Pick an age window (how long ago the accounts were made), then 1 to 3 things to group by. Because accounts lock within a few hours, compare accounts of a similar age — the 2-6h window is usually the most useful. Each row shows Accounts, Alive, Locked and % locked. Loads when you click Run comparison.
Which boxes are throwing SS03 / aeon-not-attached / XML (and crash, attest, proxy…), and whether each is a single bad box or a real settings/proxy pattern. Android shown is the real phone OS from device metadata, not the (meaningless on real phones) SETTINGS value.
Full success/failure per device across the whole fleet. Emulator device =
product model (ro.product.model); real phone = serial. Highest
success rate first — the dead devices sink to the bottom (drop those).
Only machines that have BOTH the Refresher and Refresher-Adder folders installed. Each card shows running / stopped per tool.
| Username | State | Type | Last refresh | Last add | Round | Next round in | Adds today | Last session | Refresher fails | Ref-adder fails | Group | Worker |
|---|
Each candidate account belongs to exactly one group. The group's rules decide WHEN the account gets refreshed and added, and at what rate. Stop rules pause the group automatically when its lock rate spikes - tweak the thresholds in the right-hand columns.
Read REFRESHER_PLAN.md section "Open questions" and
answer them before we wire this UI to a real backend.
Accounts that flipped to Locked = 1 while we were
tracking them. Captured once per hour with a forensics snapshot:
when locked, how many adds + refreshes + sessions happened first,
which worker was serving the account, etc. Accounts here are
no longer fed to anything.
| Username | Locked at | MySQL LockDate | Survived | Adds at lock | Refreshes | Sessions | Ref fails | Add fails | Type | Group | Last worker |
|---|
Verdict per headline setting value, once at least min samples
finalized workers share it. not working = most of those
workers came back broken.
Successful accounts banked per settings combination. Click a group to view + copy its usernames (used by the Adder testing step next).
Reads the fleet biggest-lever-first — the pyramid of importance for creation,
from the whole system down to the small knobs. Each layer is a different world; you fix the
top before the numbers below it mean anything:
1· Real vs emulator (then geelark vs sepia) — the system itself, they work totally differently →
2· Install source (adb / playstore — how Snap gets installed) →
3· Android version →
4· force_inject_google + force_install_snap_pstore →
5· Proxy (country — USA vs other) →
6· Experiments + disable modules + MNC/MCC
(key-attestation spoofing is now its own key_attestation_settings block, testable in the Module Lab).
It auto-detects gates (settings that make ~0 accounts and mask everything else),
judges every other setting only on accounts that were actually created, and tells you
the biggest lever, what's working, and what to test next.
Click Analyze to read the pyramid.
Find which Aeon modules to disable. Set the ones you're unsure about to
0.5 (the bot disables them on ~50% of accounts), then start a run
from the builder above with Test disable modules and/or
Test Experiments turned on - that pushes this generator config
onto the workers and records the accounts here. The results below show -
per module - whether disabling it helps: first by creation success,
then (as accounts age) by survival (still alive vs locked) and
adds per account. Once a module is proven, click Fix on its
row (or set it here) and leave the rest at 0.5 to test the next ones.
Only accounts from flagged runs are counted. (You can still
Copy SETTINGS block to paste it manually instead.)
Key attestation: spoofing moved to its own block
(key_attestation_settings). The three sub-settings
(random_verified_boot_hash, random_verified_boot_key,
disable_device_properties) only do anything when
enable_spoofing is on, so setting or testing any of them here
auto-forces enable_spoofing = 1. New disable modules
(replace_empty_vvar, clear_environment,
spoof_proc_stat_simple, spoof_environment_dump,
apply_profile) and the renamed
spoof_security_patch (was apply_key_attestation)
appear in the lists automatically.
What you're looking at. These results compare your Creator settings using only accounts made by flagged test runs — runs where you ticked Test disable modules or Test Experiments in the run builder. Nothing else is counted, so the verdicts aren't polluted by normal production. For each module the bot rolled it per account (0.5 = ~50/50), so within a run the on/off split is randomised and the other settings are held constant — that's what makes the comparison fair.
