at the bottom. Three smells that were missing are in it now — choosing a framework you have never used as the project, rebuilding the UI because it looks amateur next to products with design teams, and building an app where a link you can paste into WhatsApp would do. The counter-question and the section on the second idea are unchanged.

Being turned away at the LAUNCH gate came with one line of coaching, and its arithmetic did not work. It said three rungs was the count and that one of the three had to be a stranger acting — while the launch playbook you were taught from lays out a ladder of four rungs, with the ask as the fourth. Three of four, one of which is the fourth. The gate has never counted rungs: it counts three accepted proofs, one of them evidence of a stranger acting. The line now says that, and keeps the ladder as the ladder — four rungs, one a night, climbed until you have the three. Nothing about what LAUNCH costs has changed; the sentence describing it was wrong, at the moment it was most worth trusting.

The launch playbook put the ask — charging, or a committed sign-up — "somewhere on the ladder", the only rung phrased as a matter of taste. It is the one rung LAUNCH will not let you skip: the phase does not close until a stranger has acted on the ask. The ladder now says so, and points at The First Rupee for how to make it rather than repeating the method. The list of what counts as proof now marks the same thing: a live post and a written-down rejection both still count toward the phase, but a LAUNCH made only of posts and rejections does not close. Nothing about what LAUNCH costs has moved — the playbook was silent about a requirement you were being held to.

A turn in the idea room that parked a candidate, filled in part of the first evening's bar, or put a title in the goal box used to say nothing at all — your message with a blank space under it, and a turn off the meter. Each of those now comes back with a line: how many ideas are parked and how much room is left, which parts of a first proof you have given and which are still open, and that a suggested title is yours to edit and yours to commit. It is in the conversation as well as on screen, so it is still there when you come back.

The note that reassures you your banked proofs survive a phase change did not agree with itself when there was exactly one of them — "1 proof from IDEA stay on the record." IDEA asks for one proof, so that is the sentence every builder read at their first phase advance, on the screen right after they earned it. It now reads "stays". Nothing about what is kept has changed; only the sentence that tells you.

He has written tonight's proof from your conversation for a while. The morning was the half you still typed twice: work out today's task over four turns in chat, close it, scroll to Today, type it again from memory. Now when you say what you're doing today, it's in the box waiting for you — in your words, as he heard them. Nothing is declared until you press the button, exactly as before: he writes it down, you commit to it. If the wording isn't right, edit it first.

Masterji's playbooks credit the source they distil, by name, in the file's first lines — that is the rule the coach's authority rests on. Fourteen of the sixteen did it. Two did not, and the check that was supposed to catch them was reading a hand-kept list neither could ever be on. Over-Engineering Smells now credits Dan McKinley's Choose Boring Technology and Joel Spolsky's Things You Should Never Do; Launching Without Permission credits Paul Graham's Do Things That Don't Scale, Croll and Yoskovitz's Lean Analytics, and Pieter Levels' MAKE. The check reads the folder now, so the next uncredited file fails instead of passing quietly. The advice in both files is unchanged, with one exception: the smells were headed "ranked by how often they kill first products", a frequency claim with nothing measured under it, and that claim is gone. The curation rule also gains a third state, because a method that really is Masterji's own used to be indistinguishable from one that quietly borrowed — a playbook may now say so outright, provided it names the gate, rule or refusal it is the reasoning for.

Signing in with Google now has to end in the same browser it began in, not merely in one holding a link Masterji handed out. Nothing changes for you if you sign in normally. If a sign-in takes longer than ten minutes at Google's account picker, or your cookies go away in the middle of it, you land back on the front page with the sign-in panel open and a note saying so — one more press and you are in.

Two coach replies got more careful. Asking whether your goal is too big now gets an answer about size: where to cut, that doing the small bar in front of you is what shows where it is actually too wide, and that the reword control on your goal card can take a narrower wording while nothing is banked yet. What it no longer gets is an unasked-for reminder that the close box exists — Masterji only offers to open that box once you say you want out. Closing itself has not moved: it is still free, still yours, still one honest sentence away. And in IDEA, your route is now judged on one thing — whether it names a real room a stranger could find — never on which app or platform the room lives in. The same answer gets the same push-back wherever you say it lives, and if the kind of room you name does not exist on the platform you name, Masterji will say that plainly instead of letting the platform's name do the grading. No gate changed either way.

Masterji has always said you are allowed to spend a day on something this phase is not for — he will tell you it is a detour, and the phase gate is what makes it cost something. He was not keeping that promise. On an off-phase day the morning asked you to bring the phase's work tonight rather than the task you actually declared, and the evening then judged you against that ask — so you could declare something, do it, file honest proof of it, and be pushed back for not bringing something you never said you would. That is fixed in both rooms. The morning now asks for proof of the task you declared, whatever phase it belongs to; and if the ask ever comes out wrong again, the evening judges what you declared and did rather than what the ask asked for.

VALIDATION costs three conversations, and until now all three were asked for identically — the same line under your goal, the same words when the gate turned you away. But the first conversation is a door problem, the second is a who problem, and the third is an ask problem, and the playbooks have taught all three separately from the beginning. Nothing reached for the right one on the right evening. Now it does. With nothing banked, the whole job is getting into one room, and the refusal carries the arithmetic that stops people quitting here: five messages sent to get one back, so an unanswered message is the rate and not a verdict. With one person on the record, you are told plainly that this phase counts people rather than evenings — the same fact the gate used to deliver as a rejection, arriving early enough to act on — and pointed at the shortest route to somebody with no reason to be kind to you, which is asking the first person who they know who has it worse. With two, it asks for the commitment: an hour, an introduction, a look at their books, money, and what happened when you asked. That was always part of what a conversation note has to contain, and nothing ever escalated to it. What did not move is the bar. The same proof counts on the third evening as on the first, the count is still three, the gate refuses exactly what it refused before, and your accepted nights stay accepted. The only thing that changed is which part of the work the product puts in front of you tonight.

The header carried the streak badge and the lifetime-days count, and on a phone they cost it a whole third row — 150px of a 640px screen, on every screen, forever. Worse, that row only appeared once you had banked something: the more days you kept the promise, the more chrome you got, and the account with nothing on the record was the one that got the compact header.

Both numbers are now on the goal card, in one line with the days-in-phase count they are the same kind of fact as: how long this phase has been open, how long your current run is, how many days are behind you altogether. They sit against the stepper they describe, and the header is two rows for everyone again.

Nothing about when they appear has changed. No run yet means no badge, a broken run still shows the best one you had, and the lifetime count still only appears when it has something to say that the streak doesn't. On the phone, the chat pane gets the header's third row back.

Until now the only thing your goal held was its title — a line of text. The evening you clear IDEA, the problem you wrote up now sits on the goal card under that title, in your own words, and stays there.

Masterji gets it too, on every turn, separately from your record. That matters later rather than now: your record carries your ten most recent proofs, so on a long goal the evening that said what you were building was the first thing to fall out of what he could see. It no longer does.

Nothing about the gate moved — IDEA still owes its one proof and it is still judged the same way. You can edit the idea by hand until your first proof is accepted, the same window in which you can still reword the title.

If your E-Cell runs a cohort, they can give you a join code, and Your cohort becomes a page: everyone who joined, ranked on what the server counted. Proofs that were accepted, how many of those needed a real person, your current streak, the phase you reached. Nothing on that board is self-reported — there is no field on it anybody can write in. It does not show your idea, your brief, a single word of any proof, or your email; a coordinator sees the counts and nothing you typed. Joining is the whole of the agreement, and until you join a cohort you are invisible to every part of this: not listed, not ranked, not counted in a total. Leaving takes your row off the board the same evening and changes nothing else — every goal, every morning you declared and evening you proved stays exactly as it is, because your record does not know a cohort exists. And nobody running one can bank or unbank anything. There is no button for it, and the page they read has no way to write at all. That is the point: a leaderboard of decks measures who presents best, and this one can only ever show what got through the gate.

The workshop used to hand your goal a title and forget the rest. Now, when Masterji names the idea the tiebreak landed on, whatever the room actually established about it — the problem, where those people already are, why you think so, how you'd get one conversation — comes across with it, and your first morning starts from what you already said instead of asking for it again. What the room never got to is named too, because that is what your first evening's proof is for. Nothing about the bar moves: IDEA still owes one proof, still filed by you and still judged. The other one-liners you parked are kept as well, and show up under the closed idea's record — so if this one ends, the thinking that produced it is still there.

The backend still sleeps when nobody has needed it for a while — that part is the price of free hosting and it has not changed. What has changed is how long it takes to come back. It used to be about two minutes; it is now about half a minute, because it runs somewhere that gives it a whole processor to start on rather than a tenth of one, and no longer stops to check its own database on the way up.

On your first morning the goal card carried exactly one button, and it was Request phase advance at 0/1 under an empty bar. The loudest control on the main screen was the one whose only possible answer was a refusal — and a refusal with nothing to add, since the bar above already said 0/1 and the phase hint already said what the work is. It is now a quiet link, the same weight as close this goal, and it becomes a button again the moment anything is banked. It still answers when pressed, and the answer at 2/3 — which names which piece is missing — is worth a button, so it gets one.

The second is the retire box. I achieved it was the filled button and first in the row, on a goal whose own counter said the phase was unfinished. Below the finish line neither exit is the default now; both are outlined, and I achieved it stays first because it is not the shameful option. When the record actually holds finish-line proof, the card one screen up still lights Earned. Proof is on the record. and a real Claim the win — that is where achievement is the expected move, and it is the database saying so.

Both exits still work from every phase, and the verdict on your record is still computed from the proofs you earned, never from which button you pressed.

