Eleven independent audits read the plan, the rules, the designs and the code. This is what they found, in plain terms. No task numbers.
Right now V1 is defined as everything working at once — every app, every feature — so one broken thing blocks shipping anything at all. The suggestion is to cut it to: make an account, recover it, message another OSL user directly, plus ONE working chat app and ONE working email. Everything else shows as coming soon until it is genuinely proven.
This is the single biggest unblock in the whole audit.
Group chats and server channels are not built. Discord DMs are the closest thing to working.
Is 'it works in DMs' enough for V1, or must it work in servers too?
Nobody can actually change a setting today. The screens you approved were the design drawing shown inside the app, with the real controls hidden behind it.
Both jobs are large. Which comes first?
Today it has no way to find out and no record to resume from. It could say 'we do not know', or refuse to send until it can be sure, or something else.
This is a product decision and it is yours.
174 surfaces have never appeared on the review site — popups, overlays, empty states, the whole view-once flow. Some belong to retired designs.
Do you want all of them, or only the ones actually reachable in the app?
A test cannot pass because it requires something you decided not to do. Your decision wins, so the test needs rewording — but until then it blocks six other tasks.
The AI cover writer cannot be built into the Windows version — the library it needs will not compile for Windows. Tasks still require Windows proof of it, which can never be produced.
Two rules pull opposite ways, so the screen can never satisfy both.
The rule says it finds things for review and never deletes automatically. The shipping app still contains the delete wording and permission. Two separate audits found this independently, which is why it is trusted.
There is no way to record that something previously marked done has failed. That is how 94% complete can sit alongside things that do not work.
Not on Discord, not on Telegram, not anywhere. The pieces for a Discord DM exist, but no message has gone from one account to another and been read back. Records that look like successful sends turn out to be tests with fake data, or runs that stopped before sending.
The overlay can attach to a channel box while encrypting for a different person. Everything looks correct and the message goes to the wrong place. 'The strip appeared' is not proof it worked.
Close Discord mid-send, lose the network, or crash the app, and OSL cannot tell what happened, cannot finish, and cannot undo.
Ten sections show the design drawing inside the app while the real controls are hidden by styling. On Friends all 29 controls do nothing, and the Friends button is invisible — the Privacy button is what actually moves you.
31 screens are approved, not 38. A counter jumped by seven and every later count inherited it. The seven wrongly counted are Home and the six Chats screens — the ones you have rejected three times.
A release-status sentence about setup not being validated renders above the title on onboarding screens, in one case overlapping it. The code itself says that sentence must not appear there.
Or remove the design stand-in so the emptiness is visible instead of hidden.
Its button is invisible today.
Send it, read the stored message back byte for byte, and decode it on the second account. Then one deliberately wrong attempt that must send nothing.
So the app can say what really happened.
It is not in the binary; the app should not imply otherwise.
The current set is the old look — they were captured four minutes before the new build existed.
And put Home and the six Chats screens back where they belong.
Each names its evidence and says plainly where something is unproven.
What OSL is meant to be, completely.
Your remaining work, with the steps, ordered by what it unblocks.
What the agents are blocked on.
The plan, the rulings, the build, the documents and the checkers.
Contradictions, impossible requirements and rules that block work without protecting anyone.
What works where, and the false pass that could bite.
What works where, and what is missing.
Which are worth doing and which are not.
Ends with the shortest path to one carrier genuinely working.
Whether OSL recovers from a failed send.
Including popups, overlays, the composer and the eye.
What to keep, simplify or cut.
One-party, two-party, groups and servers, as a runbook.
How every surface gets reviewable.
Accounts, machines, and what testing still needs.