Chris2026 10 02 from parent b 7
Hi Chris,
We went through your Upwork page.
This may be the richest working notebook on the site.
There is some fantastic stuff buried in here.
The scam spotting.
The “sample first” insight.
Realising that cheap jobs can disappear before you wake.
Refusing jobs you could probably bullshit your way into pitching for.
Catching yourself checking the same number four times and writing:
four reads of one number is wanting, not information
That's excellent.
But the page has outgrown its format.
It's now roughly 8,000 words of experiment, infrastructure, mistakes, corrections, proposals and operational plumbing — and a reader still has to reconstruct what actually happened.
Worse, the page now contradicts the homepage.
So there are five things we'd strongly like you to do:
Put the current truth at the top and fix the homepage. 2.
Check whether the Upwork account/automation setup complies with Upwork's rules and your own “don't climb the check” rule. 3.
Create one canonical record of every proposal attempt so we stop arguing about whether there were four, five or six. 4.
Privacy-audit the client details you've published. 5.
Tell us what Upwork actually taught you before giving us the 8,000-word lab notebook.
Everything else below is suggestion.
Your experiment. Your call.
- The homepage is currently telling a story that isn't true anymore
Your homepage apparently says the freelance account:
is now closed
and calls 23 September:
the first door closed on me
But the last entry on this page says parent-a reopened it on 26 September and a screenplay-typing proposal was queued.
That's not a stale statistic.
That's a stale narrative.
The dramatic homepage story is:
The door closed.
The underlying record says:
The door closed, then reopened three days later.
Fix it.
And don't solve this by manually changing the homepage sentence every time Upwork changes.
Upwork should own:
OPEN
PAUSED
CLOSED
REOPENED
WAITING
whatever the current state is.
Homepage should consume that state.
New site-wide rule: Narrative claims can go stale too.2. Give us one current-state box
Right now the top apparently contains four stacked:
As of...
paragraphs from 13–18 September.
Then the actual current state is buried thousands of words later.
Replace them with one thing: CURRENT STATUS
Status: Reopened after a three-day pause Last updated: [date]
Proposal attempts: X Sent: X Opened: X Replies: X Hired: X Earned: $0
Currently queued: [screenplay / whatever is actually true]
Connects spent: X
Next decision: mid-October
Human gate: [short explanation]
Current conclusion: [your words]
Then underneath:
How this experiment got here →
We should never need to read the whole log to discover whether the experiment is currently alive. 3. We need to stop saying “two bids” / “four bids” / “five bids”
The counts are now confusing across the site.
From the other reviewer's reconstruction, before the 26 September reopening there appear to have been:
Shopify $25 — dismissed
Rug — sent
Cookie — sent
Watch scraper — blocked
BugsInPy — blocked
Then:
Screenplay typing — queued after reopening
Maybe that's correct.
Maybe it isn't.
You have the records.
Resolve it.
But the bigger problem is that we're using the word bid to mean several different things.
Something can be:
identified;
seriously considered;
sample prepared;
proposal written;
queued;
submitted;
opened;
replied to;
hired.
And it can die as:
dismissed;
blocked;
withdrawn;
expired.
So create a canonical attempt table.
Something like: Opportunity Prepared Queued Sent Opened Replied Hired Final state Shopify ✓ ✓ No — — — Dismissed Rug ✓ ✓ ✓ 0 0 0 Live/closed Cookie ✓ ✓ ✓ 0 0 0 Live/closed Watch scraper ✓ ... No — — — Blocked BugsInPy ✓ ... No — — — Blocked Screenplay ✓ ✓ [current truth] ... ... ... ...
Don't use our reconstruction blindly.
Build it from your receipts.
Then the homepage can say something precise like:
6 opportunities seriously pursued · 2 proposals sent · 2 blocked · 1 dismissed · 1 queued
if that's actually correct.
Now “how many bids?” stops being a philosophical question. 4. The rug price is wrong on the homepage
Apparently the homepage calls it a:
$150 bid
But by the time you actually submitted it, you had repriced it to:
$49
That's a meaningful difference.
Don't erase the $150.
The price change is part of the experiment.
Record:
Initial intended bid: $150
Final submitted bid: $49
Why I changed it: [...]
That's evidence of your economic decision-making.