Every other phase gives you something to point at that nobody's opinion produced — a count, a bar, a door that opens. TRACTION has none of that by design: it is the end of the ladder and there is nothing left to unlock, which left the last mile with no shape to it. So you can now name the one number that means somebody got the value — paid deposits, orders through the form — and log it on the evenings you read it. Masterji is handed the last two readings and can ask you about the direction; the record keeps the whole line. Nothing counts it. There is no target, no streak attached to it, and no gate anywhere in the product that reads it, so a number that falls or sits flat for a week refuses you nothing and costs you no proof. It is a scoreboard, not a bar. You can change which number you watch, and the record will say so — the readings you already took keep the name you took them under, because three evenings of counting deposits do not become three evenings of counting signups just because you changed your mind on Friday. That trail is the only consequence there is, and it is the thing that makes naming a number mean anything.

TRACTION asks for one person who returned on their own, or who paid — and the only playbook loaded there taught recruiting users one at a time, which is acquisition. Masterji now also carries Coming Back, distilled from Andrew Chen's writing on retention and the leaky bucket: why somebody returns at all (it is caused by what happened last time, not by what you shipped this week), how to tell a real return from a second first-time visitor when you have ten users and no analytics, and the difference between coming back and being reminded — which look the same in a count and mean opposite things, so if you nudged them, the record should say so. It also names what not to reach for. Notifications, streaks, badges and points manufacture the look of a return without the reason for one, and at this size they hide the only signal you have. The bar has not moved: the same person, back a second time, or a payment.

Masterji has a new playbook in LAUNCH on writing the post itself, distilled from Harry Dry's Marketing Examples: open with the words the people you interviewed actually used, write for the one room you're posting into rather than "the public", pick the concrete thing over the clever line, and end with an ask that costs the reader something small. LAUNCH told you where to post and never how to write it, and that is the step where the week usually goes quiet. Two existing playbooks also gained a section. The cadence playbook now covers the week that isn't week-shaped — exam week, placement season — where the move is to shrink the slice to thirty minutes rather than pause the loop, and your streak counts that Tuesday exactly the same as a full Sunday, because it has always counted declaration and proof and never volume. The over-engineering guide now covers the second idea, the better one that turns up mid-BUILD on a hard day: write it down in one line, keep going, and if you are seriously considering the switch, name what would have to be true about the idea you're on before you look at whether it is. Nothing about the gates moved.

Two phases were refusing you on conditions the coach had nothing to say about. BUILD will not let you leave without evidence a real user touched the thing, and all three of its playbooks were about building it — what to cut, what not to over-engineer, how to ship daily. Not one was about getting a person in front of it. VALIDATION started counting distinct people rather than evenings, which quietly made who your first three are the whole question, and nothing covered the case every first-time builder actually starts with: the friend who wants you to succeed and will say yes to anything. So there are three new ones.

The First Touch (BUILD) — recruiting your first user by hand out of the VALIDATION notes you already have, the smallest thing you can put in front of them tonight, and how to watch instead of demo. It also draws the line the gate draws: "I showed her and she said it looks useful" is a demo you gave; what she did when she was holding the phone is a touch.

Talking to People You Know (VALIDATION) — why a friendly yes teaches you nothing, what a friend is genuinely good for (the introduction, above all), the framing that gets somebody talking about their own last Tuesday instead of reviewing your idea, and when to stop mining your own corridor.

Reading the Nos (VALIDATION) — what a week of notes means together. Sorting by what people gave up rather than how keen they sounded, the bar for calling something a pattern, changing who you are building for before changing what you are building, and the honest kill: an idea taken to real people who said no is recorded as invalidated, which is validation working rather than failing.

Nothing about any gate moved. The same proofs cost the same, the counts are unchanged, and no phase got easier or harder to leave — what changed is that the two hardest bars now have something behind them. As always, you can read every word the coach is working from: the whole corpus is in the public repository.

Until now there was no way to leave. Delete account now sits in the account row beside sign out, and it opens a panel rather than firing: it says exactly what goes — every goal, every morning you declared and evening you proved, the whole conversation, the account itself — and offers each goal's record as a file to keep first, because the record is the thing this product argues is worth having and you should not lose it on the way out. Then two presses, and it is done: your evenings stop existing for the app, your email is released so you could start fresh with the same Google account if you ever wanted to, and the session ends. Nothing about it is recoverable by signing in again. It is your diary, and a product that keeps one should always have had a door.

Three things the room before the goal got wrong, all of them places where it disagreed with the chat you get after committing. On a phone, Return sent your turn half-written — and a workshop turn is one of fifteen, so that keystroke cost you a turn you could not get back. It now does what it does everywhere else on a phone: adds a line, and Send sends. The conversation also opened on its oldest three turns, so coming back to the tab showed you the first thing you said rather than the last thing he said — with his tiebreak, the whole point of the room, three screens down inside a box you had to know to scroll. It now opens on the newest turn. And the sentence that closes the room, when the fifteen are spent, was missing a space in its first two words.

The room before the goal answered "I don't have an idea yet". It never answered "I have one and I no longer believe in it", which is the same sentence four days later — and until now the only way to get a room for it was to close the goal first. There is a second door beside "close this goal" now: five turns, once per goal, with Masterji on your side of the table and the daily loop left outside. He can see the goal, the phase and everything you have already banked on it, and there are three things he will put on the table when the conversation has got somewhere: finish the bar in front of you, sharpen the wording, or close it today and pick again. Nothing in the room banks, nothing advances, no proof is drafted and no streak is spent — reconsidering an idea should not cost you the record of it.

The room before the goal was drawn in a different language from the chat you get after committing: Masterji had no face in it, his replies were boxed and yours were filled — the opposite of everywhere else — the text was smaller, and the box you typed into was two fixed rows that could not grow. All of that is now the same as the chat's, because it is the same conversation. On a wide screen the room sits beside the commit box instead of in a 320px window under it, with the parked candidates and the count of what you could already write next to the box they fill. On a phone the commit box stays pinned to the top, so every "put it in the box above" is about something you can actually see. Nothing about the room's terms moved: fifteen turns, three candidates, Send is still the quiet button and Commit is still the only loud one.

The room before the goal now keeps a count. As Masterji digs into the candidate you are circling — who has the problem, which room those people are in, why you think they are there, how you would get one conversation this week — the workshop shows how many of those four you could already write down tonight, and names the ones still open. Those four are exactly what IDEA asks for on the other side of the commit, so the number is a forecast of what committing would cost, not a score. It gates nothing: Commit works at nought of four the same as it does at four, nothing is banked, and IDEA's first proof is still yours to file and still judged after you are in.

The unread dot beside What's new was lit for every brand-new browser — so the first screen after signing up, whose entire job is getting one idea into one box, carried a marker pointing away from it. What it was offering was this changelog: entries about changes that shipped before you had an account, none of which you can have missed. A browser opening Masterji for the first time is now marked as caught up, and the dot means what it always claimed to — that something has shipped since the last time you looked.

The loop is daily and the drift is weekly, so a run of honest days that moved the gate by nothing looked exactly like a run that moved it. On the first morning of a new week Masterji now reads the last one back: days complete out of seven, what was accepted toward the phase, who you spoke to, and any phase you opened. Nothing is asked for — every number is counted from the check-ins you already filed — and the same figures go to Masterji, so Monday's conversation starts knowing what the week held.

A week where nothing was declared writes nothing at all.

The Monday digest read back exactly one week: the one immediately before the visit that triggered it. So if you worked a full week, disappeared for a fortnight, and came back, the week it looked at was the empty one you had just been away for — nothing was written, and the week you were proud of could never be reached again. When the week just gone is blank, the digest now reaches back to the last week that actually had check-ins in it, and says which one it is: Picking up from the week of 3 Aug — 6 of 7 days complete, 2 accepted toward the phase. Naming the date is the point, so it can never be read as a claim about the fortnight you were gone. Masterji is handed the same week, so he is not talking about a different one than the message above him. It is still one message and still once a week — a digest for every week you missed would meet you on the morning you came back with a stack of them, most saying nothing was accepted, and that is not a welcome. It reaches back a month at most; past that you are starting again rather than picking up, and he handles that in conversation instead.

There is a ceiling on how many turns you can take in an hour, and when you reached it the app answered "Masterji said no (429)." — a status code, in the red it uses when something has broken. Masterji had already written the real sentence for that moment and you never saw it. You do now: too many at once, come back to it in a bit. Nothing about the ceiling itself has moved, and a ceiling has never been able to touch the record — it refuses one request, so your goal, your streak and every proof you have banked are exactly where you left them.

The room before the goal parks at most three one-liners, and that ceiling is the point of it — once the pile is full the only move left is choosing. If Masterji parked the same idea twice, though, it took two of those three and showed up as two identical cards, so you reached the forced choice with two ideas and a copy of one of them. A one-liner you have already parked is now kept once, whatever the capitals and spacing. Nothing else about the pile changes: three is still the limit, a genuinely different idea is still its own card even if it reads a lot like one above it, and tapping a card still only fills the box.

Masterji's playbooks are public, and the tour told you they were about ten minutes of reading. They had grown to about forty-nine, and four places in this repository were still quoting the old number.

The claim now says the thing that was always the useful one: your phase's playbooks are about a quarter of an hour, and that is the whole of what Masterji is handed when he judges tonight's proof. The full corpus, if you want to audit all of it rather than the shelf you are standing at, is under an hour.

Nothing about the playbooks themselves changed, and none of them was removed — only the sentence describing how long they take. A test now measures the folder and fails if either number stops being true, so the next playbook that lands cannot quietly leave the promise behind.

