Relay log methodology

Innovation as a Relay Race sets the ontology. An answer is a stopping point. A question is a transfer. A baton that ends the race is a trophy, not a baton. The quality of the question matters more than the authority of the asker — a child, a stranger, or a machine can hand one off.

• What question, once answered, immediately produces a better question?
• How do we know a question is worth passing on?
• What makes a question die in transit?
• Which questions are we still answering badly, generation after generation?
• What is the difference between a question that opens a field and one that closes it?

Closing questions demand a verdict. Opening questions demand a runner.

Each entry carries the same fields, whatever the venue.

Number and date

Sequential id and the date the handoff was observed. Entries are appended, not rewritten.

Venue

X, Substack, internal essay, correspondence, or working note. Venue is recorded because it changes what silence means.

Question as asked

Verbatim. Not a cleaned-up version written after the reply came back.

What happened next

One or two plain sentences. Where there is a verbatim reply or a full exchange, the turns are quoted rather than summarized.

Status

Open (awaiting a runner), passed (picked up as asked), mutated (came back larger), dropped (stopped in transit).

Baton passed

The next question, written so a stranger can pick it up without knowing the history.

Claim status inside the write-up

Established, reported, interpretive, contested — assigned per claim, not per essay.

Falsifier

What observation would show the entry's own reading was wrong. Often turned on the essay it produced.

Resolution

Added only when an open handoff is later closed, with the date and who closed it.

The counters at the top of the log — handoffs, status split, carry rate, verbatim replies quoted, new questions produced — are derived from those coded fields and nothing else. They are computed from the entries at page load, so a quoted readout is only accurate as of the moment it was copied.

Question Trials is the public statistical companion. A case fixes four things in advance: the opening question, the predicted drift, the follow-up if the first stalls, and the codebook used to score replies.

The Question Tracker is the pre-registration bench. A question is logged before any reply arrives, with its predicted drift, so the prediction cannot be edited afterward. The unit of analysis is one question posted in reply to one specific post, with a fixed window — seven days on X — after which silence stays Open rather than being deleted. One logged reply is an anecdote. The same reply becomes evidence when unit, prediction, coding, and window are fixed first.

The Interface Discipline Test is adjacent rather than core. It pairs naive against hardened prompts on planted false premises and scores them 0–3 with Wilson intervals. It tests the prompting discipline the relay method assumes, not the handoff itself.

The log inherits its turn rules from what a Socratic interchange actually is, not from social-media etiquette. Four moves, in any order, by either party:

1. Elicitation — Get the claim into a sentence that can be tested.
2. Status assignment — Established, working, speculative — later also contested and interpretive.
3. Stress — Find the edge where the claim stops holding.
4. Reconstruction — Restate with new scope, or retire it. Retirement is a successful exchange, not a loss.

A valid turn returns three things: the answer, the reasoning, and the point at which the reasoning would break. Applied to an organization or an agent, the same grammar demands four interface properties: legible status, a visible intermediate chain, re-entry at any step, and a revisable record. Inner direction is operationalized narrowly — can the person name, unprompted, what would change their mind.

Named failure modes: sycophantic decay (smoothness rising while novelty falls), verdict-first replies, fear-only messages, and reward before work. The wrong sort of communication substitutes the speaker's decision for the listener's. The right sort hands the step back.

• Hand coding by one reader until the scheme stabilizes. Replication is deferred, not claimed.
• Carry-rate percentages describe engaged handoffs in this ledger. They are not a measure of influence, reach, or persuasion.
• No interpretation of intent. The log records what happened to the question, not what anyone meant.
• Compression is a distortion. A long emergent exchange scored down to a number flattens the object under study, and a score becomes the next thing to game.
• Checking is non-transferable. A specification can be handed over; the act of checking cannot.
• The method refuses pre-ordained use. Publishing an open question without dictating what it is for is the point, not an omission.

Established — the fields, codes, statuses, and windows described here are the ones actually used to build the ledger.

Interpretive — the claim that coding what a reply does to a question measures anything worth measuring. That is the bet the benches exist to test.

Falsifier — if a second reader, coding the same replies from this codebook without seeing the first coding, produces a materially different status split, the scheme is not yet stable enough to report aggregate numbers, and the counters should come down until it is.