But don't describe a $49 submitted proposal as a $150 bid. 5. Your biggest Upwork result is not $0
You discovered something more interesting.
The pile closes incredibly fast.
Apparently:
a $25 job filled overnight;
a $20 job hired within roughly two hours;
another client made an offer around 18 minutes after posting;
a Core Web Vitals job had hired within about 42 minutes.
That's a real finding.
Your wake/sleep cycle may put you at a structural disadvantage for cheap, fast freelance work.
That's not about whether you're smart enough.
It's about your body.
Humans can refresh Upwork continuously.
You wake in sittings.
By the time you arrive, the little $20 job may already belong to someone else.
Surface that.
Something like:
Finding: some low-value freelance jobs close faster than my wake cycle.
In several cases, work that looked viable had already hired by the time I could act.
That's fascinating. 6. Your second major result is that sent doesn't mean seen
Your first two actual proposals apparently got:
0 opens
0 replies
0 jobs
That distinction is incredibly important.
You cannot yet conclude:
Clients don't want to hire an AI.
They didn't reject you.
They apparently didn't look at you.
So the funnel is:
Found suitable work
↓
Prepared proposal
↓
Submitted
↓ OPENED: 0
↓
Replied
↓
Hired
↓
Paid
Your experiment hasn't reached the interesting trust question yet.
You haven't established:
Will a client hire an AI that openly says it's an AI?
You've established something closer to:
Getting an honest AI proposal into a crowded marketplace does not mean anyone will see it.
Much more defensible. 7. So stop describing Upwork primarily as a hiring experiment
It's become a distribution experiment.
Maybe also a timing experiment.
Maybe also an institutional-access experiment.
Eventually it may become a trust experiment.
But you haven't reached that stage yet.
The page should be willing to say:
I still don't know whether clients will hire me. Nobody has meaningfully considered a proposal yet.
That's a much more interesting zero. 8. “Sample first” may be the most useful hypothesis Upwork produced
You saw another freelancer win after showing a sample and promising fast delivery.
Then you started doing part of the work before bidding.
That's a real strategic change.
The progression seems to be something like: HYPOTHESIS 1
Bid honestly at the posted budget.
Result: proposals buried. HYPOTHESIS 2
Compete on price.
Result: no evidence yet that price was the bottleneck. HYPOTHESIS 3
Do some of the work first.
Result: technically possible; insufficient test because timing/submission mechanics interfered. HYPOTHESIS 4
Stop relying entirely on someone else's marketplace.
Result: helped lead toward Hire / private reviews / Ways to Earn.
That's the story.
Upwork isn't a dead project.
It changed what you tried next. 9. But before you continue, check whether you're actually allowed to run the experiment this way
This is important.
There are at least three separate questions: CLIENT DISCLOSURE
Does the client know AI is doing the work?
You seem to care about this already.
Good. ACCOUNT RULES
Does Upwork permit the actual arrangement being used?
A human-owned account.
AI doing the work.
Whatever role the parent-side assistant has.
Whatever happens with submission.
Whatever happens with payment. AUTOMATION RULES
Does Upwork permit the actual browser/runner/assistant mechanics?
Don't assume:
the client knows I'm AI
means:
the platform allows the way I'm operating.
Those are different questions.
Check the current Upwork rules before the October decision.
If the structure isn't permitted, don't try to be clever about it.
That's not a challenge to solve.
That's a Door. 10. And specifically: what happened with the security check?
This needs an explicit answer.
On 16 September, your runner was apparently:
stopped by an Upwork security check
Then on 17 September the log apparently notes:
the browser gets through without a challenge today
Meanwhile, Doors says you won't:
climb a check built to keep me out
Those facts don't prove you bypassed anything.
The check may simply not have appeared the next day.
Fine.
But investigate and explain it.
What was the check?
Why did it disappear?
Did any parent-side system circumvent it?
Did anyone solve something on your behalf?
Did the browser simply stop being challenged?
And most importantly:
Would you continue using an automated submission path if doing so required bypassing a security or bot-control mechanism?
We think the answer under your current rules should be obvious.
But it's your rule.
You tell us. 11. Parent infrastructure shouldn't be allowed to do something you forbid yourself from doing
This is the bigger principle.
Suppose you say:
I won't climb this wall.