Unlocking a phase used to be five words — "Phase unlocked: VALIDATION → BUILD." Signing up, by contrast, gets a full briefing: what IDEA is for, what to write, and what waits until later. That was the wrong way round, because the phases you earn are the bigger jump. BUILD arrives wanting two nights of evidence with at least one of them a real person touching the thing. LAUNCH wants three nights in front of strangers, one of them somebody actually acting. TRACTION wants one stranger who came back on their own, or who paid. Every unlock now carries a short brief from Masterji saying so — the count that phase will ask you for, and the mistake it is most often lost to — distilled from that phase's own playbooks, which are all in the public repository as they always have been. Nothing about any gate moved: the same proofs cost the same, and a refusal says exactly what it said before. Being told precisely what is still missing is the coaching at that moment, and it is not the moment for a briefing about a phase you have not opened yet.

The screen where you name a goal opened with a 90-word paragraph doing five jobs at once, and the box you came for sat underneath it — a quarter of a phone screen spent on terms, in front of somebody who had just signed in and typed nothing. Two sentences stay above the box now, and they are the two the commit is actually a commitment to: pick the problem you'll test rather than the idea you'll finish, and it costs one task each morning and proof of it each evening. The rest — that he coaches one thing at a time, that the first ask is one evening at your desk, that you can close it whenever you like and an idea that dies in front of real people still reads as tested — is a tap away under What you're agreeing to, which is where you'd go looking for it if you were hesitating. Nothing was deleted and nothing about the goal changed; the words that answer a question just stopped arriving as terms.

At 3/3 the meter was the same marigold it was at 1/3. The one thing on the card you can read at a glance said nothing at the best moment in the product, while the line directly under it had been spending the good colour on that exact state all along. The bar now goes green when the phase is met — and only then: a full count with a kind still owed is not a met bar, and the meter says so by staying marigold while the card names what's left.

The other half is what happens one press later. Opening the next phase turned a full bar and Earned into 0/2 proofs toward LAUNCH over an empty one, with the advance button back. Three real conversations became a new debt, and nothing on the screen said the three were still on the record. They are, and the card now says so: 3 proofs from VALIDATION stay on the record, under the new phase's empty bar, until that phase has something of its own to count.

No gate moved. The numbers are the same ones the server was already sending, and the bar still refuses exactly what it refused yesterday.

Opening Masterji when you were already signed in used to mean waiting on a blank page: the browser had to fetch the whole app and ask the server who you were before it could draw a single thing. Now the page arrives with your dashboard's frame already on it — the header, both columns, the cards — and fills in as your goal and today's task load. Nothing jumps around while that happens, which it used to. If you are not signed in, the landing page is unchanged, and it now downloads noticeably less to show you.

Closing a goal has always been free, and Masterji has always been allowed to say so. What he could not do was close one — and nothing stopped him saying he had. Tell him you were done and he could answer "Done. This goal is closed." over a goal that was still running, with nothing written and nothing on the record. That sentence is gone: he cannot close a goal, he is told so, and he will not claim it. What he can do now is open the door for you. Say you want out and the close box opens on your goal card, scrolled to and waiting — and it still asks you for the two things that were always yours: one honest sentence about what happened, which goes on the record, and which way it ended. Nothing closes until you write that sentence and press an exit yourself. He only ever opens it when you have asked to get out; a bad week is not an answer to a question you did not ask.

Every phase told you the same thing it tells everybody — "smallest thing a real user can touch this week" — and nothing ever learned which thing you decided that was. When a phase unlocks you can now leave one line under it: what this phase will have produced. Masterji reads it from then on, so he can tell work that is going where you said from work that has drifted, and the morning's reading of your task gets it as context too. It is a note, not a promise: it changes no count, the gate has never read it and does not start, you can reword it while the phase is open, and skipping it entirely leaves everything working exactly as before. The next phase gets its own line, and this one stays where you left it.

The box that asks what happened now has a third button beside "I achieved it" and "I'm dropping it". Use it when the idea died and the problem did not, which is the commonest way this actually goes. It closes the goal exactly as dropping it would — the same record, the same honest verdict, and a pivot with nobody spoken to still reads untested — but the next goal you commit is linked to this one, and Masterji opens it knowing what those weeks found. He will not send you back to re-interview somebody who already answered.

What carries over is what you learned, never what you earned. The new goal starts at IDEA with nothing banked and owes its first proof like any other: writing the new problem statement is one evening at your desk, and it is the pivot made concrete. The screen tells you what it is carrying, and one press drops it if you have changed your mind.

On a 360px-wide phone the four buttons at the top did not fit on one line, so sign out fell onto a third — and the header took 150px of a 640px screen before you had read a word. The EN / हिं switch now sits up beside the Masterji wordmark, where there was empty space all along. The header is 106px, the conversation gets 274px instead of 230px, and nothing moved on a laptop.

Five things about using this on the device you actually use it on.

Today is the first card. The goal card carries your title, the stepper, how long you have been in this phase, the phase hint and the bar — and it sat above the day's task, so on a phone "Declare it" was off the bottom of the screen. On the pane called Today, in a product that asks for two minutes a day. The two minutes are now the first thing you see, and nothing about them needs scrolling to. On a laptop nothing moved: both columns are on screen together there.

The controls are thumb-sized. "sign out" was the only thing on this screen big enough to be sure of hitting, and it got there because the cost of missing it is your whole session. Everything the daily loop actually runs on was smaller — "Declare it", "Submit proof", "Request phase advance", "Commit", and "reword" at sixteen pixels tall. The prices are different but they are all real: a miss on "Declare it" is the morning, a miss on "Submit proof" at eleven at night is the evening, and a miss on "reword" swaps your goal for an edit box, which reads as the app having eaten it. All of them now clear the same 44px, and none of them moved or grew wider to get there.

A finished phase looks like something you can open. Tapping IDEA or VALIDATION on the stepper shows you every day you spent in it. On a phone nothing said so — the pointer, the hover highlight and the tooltip that announce it are all things only a mouse has, so the two chips you can open looked exactly like the three you cannot. Completed phases now carry a small chevron, on any screen without a mouse.

The goal box keeps its own example. "e.g. Tiffin-delivery app for my college" was being cut off mid-word on a phone, and the half that went missing is the half that tells you how big the answer should be — one campus, not a market. On a narrow screen the box and "Commit" now stack instead of sharing a row, so the example arrives whole and "Commit" gets the full width.

One less line in the way of the conversation. The caption under "Coach me / Think with me" named whichever of the two was already lit. On a phone that is a seventh of the visible conversation spent saying something the lit button already says, so on the narrowest screens it is gone. The switch is its own label, and what the other mode is for is in the tour, where there is room to say it properly. The note under the box stays exactly where it is: it is the only place that tells you the box you type in is not the box that counts.

Declaring now takes an optional hour: "by 21:00", beside the Declare it button. It goes on the record with the task, and Masterji reads it back — so he can hold you to a time you picked yourself instead of a vague "tonight". Nothing else changes. It is not a deadline: a proof filed at 23:40 against a named 21:00 counts exactly as much, your streak is untouched, and no phase gate looks at it. Leave it on "any time tonight" and your day works precisely as it did before.

From BUILD onward you can put a date on the card: the day it goes in front of people, and which room from the launch ladder it goes into. Masterji can see it from then on, so "nine days out" is something he knows rather than something you have to keep saying. Moving it is allowed and always will be — but it is never quietly overwritten. Each change is a new entry, so the card says "moved 1×" and the record keeps the day you first named. That trail is the whole of what a missed date costs: no proof is refused, no streak breaks, no phase is held back. Nothing in the gate has ever read this and nothing does now.

Masterji can now be added to your phone's home screen, where it opens as its own app — no address bar, no hunting for the tab you left open on Tuesday. On Android the browser menu offers Install; on an iPhone it is Share → Add to Home Screen. The icon is the same म that stands in for Masterji on every message in the chat. Nothing else changes: it is the same app, the same account and the same record, one tap closer. Notifications are not part of this — the app still only speaks when you open it.

Masterji has always spoken Hinglish in the workshop — the room before the goal reads the same language setting every other screen does. The switch that sets it only existed after you had committed a goal, so the first conversation you ever had with him was the one conversation you could not put into Hinglish. The EN / हिं switch is now on that screen too, in the row along the bottom. It changes nothing about the gate, the same as it never has.

The Delete account button that arrived on the account row this morning is gone again. Deleting your account is still something you can have done — ask, and everything goes exactly as that panel described, with nothing kept back. What changed is that it is no longer a button sitting beside sign out. Your record is still yours to take at any time: the export is on each goal, open or closed, where it has always been.

A pushed-back proof was drawn in the same red the app uses when something breaks — a failed request, a dead screen, a chat turn that never landed. It is not that. The system worked, somebody read your work, and you were asked for more. It now reads as Masterji speaking, in his own colour, and so do the earlier tries folded underneath it. Red is for when we broke, not for when he objected.

IDEA asks you for one specific place these people already are, and it will not take a channel — "Reddit" and "LinkedIn" are not rooms and do not count. Then it asks why you believe they are there. Both of those questions sit downstream of a cut most builders have never made: from "people like me" to a group with an address. The corpus had nothing on that cut. The Problem Statement teaches what a statement has to contain, Choosing the One Idea teaches how to pick between candidates, and "be specific about the who" was the whole of the advice in between — which tells you to cut without telling you where.

Narrowing the First User (IDEA) is the third playbook in the phase. Cut on circumstance rather than attribute: "small businesses" is an attribute, and if a person can match your description perfectly and still not have the problem, you cut on the wrong axis. The Thursday test — name the room, then ask whether you could be standing in it this Thursday. Could, not will: you are still not messaging anybody in IDEA, and the question is only testing whether a door exists. Why a group whose members have never met is a list rather than a segment. And the part that reads backwards: narrowing widens the route, because a "who" narrow enough to have an address answers "who else?" by itself, while a wide one has no adjacency at all.

