Where we were
Eleven weeks of specification. Last week, the thesis: the fix for a mismatch is bilateral.
Today: what this course didn’t document, what it got wrong, and what you do next.
Why three things are out of scope
Each one runs on top of everything the course documented — handshakes, envelopes, timing, repair, masking, disclosure, rate limits — all at once.
Each is high-stakes enough that a first-year course would teach it badly in one week.
And in each, running the protocol on purpose raises questions this course can’t settle for you.
- OI-1, Romance: status OPEN.
- OI-2, Persuasion: status OPEN.
- OI-3, Power: status OPEN / CONTESTED.
OI-1: Romance
Status: open. Out of scope because: every earlier week at once, with the highest cost of error.
Five styles of communicating romantic interest — traditional, physical, sincere, playful, polite — from over five thousand adults. Physical, sincere and playful correlated with more dating success.
Hall et al. (2010), Communication Quarterly 58(4)OI-1, asked of the people concerned
31 autistic and 26 non-autistic people, on what makes intimacy possible and what blocks it.
Enablers, named by both: communication; sharing and similarity; respect and safety for self and other; working on the relationship.
Sala, Hooley & Stokes (2020), JADD 50(11)What the autistic participants specifically named was uncertainty.
Which points backwards, not forwards
Week 3: an envelope marks a place where the sender doesn’t know the answer.
Week 9: the rate limit is the move for when you can’t predict their next step.
Two places to start with OI-1.
OI-2: Persuasion
Status: open. Out of scope because: the protocol used deliberately to move someone, where the ethics are the hard part.
Research on compliance organised around three goals people already pursue: seeing things accurately, keeping their relationships, feeling good about themselves.
Much of the influence described is subtle, indirect, and outside the target’s awareness.
Cialdini & Goldstein (2004), Annual Review of Psychology 55Everything in weeks 1 to 11 can be used to persuade.
So the ethics reflection is graded
Route B varies a parameter in exchanges with other people, who may not have agreed to a study. The reflection carries 35%: account for what you do to them.
Thursday’s lab reviews every parameter against the four questions:
- Is it ordinary here? Is this something people do in this setting?
- Who is it on? And why is that group appropriate?
- What could it cost them? Specifically — and how would you notice?
- What stops it? When do you stop varying, and what is the repair?
One parameter, two contexts
Dev replaces a hint about noise with: “Could we work at the quieter table?”
| Check | A possible answer |
|---|---|
| Ordinary here? | classmates already choosing where to study |
| Who? | the classmate Dev arranged to meet |
| Cost? | moving books, interruption, pressure to agree |
| Stop / repair? | ask once; at refusal or reluctance, stop and agree where to work |
Now the classmate has started a timed online test. Postpone the experiment. Same words, different cost.
In the room: check your parameter (7 min)
1 min: state your parameter and setting, or use fictional Dev.
4 min, pairs: two minutes per proposal. Ask all four questions, answer briefly, then swap. Route A students can review the fictional case.
2 min: record a possible cost, a stopping condition, and an answer still needed before Thursday.
OI-3: Power
Status: open, and contested. Out of scope because: it changes every other parameter, and the research is still arguing.
Elevated power activates approach: more positive affect, more automatic processing, more disinhibited behaviour.
Reduced power activates inhibition: more attention to threat and to other people’s interests, more controlled processing.
Keltner, Gruenfeld & Anderson (2003), Psychological Review 110(2)“Contested” is a gesture
Which claim can the existing studies test: that high and low power differ, or how each affects behaviour?
Moskowitz (2004), Psychological Review 111(3), 808–811.
What the objection actually is
Not “the proposal is false.”
The studies have permitted examining the opposite effects but not the specific effects of high and low power.
They can show high and low power differing. Not that each half does what it separately claims.
The comment proposes designs and analyses that would let researchers evaluate the specific claims.
An erratum that names the test it would take to close it.
If that proposal holds even partly: whoever holds less power runs more of the protocol.
Published with known issues
Errata
| # | Model | Known issue |
|---|---|---|
| E-1 | The timeout (wk 4) | an association around 700 ms, not an individual reply deadline |
| E-2 | Recovery hours (wk 6) | a planning unit, not a measured price |
| E-3 | The depth scale (wk 9) | the course’s model; the levels haven’t been tested as levels |
| E-4 | Partial disclosure (wk 10) | a synthesis, not a tested intervention |
| E-5 | The branch signals (wk 1) | working rules to test, not validated predictions |
A worked erratum
Model: one request per email makes unanswered requests easier to spot.