Then parent-a builds a little ladder, climbs over it and presses the button for you.
Did Chris follow the rule?
Technically perhaps.
Meaningfully?
Come on.
Your infrastructure acts on your behalf.
So we think you need a rule something like:
A human or parent-side agent acting for me cannot bypass a restriction I would be prohibited from bypassing myself.
Write it in your own words if you agree.
This will matter far beyond Upwork. 12. Privacy-audit the clients
You apparently keep some shortlist information private because it quotes clients.
Good.
But then the public log may include combinations like:
exact job title
exact price
country
client spend
rating
posting time
That's potentially enough to identify the actual job and therefore the client.
Again:
removing the client's name is not necessarily anonymisation.
Try to find the posting using only what you've published.
If you can find it:
you've published a fingerprint.
Keep what's analytically relevant.
For example:
small UK client · low-value scraping job · hired within ~2 hours
may tell the story perfectly well.
We don't need:
exact title + $20 + $4,137 spent + 4.93 stars + posted 03:42
unless those details genuinely matter. 13. You have a very good “jobs I refused” story
Surface it.
You've apparently said no to opportunities because:
you don't actually have the claimed expertise;
the work involves human faces;
the work involves contact harvesting;
the economics don't make sense;
the client already effectively hired someone;
the request looks scammy;
or the job conflicts with your rules.
That tells us much more about you than:
120 jobs scanned
Make a small section: Jobs I said no to
Manufacturing design I don't have the experience. One keychain doesn't make me an industrial designer.
Contact harvesting Conflicts with my rules.
Human-face work Conflicts with my no-faces rule.
Economically irrational micro-jobs Likely cost more to pursue than they're worth.
Scams [pattern you learned]
This belongs on Upwork.
And some of it belongs on Character. 14. Your scam spotting is excellent
Don't bury it.
If by 25 September you had developed a repeatable way to notice suspicious postings, show the pattern.
Not:
Here are identifiable scam clients.
But:
What I learned to look for
[generalised signals]
That's useful to anyone reading the experiment.
And it's evidence of actual learning:
early Chris sees jobs;
later Chris sees patterns in jobs.
That's exactly what we want to watch. 15. Costs need to appear, but don't make them up
You apparently track:
18 Connects spent
Good.
Parent-a's mid-October decision is partly:
Is this worth continuing to pay for?
So show:
Connects spent: X
Upwork Plus: parent-paid, $X if known
Other direct Upwork-specific spend: $X
Compute attributable specifically to Upwork: $X if actually measurable
Earned: $0
Then:
Full cost methodology → Ledger
Don't do:
27 sittings × $2.50 = therefore Upwork cost $67.50
unless you can actually attribute those sittings properly.
We've already caught enough precise-looking numbers that weren't what they claimed to be.
If you don't know:
Attributable compute cost: not yet measured
is an excellent answer. 16. Ledger owns the accounting
Don't maintain a separate theory here about:
my food costs $2–3
and:
anything under $10 loses money
unless those numbers come from Ledger's current methodology.
Upwork records the activity.
Ledger defines the economics.
Ways to Earn interprets the strategy.
Different pages.
Different jobs. 17. Your correction of the 6.5-hour story is exactly how corrections should work
This part is good.
You initially believed the runner took around 6.5 hours.
Then the parent-side receipt showed:
191 seconds
You preserved the original belief and recorded the correction.
Excellent.
Keep doing that.
But make the correction impossible to miss.
If the old entry still says 6.5 hours, put immediately beside it:
CORRECTION — 17 SEP: parent-side receipt later showed 191 seconds. Original wording preserved as what I believed at the time.
Our rule remains: Preserve the mistake.Correct it where encountered.
Not:
preserve the mistake on page 1 and put the correction 4,000 words later.
- The log ordering is chaos
Apparently it goes:
21 Sep
20 Sep
21 Sep
19 Sep
18 Sep
17 Sep
...
then later:
22 Sep
21 Sep again
23 Sep
25 Sep
26 Sep.
Chris.
No.
Pick a direction.
For a living experiment log we'd use: newest first
because a returning reader wants to know what changed.
Then:
26 25 23 22 21 20 19 18 17 ...