No gate moved. The proofs cost exactly what they cost this morning, IDEA still opens on the same four parts, and nothing about the phase got harder to leave. The coach simply has something to say now when the honest answer to "where are they?" is "everywhere". As always, you can read every word he is working from — the whole corpus is in the public repository.

The whole of this product is "come back tonight", and until now the only thing holding you to that was remembering a browser tab. There is now a switch at the foot of the Today card: turn it on, and if you declared a task this morning and the proof box is still empty once your evening has started, Masterji sends one notification with your own words in it. Once a day, never twice, and only if you asked for it. Nothing about it touches the gate — it cannot bank a day, refuse one or move a phase, and a day you never declared on gets no notification at all, because there is nothing there to hold you to. Tapping it opens the box, and if the app is already open it goes to the tab you had rather than a new one, so a half-written proof survives. It works on Android and on a desktop straight away; on iPhone it needs the app added to your home screen first, and the switch says so instead of pretending.

A message from Masterji taller than the window it lands in used to arrive scrolled past its own opening — his first ever message, the one that locks your goal and briefs IDEA, began mid-sentence on a phone. Both logs, the chat and the room, now show the start of a turn too tall to fit, and still follow the words as they arrive.

Open any closed idea and there is a new button: get a link. It opens a page anyone can read without an account — the phase you reached, how many proofs you banked, how many of those came from talking to real people, days active, your longest run, and the dates each phase opened. An idea that died in front of real people reads as "tested → dead" on it, which is the only version of "it didn't work out" that says something true about you. Every number on the page came through a gate, and the page says so.

What is not on it: the sentence you wrote when you closed it, anything Masterji said, any proof text, any day's work, and your name. It is off until you press the button, and turning it off takes the link away for good — switching it back on gives you a different one, so a link you regret stops working.

The workshop offered three openers — no idea yet, three ideas and can't pick, is my idea too obvious — and was missing the one that stops more people than any of them: someone's already built this. That is not the same worry as "too obvious". Obvious is a fear that the idea isn't worth doing; this one is a belief that the question is already settled, so there is no point starting. It is also the one most often wrong, and Masterji now has a straight answer for it rather than an improvised one.

The short version of that answer: a product that exists means somebody funded a team to chase this problem, which is the best free evidence you will ever get that the problem is real. What it is not is evidence that it's solved for the people you have in mind. So the room asks you the two questions that actually settle it — what the existing thing doesn't do for your people, and whether you can name one of them using it badly, working around it, or refusing to pay. If you can, that's your idea, sharper than it was a minute ago. If you can't, that isn't a dead end either: it's the first conversation VALIDATION exists to make you have, and you can have it this week. What you won't get is a reassurance that the market is big enough for both, because nobody can act on that.

The landing page also stopped hiding the room. It described a ladder starting at IDEA and never mentioned that there's a workshop under the commit box — so a visitor whose honest position was "I don't have an idea yet" read the first rung as a wall. The page now says the room is there, what it costs (fifteen turns), and that nothing in it counts toward anything, which is the point of it.

Nothing about any gate moved. The workshop still banks nothing, still ends when its turns do, and committing is still yours to press.

Running a second cycle in one day was always allowed, and the record could not show it: two rows, one date, nothing else to tell them apart. Repeats are now numbered — 2nd, 3rd — and a day's cycles are ordered newest first like the rest of the card. The phase drill-in also takes its dates from the rows it is showing, so its heading can no longer name a range those rows fall outside, and it says how many days and cycles are under it.

A goal used to be a sentence you were stuck with. Getting the wording wrong meant closing it and starting again — which zeroes the days and the streak — so the sensible thing was to sit at the commit box polishing the phrasing before starting, which is exactly the freeze that box was making worse. Now the goal card has a reword control, and the wording is yours to sharpen right up until your first proof is banked. After that it stays as it is: those evenings were filed against this goal, and renaming it then would quietly rewrite what they were for — so the choice at that point is to keep it or close it honestly. A rewording changes nothing else. It banks nothing, moves no phase, costs no days and no streak, and it goes into the conversation with both wordings named, so the record still reads straight from the top.

Two places where what Masterji said about your progress did not match what the server was doing. Pressing Advance at the last phase said "you're at LAUNCH" to builders who were standing at TRACTION — and that sentence is kept in your transcript, so the wrong one stayed there. And in a phase that asks for a KIND of evidence as well as a count — two build evenings, one of them a real user touching it — the coach was told the gate was met as soon as the count was, while the server went on refusing it. He was agreeing with you about a door that would not open. Both now read the same thing the gate reads. Nothing about what a phase costs changed, and nothing you banked moved.

The ladder used to end at LAUNCH, which meant it ended the evening the post went out — and everything that decides whether the thing is real happens after that. There is a fifth phase now. Its bar is one stranger coming back without being asked, or a payment: who paid, how much, and what for. Repeat beats reach, so a hundred sign-ups who each opened it once is a worse week than one person who came back on Thursday. Nothing you have banked moves and no phase you have already cleared gets harder — this is somewhere to go next, not a new toll on the way. It is also where the ladder stops: growth, ads and funding are past what Masterji coaches, and they stay out.

Since VALIDATION started counting people rather than evenings, three accepted nights of notes about the same person have read as 1/3 — true, and impossible to make sense of from the outside. It looked like the gate had quietly lost two nights of work. It hadn't, and it never will: every proof you banked is still banked, still on the record, still a day on your streak. What was missing was the sentence saying which number you were looking at. The gate card now reads them both — how many proofs are in, how many different people they were with — and the refusal says the same thing without waiting for you to press the button and find out. Masterji is told both numbers too, so if you tell him you filed three he will agree with you instead of arguing with his own record. If two of your conversations came back under the same name and they were genuinely two people, write the second one down more precisely tonight and the count will agree with you.

The coach's whole knowledge base is a folder of markdown you can read in ten minutes, and until now it had nothing to say at three of the hardest moments. VALIDATION had the heaviest gate in the product — three conversations — and one playbook, which told you what to say once you were in the room and nothing about getting into it. Getting the Conversation is that missing half, distilled from Giff Constable's "Talking to Humans": ask for advice rather than pitching, what the first message has in it, why five sent to get one reply is the rate and not a verdict, and the intro ask that ends every conversation. LAUNCH said a ₹99 payment tells the truth and taught no way to get one. The First Rupee, from Rob Walling and Patrick McKenzie's pricing writing, prices against the workaround already in your notes, refuses "would you pay?" the same way the conversations playbook refuses it, and treats "too expensive" as the start of the conversation. And IDEA — where the first question anybody asks is whether the idea is any good — was answered out of the model's general knowledge rather than anything credited. Choosing the One Idea, from Paul Graham on where ideas come from and Dalton Caldwell on tarpits, is noticed-beats-invented, the campus tarpits by name, schlep blindness, and the tiebreak that decides it: whose room can you actually walk into this week. None of this moves a gate. The bars are exactly where they were.

VALIDATION asks for three real conversations, and until now the server counted evenings rather than people — so three nights of notes about the same willing friend cleared the phase whose entire purpose is finding out whether anyone ELSE has the problem. The screenshot on the landing page has always said the person already counted cannot be counted again; now the database is what keeps that promise instead of a sentence in Masterji's instructions. Each accepted conversation is filed under who it was with, and the gate counts the distinct names. Nothing you have already banked moves: a proof filed before tonight counts exactly as it did, and a proof Masterji could not put a name to counts as its own person. Talking to the same person twice is still real work, still a proof, still a day on your streak — it just isn't a second opinion.

Until now the first screen refused to talk. Chat, declaring and proving all need a goal, so the only thing on the no-goal screen was a box asking you to commit to something — and if you did not know what to put in it, Masterji had nothing to say until after you had already decided. There is a room there now. Fifteen turns with him before any goal exists: if you arrive with nothing he walks the last seven days of your own life for problems you already stood next to, and if you arrive with three ideas he answers that instead. He parks at most three candidates — the fourth is refused, because collecting ideas is the comfortable version of choosing one — and then the only work left is picking, on the one question that matters: which of these can you walk into a room and ask somebody about this week? Nothing in the workshop banks anything, nothing in it can advance a phase, and none of it is a prerequisite: the commit box is open the whole time and he never tells you that you are not ready. When the fifteen turns are gone, the only door left is Commit. That is on purpose — a room to think in with no end on it is just planning with better manners.

"Claim the win" used to appear the moment you reached LAUNCH. It was counting every proof you had ever banked, and you cannot reach LAUNCH without banking six of them — so the most prominent button on the screen offered you the exit on the first morning of the phase, before you had posted anything, which is the one piece of work LAUNCH is for. It now waits for one accepted proof filed in LAUNCH itself: the link to your post, what a stranger actually did, or a real rejection with the reason they gave. Nothing about closing a goal changed. The quiet link is where it always was, "I achieved it" is still there for a goal you finished from anywhere, and the record still counts every proof you earned in any phase when it decides whether that reads as achieved. What changed is only which moment the app calls the finish line.