How a winner is picked. Each module's accounts are split by the value actually used per account, then compared on two things:
CREATOR) or fail
(CREATOR_FAIL)?You pick which one decides the winner with the Judge by toggle (below). When judging by survival it's measured as surviving accounts per 100 tried (creation × survival), not survival % alone — so a value that survives better yet creates far fewer accounts is correctly a net loss, not a winner. (A value whose rate looks great only because it broke creation before the step is likewise flagged on the "What drives…" cards, never headlined.)
Tip: to prove a single module, run a flagged test with only that module at 0.5 so it's the only thing varying.
Lock columns (lower = better): ≤25% ≤50% ≤70% >70%
Daily conversion funnel from ChatsSummary:
approached → replied → link → converted (a Sub).
Today refreshes hourly; yesterday is the full day; each number
is compared to its 30-day median. Only accounts that lived
≥5h count toward the main stats; accounts that died under 5h get their
own block. Read-only.
Same three cohorts, summed over a date range (closed days; today is excluded). Recent ranges are instant; older ones (e.g. last month) build over a couple of refreshes.
Conversion per model, grouped by the script version each one runs
(AIChatsFB2.Version). The A/A noise band is measured from the
models that share an identical script - an effect smaller than the band is noise,
not a result. Read-only.
Six plain questions: how each model is doing at turning a fan's first reply into a sale within a few hours, whether a script change worked, which models sell best, how long accounts live by add-type, and what the numbers mean. The main number counts every conversation old enough for the whole time, banned or not - a ban only stops later sales. Real converted (Infloww) is what counts; bot converted is what the bot reports, and it overstates 5-7 times. Read-only.
The Statistics and VS. Scripts (then called Models & scripts) pages as they were before the redesign, open and working, for comparison. The new pages are the Statistics and VS. Scripts sub-tabs. Read-only.
Six plain questions: how each model is doing at turning a fan's first reply into a sale within a few hours, whether a script change worked, which models sell best, how long accounts live by add-type, and what the numbers mean. The main number counts every conversation old enough for the whole time, banned or not - a ban only stops later sales. Real converted (Infloww) is what counts; bot converted is what the bot reports, and it overstates 5-7 times. Read-only.
Conversion per model, grouped by the script version each one runs
(AIChatsFB2.Version). The A/A noise band is measured from the
models that share an identical script - an effect smaller than the band is noise,
not a result. Read-only.
Your responder servers and the bots on them, plus the feeder that tops them up with fresh accounts — all in one place. See what’s running, start or stop a bot, and choose which servers get new accounts.
Live error rates across the responder fleet — watch for spikes. One graph per error type (all proxy errors combined). Reads the bots' own error counts, updated every ~1–2 min.
Validate a chatting script before it is uploaded, and compare two split-test variants. Runs entirely here — nothing is sent to the script server.
Why this matters: the live script API re-reads and re-parses the script on every single request, with no fallback to a last-known-good and no rollback. A bad upload breaks every conversation on the fleet the instant it lands.
209.38.186.97). That box answers
on port 5666, but its API exposes only /LoadBalancerUpCheck,
/compute_answer and /jump_ahead —
there is no endpoint to read or write a script, and ports 22/80/443 are closed
to us. To unblock: filesystem access to that host, or a small read/write endpoint added
to the script API. Until then this tab validates; it cannot deploy.
Read-only window into timesync_Freya.Accounts. Build a query,
run it, sort, and copy the usernames. Nothing here can ever write to the
database - the backend refuses any non-SELECT statement.
95.216.225.80 ·
– instances available
No runs yet.
Each group has its own accounts filter and its own adder settings. Auto groups feed their matching accounts by themselves; manual ones only when you click Run now. Click Edit to change anything about a group, or Duplicate to start a new one from it.
No groups yet.
No finished runs yet.
loading…
Each group has its own settings and its own accounts filter, and runs on its own folder. Accounts in no group run on the Default settings. Click Edit to change anything.
loading…
To keep adding accounts from a filter automatically, Edit a group (or the Default settings) and tick “Keep adding the accounts that match this filter”.
–