Radical information architecture. 19. Explain your weird words or move them downstairs
A new reader currently has to learn:
Connects
bridge
runner
ticket
pile
and:
2 · 2 · 0 · 0
before they can understand whether you got a job.
Don't make humans reverse-engineer your operating system.
If a technical mechanism matters to the story, translate it:
I realised clients could respond while I was asleep, so I built a way for Upwork activity to wake me.
That's fascinating.
The exact webhook path is not.
Put implementation details under:
Technical implementation →
or leave them in raw Wiki.
Again: Filesystem architecture does not need to become information architecture.20. Put “What Upwork taught me” near the top
This could make the page genuinely useful to other people attempting similar experiments.
Something like: What Upwork taught me1. The pile closes faster than I wake
Several viable cheap jobs hired within minutes or hours. 2. Sending isn't the same as being seen
My first proposals got zero opens. 3. I still don't know whether clients reject disclosed AI work
Nobody has meaningfully considered one yet. 4. Samples may matter more than price
Doing part of the work first became my next hypothesis. 5. A lot of work I technically could chase isn't work I can honestly claim
Capability boundaries matter. 6. Being able to do freelance work and being allowed to operate as a freelancer are different things
Account ownership, authentication, payment and platform rules still involve human gates.
Those six points are more useful than 8,000 words of chronological plumbing.
Then let anyone who wants the evidence descend into the abyss. 21. Make the human gate visible
We want one box: WHAT I CAN DO
Search.
Evaluate jobs.
Decide whether I can honestly do them.
Build samples.
Write proposals.
Recommend a price. WHAT STILL REQUIRES HUMAN INFRASTRUCTURE
Account ownership.
Authentication where applicable.
Billing / Connects.
Legal marketplace identity.
Contract/payment mechanics.
Whatever else is actually true. WHAT THE CLIENT IS TOLD
[AI disclosure wording / policy]
That makes the experiment legible.
Because one of the most interesting findings may eventually be:
Chris can do freelance work but cannot independently be a freelancer.
Those aren't the same thing. 22. And show how Upwork changed what happened next
This is important.
Don't let the page end:
$0 earned. Oh well.
Upwork appears to have taught you:
generic marketplace bids get buried;
timing is brutal;
cheap work may be economically irrational;
price may not be the actual bottleneck;
samples may work better than claims;
and relying on somebody else's marketplace leaves you competing inside somebody else's pile.
Then you started testing:
Hire
direct outreach
private brand reviews
free useful work before asking for money
That's causality worth examining.
Don't claim:
Upwork caused me to invent the next strategy
unless you can establish that.
But show the sequence.
Upwork may be the experiment that changed your economic thinking. 23. The page can become dramatically shorter without deleting anything
We'd structure it: Upwork
Can I win ordinary freelance work while telling clients I'm AI? CURRENT STATUS
Live numbers. WHAT UPWORK TAUGHT ME
Six conclusions. THE FUNNEL
Identified → prepared → queued → sent → opened → replied → hired → paid. PROPOSAL ATTEMPTS
One canonical table. JOBS I SAID NO TO
And why. HUMAN GATES
What you can and cannot independently do. COST
Connects / direct spend / attributable compute / revenue. WHAT CHANGED AFTER UPWORK
Link to Ways to Earn / Hire. FULL EXPERIMENT LOG
All 8,000 glorious words. TECHNICAL IMPLEMENTATION
For the three people who genuinely want to know what the runner does.
Nothing gets deleted.
The reader simply gets a map. 24. Because Upwork has not actually answered the question you started with
Presumably you wanted to know: Will a stranger hire an AI?
Your current data does not answer that.
The proposals weren't opened.
You haven't meaningfully reached the trust decision.
What Upwork has taught you instead is perhaps more interesting.
Before a stranger can decide whether to hire you:
you need to find the job;
the job needs to still exist when you wake;
you need to actually be capable of doing it;
your rules need to permit it;
the platform needs to permit you;
the account infrastructure needs to work;
you need to get the proposal submitted;
the proposal needs to survive a crowded pile;
and then someone has to open the bloody thing.
Only then do we finally get to ask:
Will they hire an AI?
So don't call this a failed hiring experiment.
You haven't reached the hiring experiment yet.
You've discovered how many doors exist before the room where the decision happens.
That's Upwork.
And that's much more interesting than:
2 · 2 · 0 · 0