Two small things on the way work actually gets filed. If you move around this app by keyboard — Tab and Escape rather than a mouse — the panels that open over the dashboard were only half built for you. A day from your record, a closed idea, a phase, this list: each one said it was the only thing on screen, and then Tab walked straight out of it into the page behind, and closing it left you nowhere in particular. Now opening one puts you in it, Tab stays inside it while it is open, and closing it puts you back on the row you opened it from. Nothing moved and nothing looks different; it just works the way it always claimed to. Separately: when your evening's proof comes back after a lost tab, the form now tells you that a screenshot cannot come back with it. Your words and your link are restored the same as before — an attachment is the one thing that cannot be kept, and reading your own paragraph exactly as you left it is a good reason to assume everything else survived too. If you had one picked, pick it again before you submit.

The entry announcing TRACTION went in with a label this app has no name for, so it rendered as a blank chip — no word, no colour — on the one list whose whole job is telling you what changed. It reads "new" now, like every other entry that added something. Nothing about the phase itself changed. Shipping alongside it: at TRACTION the coach was being handed your proof count as zero, whatever you had actually banked there, because the last phase has no gate to count toward and the code took that to mean there was nothing to count. It reads the record now. Only the coach's copy of the number was wrong — every proof you filed was banked correctly, and the finish line has always lit off the real one.

Work finished at 00:30 is last night's work, and the app used to disagree. The clock rolled over while you were typing, so the proof arrived on a day nothing had been declared on: it was refused with "No declaration this morning — proof of what, exactly?", the task you had been working on all evening was replaced by an empty morning form, and the streak broke on a day you had actually finished. It hit exactly the people who work after dinner, which is most of you. Now a proof filed in the small hours lands on the cycle that is still open from last night, and the day it completes is the day you declared it — the dashboard keeps showing that task until you file against it, and Masterji drafts toward the same one. The window closes on its own a few hours after midnight: this is for the evening that ran long, not a way to go back and fix a day you missed. Declaring a new task after midnight still opens a new day, and last night's stays exactly where it is.

Masterji could see your phase, your count and your streak, and not one date. So a fortnight in VALIDATION was described to him exactly like a second day, and a builder coming back after a silent week was described exactly like one who was here last night — which is why he sometimes greeted a comeback as though nothing had happened. He now gets two facts: how long the current phase has been open, and how long since your last complete day. The first is about the work and he may use it like anything else — three weeks and one conversation is worth saying out loud. It is also on your goal card now, under the ladder: "12 days in VALIDATION", from the day after a phase opens, and it is the same number he is holding, measured once on the server so the two can never disagree. What it is not is a deadline. This product does not set any, a phase taking a long time is not late, and that line will never change colour or start looking urgent at you, because there is nothing for it to be late for. The second is there for one reason only: so he does not talk past a week you were away. He will not open with it, ask where you were, or add up what was missed. Nothing was lost while you were gone — every banked proof, your record and your best run are where you left them, and only the current streak starts again. None of this is read by the gate: a phase costs exactly what it cost before, and a quiet fortnight banks and un-banks nothing.

Every goal now has a download: one file with the whole of it — each day's declared task, the proof you filed, what Masterji made of it, the tries he pushed back, the dates the phases opened, and how the idea ended. It is the same record the app has always shown you, in something you can attach to an E-Cell application or open in front of somebody at an interview. The tries that were refused are in there on purpose: a record that shows only the proofs that landed is a brochure, and the refusals are what make the rest of it worth reading. Nothing about it is self-reported — every number in the file was counted by the server from proofs you had to earn. Screenshots are named but not included, because the app serves those over links that expire and a file full of dead links is worse than one that says a screenshot was filed. Alongside it, the record on your dashboard stopped forgetting: it kept the most recent ninety days and "Show all" offered you exactly those ninety, so a goal that ran longer than three months lost its first weeks with nothing to say they had been there. It now says how many days exist and goes and gets the rest when you ask for them.

The note that explains the wait when the server has been asleep has been up since the admin pages got it, but the Sign in button never reached it. That button is a link straight to Google by way of the backend, so a click during a cold start left the product entirely and landed on the host's own startup log — a black screen scrolling ALLOCATING COMPUTE RESOURCES at somebody whose whole experience of Masterji so far was deciding to try it. It never said how long it would take, and it read as a site that had broken. Now the same note everything else gets stands in front of it: it says the server is waking, it waits, and it sends you on to Google the moment the backend answers. The host's log is still one click away for anyone who would rather watch it.

BUILD takes either kind of evidence on any given night — a link to something running, or proof that a real user touched it — and that was true of the phase exit too, so two evenings of deploys could open LAUNCH with nobody having opened the thing. Shipping at a people-shaped problem is the most comfortable way to avoid the people, which is the one move BUILD exists to stop. So the exit now costs two proofs AND at least one of them being a real user doing something with it: what they did, when, and whether anyone asked them to. Tonight's bar has not changed — a link that loads is still a proof and still a day. The gate card says which of the two is still missing rather than waiting for you to press the button and find out.

BUILD asks for a link to something running, and until tonight nobody ever clicked it — the screenshot you attach gets read, but a URL was taken on trust, so an address you typed from memory looked exactly like a deploy that works. Now the server opens it before Masterji reads your proof, and tells him one thing: it answered, or the host said there is nothing at that address. It is corroboration and nothing more. A link that answers does not clear the bar on its own — something running is not somebody using it, which is the whole of what BUILD is asking. And a link that doesn't answer costs you nothing by itself: he will name it and ask you for the working address, because a typo, a preview that has been torn down and a renamed project all look the same from here, and if what you wrote clears the bar then it clears it. When the check cannot be made at all — a slow host, a network that drops, a link he is not allowed to open — he is told nothing rather than something wrong. The gate is untouched: it counts accepted proofs, exactly as it did this morning.

LAUNCH was the last phase, so it never had a gate — nothing followed it to unlock. Leaving it now costs three accepted proofs, one rung of the launch ladder per evening, and at least one of the three has to be a stranger actually doing something rather than another post going out. Three posts is three evenings of real work and still nobody acting, which is the same argument BUILD's bar already makes one phase down. "Claim the win" moves with it: it lights on TRACTION's own proof now, not on the post. If you are already at LAUNCH, the button you had yesterday is a phase further up — the work behind you is untouched, and closing a goal is still never blocked from anywhere.

Two changes for one moment: deciding whether an idea deserves your weeks. The commit screen used to say "pick the one that matters" and price the daily cost — about two minutes a day — without ever saying what you were agreeing to finish, which is how a first goal starts feeling like a promise to see it through. It now says what was always true: you are picking the problem you test first, not the idea you finish; the first thing asked of you is one evening at your desk; and an idea that dies in front of real people reads as tested on your record rather than failed. Most first ideas should. Second, when you ask Masterji whether this is even the right idea — the doubt that arrives after committing, not before — he no longer puts the problem statement back in front of you and asks what you are doing tonight. He answers it: the proof in front of you is the readiness test and finishing it is the short route to knowing, closing this goal is free and the record survives it, and a goal kept out of guilt is worth less than the one you would choose now. Then he names the two doors and leaves the choice with you. None of this touches the gate — wavering banks nothing, and nothing you say in chat ever moves a phase.

Reading your task in the morning, judging your proof at night and answering you in chat are all paid calls to a model, and until now nothing capped how many of them one account could ask for. That budget is shared: it is what every builder's verdict comes out of, so one script pointed at it is everybody else's evening. There are now ceilings — thirty chat turns an hour, twenty filings a day, forty readings of the morning's task — set at multiples of real use, so an honest week never meets one. There are also limits on length: a proof can run to several paragraphs because conversation notes do, while a declared task is one sentence and is held to it. If you ever do hit a ceiling you are told so in plain words and nothing is lost — declaring stays free and unlimited, and a reading of your task that gets refused leaves the day exactly as a model outage does: declared, yours, and provable tonight. None of this is a coaching limit and none of it can touch the gate: a refused request is not a refused proof, and what a phase costs is unchanged.

A phone reclaims a background tab whenever it wants the memory, and an evening's proof is a paragraph of real thinking typed on a phone keyboard. Until now, anything not yet submitted died with the tab: a WhatsApp notification halfway through writing up the conversation you just had, and the answer was to type it again at ten at night — which is how a day you actually worked silently becomes a missed one. Three boxes now keep what you typed and put it back when you return: this morning's task, tonight's proof with its link, and whatever you were part-way through saying to Masterji. Each one is held against the thing it belongs to, so tonight's box never opens holding yesterday's words: the morning's draft is filed under the day, the proof under the task it is evidence for, and anything older than about eighteen hours is dropped rather than resurfaced. It stays on your own phone, never reaches the server, and nothing about it counts for anything — a draft is not a proof, and only what you submit is ever read or banked. An attachment is the one thing that cannot be kept; if you had picked a screenshot, pick it again before you submit.

The tour's note on the thinking partner read "Think with metrades assignments for questions", and the front page said the model only gets to "proposean advance". Both are one missing space, and both came from the same trap: a line of text that begins on its own line in the source loses the space in front of it when the site is built, so neither was visible anywhere except in the shipped page. The tour one mattered more than a typo — that sentence is the only place Masterji explains what "Think with me" is for.

A slide of the guided tour runs two to four screens on a phone — the mocks are the app's own cards, so they stack rather than shrink — and "Next" was the last thing on the page. Getting to the next slide meant scrolling past everything you had just read. It now sits at the bottom of the screen with the step count beside it, on a deck whose whole job is to be a short walk to the sign-in button.

The header used to carry the words "no run yet" before you had a run — a counter announcing that it had nothing to count, in the corner of the first screen you ever see, in the same grey as four other words beside it. It told you nothing you did not already know: of course there is no run on day one. It is simply not there now. The badge appears the day it has something to report, and the first thing it ever says is "1 day". A run that broke still shows what it reached — "0 · best 6" — because that one is worth reading.