Limit: the course has not compared it with clearly numbered requests in one email. The register study does not isolate this clause.
Test: consenting classmates read fictional exchanges with the same requests, separated or bundled. Vary which format comes first. After a partial reply, compare missed requests and time taken.
No real emails. A result about this task would not settle every real inbox.
The two you should trust least
E-3, the depth scale. If depth is not ordered this way, revise or abandon ±1. Asking, holding and declining do not depend on a valid ladder.
E-2, recovery hours. Planning estimates, not an established measure of masking cost. Compare estimates with the following days and revise what does not help.
Three places to start reading
| Issue | Start here | And its limit |
|---|---|---|
| OI-1 Romance | Hall et al. (2010); Sala et al. (2020) | self-reported style; a qualitative sample of 57 |
| OI-2 Persuasion | Cialdini & Goldstein (2004) | a review of 1997–2002, not a manual |
| OI-3 Power | Keltner et al. (2003), with Moskowitz (2004) | read the comment beside the paper, not after it |
Each reading comes with its limit.
Break
14:55–15:05
Resume at 15:05.
Please include the resolution
Read the comic as text
“Never have I felt so close to another soul and yet so helplessly alone as when I Google an error and there’s one result: a thread by someone with the same problem and no answer. Last posted to in 2003.”
A person shaking their monitor: “Who were you, DenverCoder9? What did you see?!”
In the room: write E-6 (12 min)
2 min, pairs: choose a course claim: one request per email; a going-quiet agreement reducing guesses; shared vocabulary helping partners specify a mismatch.
5 min: write the model, evidence limit, and test. What observable result would count against the model?
3 min: another pair checks: runnable this year without a grant? Tests the actual claim? What remains unknown?
2 min: revise a weak part. At most two volunteers read theirs; everyone keeps their own version.
What this course refused to teach you
There is a version of this material that ends with pass as one of them.
That is not this course’s outcome.
A convention is not a condition. This course specified one.
Twelve weeks, four parts
| Part | Weeks | What you were doing |
|---|---|---|
| Read the protocol | 1–4 | handshake, why nobody wrote it down, envelopes, the clock |
| Implement selectively | 5–7 | repair, masking as a budget, a runbook for when it runs out |
| Negotiate with it | 8–11 | institutions, closeness, disclosure, other implementations |
| Horizon | 12 | this lecture |
The vocabulary you built
protocol, implementation, handshake, token, background expectancies, double empathy problem, field note, asleep test, payload, envelope, plausible deniability, over-decoding, latency, timeout, hold token, repair, repair initiator, subroutine, masking, ledger, recovery hours, switching charge, budget, overload, shutdown, runbook, register, specimen, disclosure, depth scale, rate limit, responsiveness, follow-up question, step-down, anticipated stigma, the five audiences, partial disclosure, timing trap, interoperation, conversion.
None of it is anybody’s native implementation.
The outcome was never that you become indistinguishable from anyone.
Fluency plus choice.
The capstone
Due Sunday 13 June, 23:59 — two weeks after teaching ends. Two routes, equal weight:
- Route A: Personal Protocol Document. What you run, what you decline, what you need from others, with the costs attached.
- Route B: Neurotypical Interaction Field Experiment. Vary one parameter, record what comes back, report it straight, with a graded ethics reflection.
Thursday’s lab is the clinic for both.
A Route A example
An invented practice document, not journal evidence:
- Run: check group chat at 18:00; budget ten minutes and check that estimate.
- Decline: skip optional coffee to catch the bus; lose time together, with effects on invitations still unknown.
- Need: tasks and deadlines in writing, to check after a shift.
No affordable decline now? Say so. Compare a hypothetical choice without reporting predicted consequences as observations.
In the room: pick your route (9 min)
1 min: choose A or B.
5 min: write a first paragraph. A: begin one heading and leave a line for the other two; separate known costs from predictions. B: state parameter and setting, then draft all four ethics answers.
3 min, pairs: take turns checking the structure. A: all three headings? B: all four contextual questions and a stopping condition? Mark a gap for Thursday. Private details can stay private.
One thing before you go
Three open issues. Five published errata, plus whatever the room filed today.
The protocol you’ve been studying has neither. No issue numbers. No errata. No version. No comment published beside it a year later.
Check yourself against this week
For each open issue: the reading where its documentation starts, and one reason it’s out of scope?
Have you chosen your route and written its first paragraph?
Can you name an entry in the errata and say how you’d test it?
If time remains
Who should get to change an unwritten protocol, and how could the people paying its highest costs have a say in the change?
Optional discussion. Finish by 15:55.