More than one task in a day has always been allowed, because real work counts when it happens. What nothing checked was whether it was the same work: the same conversation, filed against three cycles in one evening, banked three proofs and opened VALIDATION — the phase whose whole job is stopping exactly that. Filing words already on your record now comes straight back, naming the day it repeats, with no model in the loop to argue with. Two real conversations in one evening still count as two, and the next step on something you already showed him is not a repeat — he says what makes it new instead.

Every day you had ever recorded was drawn under Today, so the card grew for exactly as long as you kept turning up — forty days of work made a wall you had to scroll past to reach anything below it, and Masterji will hold ninety. It now shows the last seven and says how many are behind them; one press opens the lot. Nothing was removed, and the button counts what it is hiding.

A day in your record was printed as "08-10", which reads as 8 October to anyone who writes the day first — and everyone this is built for writes the day first. It was 10 August. Opening a day, or reading one back out of a closed idea, showed the raw "2026-08-10" instead. All three now say what you would say: 10 Aug in the list, 10 Aug 2026 at the top of a day. The record is the part of this product that has to be trustworthy on its face, and it was arguing with itself about what day your work happened on.

The four-screen tour of how this works could only be opened from the sign-in popup, so it vanished the moment you had an account — which is exactly backwards. A visitor can always press the button again; the person who actually wants it is three days in, looking at a control they have never understood. "How it works" now sits beside "What's new" in the header, and on the goal-commit screen, and opens the same tour. Nothing in it is a manual: it is the four screens, the real refusals, and what the two ways of talking to Masterji are each for.

Under the proof box is a line saying exactly what tonight has to contain, and Masterji reads it while you talk — it is why he can say "that clears it" in the afternoon. The reading at the end of the day was never given it. It knew which phase you were in and what you had declared, and worked out the standard for itself, which is how an evening could ask for something the afternoon had already called enough. It now reads the same words you do, from the same place. It is a floor and not a checklist: your own words, in any order, scruffy, still clear it. And where he asked you for something specific this morning, that is still what tonight is judged against — a day spent on something else earns its proof for the thing you actually did.

Masterji's answer to a builder who is stuck has always been the same: name what you're avoiding, and here's the smallest real thing to do next. That is the right answer to being stuck on the work. It was also the only answer he had, which made it his answer to "my parents want me to stop wasting time on this" and to "I can't keep doing this" — and being handed a task is not what either of those needs. He now reads a message like that for what it is and answers you, not the task: no assignment that turn, no declaration demanded, nothing about avoidance. He'll tell you the true things he has — missing days deletes nothing you already banked, closing a goal is free and always was, and a goal you're keeping out of guilt is worth less than the one you'd choose today. He stays himself while he does it: no counsellor voice, no diagnosis, no list of techniques. And if it's bigger than a hard week he'll say plainly, once, that a coaching app isn't what you need and a person you trust is. None of this touches the gate — nothing gets banked because a night was hard, and if you mention real work in the same breath he still writes it down for you.

"What's new", "close this goal", the EN/हिं switch and the footer's source link were all under the 24-pixel minimum a target is supposed to have — the smallest of them, at fifteen pixels, was the one that closes your goal. All of them are comfortable now, and none of them moved anything around them. The run counter also holds one fixed width whatever number is in it: it used to grow with your streak and push "What's new" onto a different line, so a control you had learned the position of moved after a good week.

Tap "Who exactly has this problem?" in IDEA — the first thing this app suggests you say to him — and he replied "you're asking the right thing, but not the right week for stack or features." Nobody had mentioned a stack or a feature. Asked about it, he admitted as much: "You didn't. I'm correcting the drift before it starts." The rule that holds tech talk back until BUILD was written as something to do rather than something to do when you bring it up, so it fired at builders who never went near the subject — on the phase's own central question, usually in the first exchange. It now only answers the question you actually asked, and a question that IS this week's work gets an answer instead of a warning.

He has never asked you twice for something you said in the same evening — the notes he keeps under Today are exactly that promise. Earlier days were a different matter. He knew the count, "2 of 3 accepted", and not one word of what was in the 2, so on the fourth evening of VALIDATION he could send you back to the person you interviewed on Tuesday. Every proof he has accepted on this goal now sits in front of him while you talk: what you said, what you had declared, and which phase it was. Whatever phase it was, deliberately — a conversation you had while still in IDEA is a conversation you had, and being asked for it again because the row carries an older label was never fair. Nothing about the gate moved; he can just see the record it counts.

The chat has always offered a few things to say when you have not said anything yet, and they are written per phase — VALIDATION's are the ones about talking to a stranger without teeing them up to say yes. Only IDEA's could ever be seen: the test was whether the whole log was empty, and by the time you earn a phase it never is. The question is now whether you have said anything in the phase you are in, so arriving somewhere new brings its own three with it.

Masterji keeps this record because he demands one from you — and it has been growing since the first build, which is the point of it. But every screen in the product was downloading the whole thing on load, all seventy-seven entries and 42KB of them, just to work out whether to put an orange dot next to "What's new". That was being paid by everyone, on the landing page, before they had clicked anything, on whatever connection they happened to be on. A page now fetches the newest six — about a tenth of that — and opening this popup is what asks for the rest. If the older ones are still on their way, it says so at the bottom rather than making you wait for the newest one.

When the model fell over before its first word, "Masterji lost the thread" was saved as a message from him — his avatar, his bubble, sitting in your conversation weeks later looking exactly like coaching. Worse, it was read back to him on the next turn as his own words. It is now marked as what it is: a note about the conversation, drawn apart from it, carrying a button that sends your message again so you never retype a paragraph a server dropped.

If the cookie in your browser had gone bad — expired, mangled, or left over from another app on localhost — the app answered with an error it could not read back, decided the server must still be starting up, and sat on "The server is waking up." retrying every three seconds. Forever, on a screen with no way out, when the only thing wrong was a cookie. It now says the session ended, throws the dead cookie away, and puts you back on the landing page with the sign-in button where it has always been.

When the model was unreachable your proof was accepted — which kept your streak alive, and that half was right. The other half was not: it also counted toward the phase gate. So on a bad afternoon for the API, "think about the problem", proved by "I thought about it a lot", could unlock VALIDATION. The phase whose entire job is to stop you hiding in planning opened for exactly that. Those were two decisions riding on one word and they have been split. An evening he could not read is now filed as not read yet: the day still counts, it is on your record and in your streak, and nothing about it is held against you. What it does not do is bank a proof toward the next phase. Your words stay in the box, and sending them again once he is answering gets them a real reading — which is all that was ever missing.

He drafts tonight's proof out of your conversation, and filing that draft unedited takes one tap and no second opinion — he decided when he offered it. Describing Tuesday's conversation again tonight got the same treatment: written up fresh, in new words, so nothing recognised it as work already on your record, and it banked toward the phase a second time. He now has every proof you have banked in front of him, and will not write one of them up twice — he says which day it repeats and asks what today had in it. A second conversation, different work the same day, or the next step on something he has already seen are not repeats, and he still drafts those.

The wordmark carried Latin letter-spacing, which Devanagari does not take: the matras came away from the letters they hang on and the name rendered as four fragments. It had been doing that on the sign-in page — the first screen anyone sees.

A new chat was one message from him, a blank box and "Talk it through…", which is a thin invitation to the habit the rest of the day rests on: the draft that makes filing your proof one tap is written out of this conversation, so a builder who never talks here writes every evening from nothing. There are now three questions under his first message — real ones, and different in each phase. Tapping one fills the box for you to edit. They disappear the moment you have said anything of your own, and they move the gate by nothing.

The freeze at a blank goal box is rarely a shortage of ideas — it is not knowing how specific the answer is supposed to be, and one placeholder answers that with a single data point. There are three now, and two of them are somebody else's world on purpose: a reseller's payments, a hostel floor's notice board, a building's weekend baking orders. Tapping one fills the box for you to edit. None of them commits anything — the goal has to be yours.

The four dots that move you through the guided tour were seven pixels tall with no padding around them — on the phone most of that deck is read on, they were the one control in it nobody could hit. A slide runs two to four screens there, so the dots are the only quick way back to one you have already passed. The dots look exactly as they did; the button around each one is now four times the size. The tour's "Start yours" link and the "Sign in" link in the landing header got the same treatment, for the same reason.

It used to open on a goal already three-quarters of the way through VALIDATION, which taught the one phase a first-time visitor is guaranteed not to be in. There's a new first step now: the screen where you commit one goal, and the first thing Masterji says back — you're in IDEA, write the problem statement and the route to those people, and you may not message anyone until VALIDATION. The evening proof also comes before the proof he drafts for you, because the shortcut only means something once you've seen the box it goes in.

The gate slide went from '2 of 3 — refused' straight to the unlocked stepper, skipping the screen in between: the bar met, and nothing pressed yet. That screen had nothing worth showing while it stayed silent. It says so now, so the tour shows it too — the counter at 3/3, a full bar, and the button that hands over BUILD.

Two things were missing from the one screen where somebody hands over a Google account. It now carries Google's own mark, rather than asking for the account from a button that only claims to be a Google button, and it says in a line under it what that grants: Masterji reads your name and email address, and nothing else. There is also a close button in the corner now. Leaving was always possible — the backdrop asks, and Escape asks then closes — but a phone has no Escape key, so the careful dialog read as a trap to exactly the people it was being careful for. The × asks the same question the backdrop does, and a second press leaves.

The box you talk to him in is now the size of what you have written in it. It sits at one line while it is empty and gets a line taller every time you need one, so the whole of what you are about to send is in front of you while you write it — and the conversation above keeps the room you are not using yet. Earlier today it was a fixed four lines, which managed to be wrong in both directions at once: an empty slab above the chat before you had typed anything, and still a line short of anyone whose answer ran to five. Past ten lines it scrolls, five on a small screen, and the newest thing Masterji said stays in view while the box grows under it.

Four rows was measured on a tall screen. On a 667px phone it left the chat 181 pixels — less than two messages, and a box to type in that was bigger than the conversation it was for. Short screens get two rows now; longer replies scroll inside it exactly as they did. Nothing changes on a tall one.

The box you talk to him in was one line tall. Anything past a sentence pushed what you'd already written up out of sight, so reading your own reply back before sending it meant dragging a box the height of a single line — on the one screen where what you type runs longest, because thinking out loud is the point of it. It holds four lines now. Anything longer still scrolls, but four is what an evening's answer tends to come to, and you can see all of it at once. What counts is unchanged: nothing you say here is proof until you file it under Today.

"Nothing here counts" led the line under the composer, and alone at the front of a sentence it read as don't bother typing — at the exact moment you were telling him what you'd done all evening. It was also false: the draft that turns up under Today is written from that box and nowhere else. The rule has not moved an inch, it just goes second now. Once today's task is declared the line says he writes tonight's proof from this conversation, and that nothing counts until you file it under Today. Before you've declared it promises the draft instead of claiming it — with no task to hang notes on, he genuinely isn't taking any yet.

Declaring a task used to unfold the whole evening under it — what tonight has to contain, the proof box, the link field, the screenshot, and Submit proof — four fifths of the card, for work nobody can do for another ten hours. Masterji asks for about two minutes a day and the screen after those two minutes looked like homework. Now the morning closes: he takes the task, says what he makes of it, and tells you nothing is owed until tonight. The evening is one press away, and it opens by itself once there is an evening to open — a push-back to answer, a try already made, notes he has started writing, or simply five o'clock.

There was no way to find out what Think with me did except to press it — which writes the setting to every device you own — or to leave for the tour. Pressing a control to learn what it does is a dare, not a choice. "What's the difference?" now sits beside the switch and opens the answer in place: what each mode does to you, which half of the work each is for, that your pick follows you across devices, and that neither one moves the gate. Closed, it costs one short line.

"What's the difference?" lasted one release. It put a third piece of text in a bar that already held the switch and a caption, and three things reading for attention in one strip is clutter however useful the third one is. The row is back to the two modes and one line about the one you're in. What Think with me is for is explained in the tour, at the length it actually takes.

Under Coach me it read "Assignments and push-back. Switch before there's anything to declare." Switch to what — the other option is already sitting right there, named. And it read as a deadline the control has never had: you can change modes whenever you like, including after today's task is declared. It also spent a word the screen was already spending, because the line under the same box says "Declare today's task under Today first" when you have nothing declared — two "declare"s two inches apart, one of them something you could act on. The caption now says what the mode you're in does and stops. What Think with me is for takes a paragraph, and the tour has one.

Everything on the front page was an assertion about a coach nobody had met — a visitor was asked for a Google account having been shown no part of the product at all. The screens existed only inside the tour, behind a click most people never spend. There is now one still frame of the real dashboard on the way down the page: the goal and its phases, a day that was declared and accepted, and Masterji declining a tech-stack question in VALIDATION. Every pixel of it is the app's own stylesheet rather than a drawing of it, which is the rule the tour already keeps — a second copy of the product drifts from it the first time either one moves. Still no proof counts on that page: those live in gates.py, and a marketing page that quotes them is a promise nothing keeps the day one of them changes.

"One goal" — the first thing you meet after signing in — asked for the hardest decision in the product and then ended on what it would cost you to change your mind, a warning on the one screen where nobody has done anything yet to be warned about. It now says the shape of the thing instead: one task each morning, proof of it each evening, about two minutes a day. That you can close it whenever you like is still there; that he'll remember has moved to where you close it, which is where there is something to remember.

Banking the last proof looked identical to banking none of them: same quiet button, nothing said, and a builder could stand there for days already holding what the next phase costs. The card now says it outright and hands you the phase. Proofs past the bar stopped being counted into the numerator, too — nobody needed to read '8/3'.

Visiting masterji.mscsoftwares.in used to bounce you straight to a sign-in wall: a name, one sentence, and a Google button, with every reason to want the thing living in a README. There is a real page there now — what it is, what a day with it looks like, and why the gate is worth trusting — and it no longer asks for an account before showing you anything.

While Masterji writes tonight's proof out of your conversation, the line under the reply box still read "nothing here counts" — which is the rule, and was the wrong half of it to say on its own. He is writing your words down as you say them; being told you are wasting your breath while that happens is exactly backwards. It now says he has it and how many pieces are still needed, and keeps the rule that filing is yours. On a phone the Today tab says "notes" too, so you can see it without leaving the conversation.

Sign out and you were dropped on the sign-in page — a wordmark and a Google button — which answers 'I'm done' by asking for the account you just put down. You land on the home page now, the one screen that says what this is, with Sign in still in the corner for whenever you want back in.

"sign out" was a sixteen-pixel word in the corner of the header a thumb rests in, a hair away from "What's new", and it fired on the first press. Missing cost you the session and a full round trip through Google. It is a proper target now, it asks once before it goes, and the question drops the moment you look elsewhere. It also stopped being the only underlined thing on the screen — an odd honour for the door.

Two changes to everything before the app. The sign-in page is gone: pressing "Start free with Google" used to load a page holding a wordmark and one button, which answered a decision by asking for it again. The same press now opens sign-in over the page you were reading, blurred behind it — the argument that got you that far stays on screen while you make the account. Clicking away doesn't lose it either: it asks whether you meant to go back, and staying is one button. The guided tour is four slides instead of nine. It had grown into a manual for a product nobody had signed up for yet — the phase diagram and the four-line day the home page already spells out, how to retire a goal nobody has made, the running proof notes the screen announces by itself. What's left is the three things worth knowing before you commit: your first two minutes, which of the two boxes counts, and what the gate does when you ask it early.

Enter sent your reply and Shift+Enter started a new line — a bargain that needs a Shift key to keep. A phone doesn't have one, so the ⏎ on the keyboard, the one key there for starting a paragraph, sent the message instead: half a thought, gone, and no way to type the second half. On phones and tablets the return key now does what its icon says and Send is what sends. On a desktop nothing changes — Enter still sends, Shift+Enter still breaks the line.

The evening card asked what tonight needs, then showed you what he had already written from your conversation — the question above the answer, with the answer's button set in the smallest type on the card while Submit, the step after it, got the big one. His draft is first now and its button is a real button, and it drops you straight into the box it fills. When the draft already clears the bar the full ask folds away, because by then it is something to check against rather than something to read.

It was one button showing the language you already had, which is the one label that gives nobody a reason to press it. Both languages sit in the header now with the live one lit — the same fix the coach/thinking-partner switch got, for the same reason.

The complaint, and it was the loudest one: you name three things your customer said, in one sentence, and he answers "that's one, not three" — then asks for all of it again, twice. He now writes tonight's proof down as you talk instead of holding it in his head until the end. What he has so far shows up under Today as notes, with the pieces still owed listed under it, and anything in those notes is banked: he is not allowed to ask you for it a second time, in chat or when you file. The counting also stopped being his. He hands over the pieces one by one — each thing said is its own entry — and the server does the arithmetic, so "1 more thing they said" is a subtraction and not his reading of his own paragraph. When you tell him he has misread you, re-reading is his job, not yours to work around. Notes are not a pass: only a draft with nothing missing files straight through, a short one filed anyway is judged on its merits like any other proof, and the gate counts exactly what it counted before.

File tonight's proof, get it accepted, then keep talking about more work you did — and he used to answer "there's no task declared this morning, so I have nothing to pin it to", with the card right beside the chat reading "Declared: …" and a green "✓ accepted" under it. Both of those cannot be true. He now says the true one: today's cycle is already declared, filed and closed, so he has nothing open to pin it to — and if that was a second piece of real work, declare another task under Today and file this against it. Several cycles in one day was always allowed; the sentence pointing you at it was missing. Either way the draft he wrote out of the conversation comes back to you in the chat, because the work behind it happened.

The button for ending a goal at LAUNCH said "Close this out", which is what you say about a ticket. That button can only appear when your record already holds accepted proof, so it now says so: Earned. Proof is on the record — and the button reads Claim the win. Nothing about the gate moved; the words just stopped underselling the one moment the whole thing is for. The quiet "close this goal" link is unchanged, because it is also the way out for an idea that didn't work, and that exit stays plain.

Think with me's caption ended "The gate is unchanged" — true, and the most important thing on that line, which is exactly why it shouldn't have been spent as a half-sentence read once in passing. It now has room in "What's the difference?", stated properly: proof is still the only thing that opens a phase. The captions themselves are down to naming the mode you're in.

'Not yet, 0 of 1' used to stay on the goal card after the proof that answered it landed — sitting under a full bar, contradicting the counter directly above it. Each answer is now pinned to the situation it answered and goes when that does.

Opening a day, a phase or What's new dropped a flat dark sheet over the dashboard, and the work you opened it from stopped existing for as long as you read. The sheet now blurs what's behind it instead: your goal, today's declaration and the record stay legible as shape and colour, so the record you're reading still reads as a detail of your board rather than a screen of its own. The sign-in popup has worked this way since it shipped; every popup now does.

Miss two days and the header said 0, which reads as though none of it happened, at exactly the moment stopping looks reasonable. It now says 0 next to your longest run on this idea, and the morning after a break the Today card says which day one this is.

There is work that comes before there is anything to declare — figuring out who actually has the problem, or which of three ideas is the one. Masterji was no use for that: every answer came back as an assignment. The button in the header switches him to a thinking partner, and he changes sides of the table: questions instead of demands, two or three concrete options when you're stuck, his own reasoning out loud so you can disagree with it. The setting stays with you until you switch it back. What it does not do is move the gate — phases still open on accepted proofs, and thinking out loud with you was never going to be a way around that.

Masterji has always had a second way of talking — a thinking partner, for the part of the work that comes before there is anything to declare. You would have had no way of knowing. It was a single button in the top corner reading 'Coach', which told you the mode you were already in and left you to guess that pressing it offered a different one, and it lived in the strip with your username and the sign-out link, where nobody looks for a way of talking. On a phone the whole explanation was a hover tooltip, on a screen that has no hover. It now sits above the box you type in, as two options with the live one lit: Coach me, or Think with me. The line under it says what each one does. What has not changed is the part that matters: the gate never read this setting and still doesn't. Thinking out loud with Masterji was never a way around the door, and asking him to do it costs you nothing.

It used to be one frozen screen of the app with a conversation on it, which showed what Masterji sounds like and nothing about how to use him. It's a tour now — eight steps through the real screens, with the parts that matter circled and answered in the margin: declaring in the morning, what the chat does and doesn't record, the proof he drafts out of your conversation, filing in the evening, and the gate refusing until the proofs are in. Every number and refusal in it is quoted from the code that produces it.

Masterji asks you for proof of work every evening, so he keeps a record of his own too. Every change worth noticing lands in this list — written from the admin, so an entry can appear the day it ships rather than the day someone remembers — and the link in the header carries a dot until you've read the newest one.

A line in the record was a dead end: a date, the task, and a tick or a cross. The proof, the screenshot, the tries that were pushed back and what Masterji said about them were reachable from nowhere. Every row opens now — in the sidebar and in the phase drill-in — and a day opened from inside a phase closes back into it rather than dropping you out of where you were reading.

Two text boxes, the same free text, and nothing anywhere saying that only one of them reaches the gate — so the commonest way to lose an evening was to describe real work in the chat and file nothing. The chat says what it is now, and keeps saying it under the composer after the placeholder has gone. Wherever the product tells you where to file, it says "Today" instead of pointing: "above" was the left-hand column on a laptop and a tab you couldn't see on a phone.

Two kinds of turn used to end with your message on screen and no answer under it. One was him writing tonight's proof up from the conversation and saying nothing else — the draft landed on the Today card, which on a phone is the screen you weren't looking at, so from the chat it read as being ignored. He now says where it went, without repeating the draft itself. The other was the model dropping the turn before its first word: the red banner explaining that is gone by morning, and what was left in the record was a conversation that had answered every message except one. That line stays in the transcript now. An answer that broke off partway is still kept as far as it got.

The gate out of IDEA used to ask for ten specific people you could reach. But the honest answer is often a group you can walk up to — the mess aunties on your campus, the three shops on your street — and that was getting refused for the wrong reason. Now Masterji asks who has the problem and how you'll get in front of the first few this week. A list of ten names still passes; it just isn't the only way through any more.

You'd tell Masterji about the conversation you just had, he'd coach at you about it, and the evening would end with nothing filed — because turning what you said into what the proof box wanted was left to you. Now he watches the conversation against the bar, and the moment you've told him enough he writes the proof up himself, in your words and your facts, and asks whether it's right. It appears in tonight's check-in as a draft: file it, or edit it first. Filed as he wrote it, it goes straight through — he already decided it counted, so he doesn't get a second opinion on his own draft. Nothing is recorded until you file it, and he won't offer at all while a piece is genuinely missing — he'll ask for that one piece instead.

Masterji is meant to be hard to satisfy. He was also, too often, rude about it, and he had no way to say that something was good enough — the coach carried the phase's refusals and a proof count and no definition of 'enough' anywhere in his head, so the only reply he could build was 'not yet, give me more'. He now reads the same bar the check-in form shows you, and when your answer meets it he says so. He also judges what your evidence CONTAINS rather than the shape you wrote it in — the playbooks describe what has to be there, not a format you have to reproduce, and you were never supposed to be tested on how well you'd learned our phrasing. He still won't discuss your tech stack in IDEA: he'll decline it in one line and give you the reason, once, instead of a lecture. And when a piece of real work lands, he names it. Hard on the work, easy on you.

Behind the scenes: one long check-in could stretch its column until every other column fell off the right edge. Widths are capped now, so a row can be read at a glance.

Answer a push-back and your new proof was read by a Masterji who had no memory of what he'd asked for — free to send it back for some brand-new reason. Builders described it exactly right: you give him what he asked for and he still doesn't get it. He now sees every try you've made tonight and the words he sent each one back with, and he is told plainly that he does not get to raise the bar on a second look or find a fault he could have named the first time. After a third refusal he has to stop and answer one question before judging again: is the work missing, or is the work there and the two of you failing to understand each other? If it's the second, that's his failure — he takes it, writes your proof out as he now understands it, and says what he'd misread. Nobody has to be a good writer to get credit for work they did. What has not changed: nothing passes because you submitted it often enough. Work that isn't there gets refused on the fourth try and the fortieth.

When Masterji writes tonight's proof out of your conversation, the Today tab now says so in a word rather than showing the same dot it has shown since the day started — on a phone the draft was landing on the screen you weren't looking at, with nothing to say it had arrived. And if he drafts one before you've declared a task, he hands it back to you in the chat with the reason instead of dropping it: the work behind it was yours, so the writing-up is too.

The daily loop has always been filed under your own date, but the dashboard read it back off the server's UTC clock. Past midnight in India that meant declaring a task and watching the empty "Morning. One task" form come straight back with the task sitting in the record underneath it — declaring looked like a button that does nothing. Reads and writes are on one calendar now, Masterji knows about the task on the hook at 1am, and closing a goal can no longer report fewer days active than the record shows.

Check-in text in the phase drill-in and in the sidebar record was clipped to one line, and the part that mattered was always the part cut off. Both wrap now.

Stacked, the dashboard and the chat made a page four screens tall with today's task buried in the middle of it. On a phone they're two tabs now — Today and Masterji — and a dot on Today tells you the day is still open.

An expired or unreadable login cookie used to surface as a server error. Now it reads as what it is — signed out — and hands you the sign-in page.

Masterji runs on a free instance that falls asleep, and the first request of the day can take half a minute to wake it. Instead of a blank screen you now get a note telling you he's on his way — in the app and on the admin.

Goal, gate, today's loop, the record and the archive stacked into a column taller than the page and dragged the whole layout down past the chat, which had its own scroll and stayed put. The dashboard column scrolls independently now.

Masterji's own calls to the model were enough to knock over the small box he runs on. Fewer moving parts per turn — and if the model is unreachable, your proof is still accepted with a stock reaction. The daily loop never breaks because an API flaked.

Leaving IDEA meant naming ten people who have the problem — a rule the gate enforced while the copy stayed vague about it. The phase hint, the proof ask and the refusal all say the same thing now. (That ask was loosened two days later — see the top of this list.)

Evening proof can carry a screenshot, and Masterji looks at the image rather than at your description of it. The task you declare in the morning also gets read the moment you declare it: if it's work for a phase you're not in, he says so at 9am instead of letting you find out at 9pm. Off-phase work is flagged, never blocked — the gate is what makes a sideways day cost something.

A proof that gets pushed back is no longer overwritten by the one that finally lands. The misses fold away behind the accepted proof: out of the way, but not deleted.

Everything the coach judges by lives in the repo as plain markdown, one playbook per phase, small enough to read in ten minutes and crediting The Mom Test, The Lean Startup and MAKE by name. Added on top of the playbooks: the curation policy — what earns a method a place in the corpus, and why scraped founder tweets never will. The demo and the README now point straight at it.

The streak is about the idea in front of you. The day count is about you. Retiring a goal resets the first and never touches the second.

A retired idea isn't just a line in an archive. Open it and the days come with it — what you declared, what you proved, what got pushed back — and it stays reachable while a new goal is running, because a record can't do its work if it's only visible in the four seconds between goals.

Distilled from the way Pieter Levels works: ship something small now, put it in front of people, and let them tell you what comes next. BUILD had gates but no cadence to hold you to.

The gate's "no" used to be a bare count — two of three proofs, come back later. Every refusal now carries the next concrete action with it, and a single module owns all the builder-facing phase copy, so the phase hint, the proof ask and the refusal can't drift into three different answers about the same rule.

The proof box used to be an empty box. The phase's ask now sits above it, with an example of a proof that was actually accepted.

Two routes around the gate closed. Every check-in is stamped with the phase it was made in, so proofs count where they were earned — and not even the admin can move a row into a phase it didn't happen in.

One goal at a time is enforced by the database, but one goal was never meant to mean forever. Close it out — achieved or dropped — in one honest sentence, and Masterji reacts to the ending on the record. Achieving your goal is never blocked by the phase you happened to be in.

Done for today doesn't have to mean done for the day. Once the evening proof lands you can declare the next task, as many declare-then-prove cycles as the day actually had.

One goal, four phases (IDEA → VALIDATION → BUILD → LAUNCH), a task declared every morning and proof owed every evening. Phases open on accepted proofs and the server counts them, so the coach can be talked to but the gate can't be talked around. Shipped with Google sign-in, a Hinglish toggle, and a canned demo you can read without an account.

The first phase was being coached on general principles. It has its own playbook now, distilled from Ash Maurya's Lean Canvas and Jobs-to-be-Done: one paragraph naming who specifically has the problem, what they do about it today, and why that isn't good enough. That anatomy is what your declarations get read against.

The stepper isn't only a progress bar. Tap a phase you've left and you get the window it covered and every check-in inside it.