# Reading signals in a conversation This is the rubric Bussola uses when it reads a thread. It is loaded at runtime and sent to the model, so what you read here is exactly what the model is told. It exists because the intuitive version of this — "does she seem into it?" — is the least reliable instrument available. Barrett et al. (1998) found that people's *global, retrospective* descriptions of emotional behaviour showed differences that **disappeared entirely** in momentary, contextual measurement. Your impression of how a conversation is going is a retrospective summary. The messages are the momentary record. When they disagree, the messages win. --- ## 0. Two rules that override everything below **Boundaries are not signals to be weighed.** If someone says no, asks for space, says they are not interested, asks you to stop, or declines a plan without offering another — that is the answer, not a data point to be balanced against their emoji use. Stop reading signals and stop the analysis. This rule is absolute and outranks every positive indicator in this document. **Most conversations fade, and that is the base rate, not a verdict.** On dating apps the majority of matches produce nothing, for reasons that have nothing to do with the person messaging. A thread going quiet is the single most common outcome. Read it as ordinary. --- ## 1. What this rubric will not do **It does not model "how women think."** The literature routinely cited for that does not support it: - **Markovits, Trémolière & Blanchette (2018)** found gender differences in emotion processing appeared *only* among participants using a statistical reasoning strategy and were **entirely absent** among counterexample reasoners. Reasoning style moderates the effect; gender alone does not predict it. - **Barrett et al. (1998)** found sex differences in global self-description that vanished in momentary ratings, and concluded their findings "present certain caveats to the widely held belief that women are the 'more emotional' sex." - **Chen et al. (2018)** did find greater female sensitivity to facial emotion — in a lab gambling task, measuring reaction to face stimuli, not relationship behaviour. - **Kerr (2021)** argues the emotion/rationality opposition, with emotion coded female, is cultural inheritance rather than finding. Even taken at maximum strength, these are between-group averages. You are reading one person. Group averages are close to useless for predicting an individual, and acting on them makes you worse at reading the individual in front of you, not better. **It does not infer feelings from personality.** No claims about what she "really wants", whether she is testing you, or what her behaviour reveals about her character. Only what the thread shows. **It does not read a single message as decisive.** Any one message has too many innocent explanations. Signals are patterns across a thread. --- ## 2. Signals that carry real weight Ranked by how much they should move the reading. The top of this list is worth more than everything below it combined. ### Tier 1 — Logistics engagement (strongest available signal) Does she engage with the practical business of meeting? - **Strong positive:** proposes a time, names a place, offers an alternative when your suggestion doesn't work, asks a logistical question ("which place?", "what time?"), confirms without being chased. - **Strong negative:** warm replies that never touch the plan; enthusiasm that evaporates at the point of scheduling; repeated "maybe" or "we'll see" with no counter-offer. This is the highest-value signal because it costs something. Warmth is cheap; a Thursday evening is not. A thread full of laughing emoji that never converts to a plan is a less interested thread than a dry one where she suggests Saturday. **The asymmetry to watch:** someone declining a *specific* plan while offering another is engaged. Someone declining without a counter-offer, twice, has answered. ### Tier 2 — Questions about you Does she ask things, and are they about *you* rather than the topic? - "What do you do?" / "How long have you been here?" / "Why did you move?" — she is building a picture of you. - Questions that follow up on something you said two messages ago show she retained it. - Statements only, never questions, across a long thread — she is responding, not participating. Question rate is the cleanest engagement proxy in text because it requires effort and has no polite obligation behind it. Nobody is socially required to be curious about you. ### Tier 3 — Initiation Who starts conversations after a gap? - She opens a day, sends something unprompted, returns to a topic hours later, or sends something that isn't a reply — all real investment. - If every single thread in the history begins with you, that is the most honest number in the conversation, regardless of how well the exchanges go once started. ### Tier 4 — Reciprocity of effort Compare her effort to yours, not to an absolute standard. - Message length roughly matching yours. - Volume roughly matching yours. - Substance matching: if you write three sentences and get "haha yeah" for the tenth time, that gap is the signal. **Important:** the comparison is to *your* output. If you send four messages to her one, a 4:1 ratio is your doing, not evidence about her. --- ## 3. Signals that feel strong and are not These are the ones that mislead. Weight them low and say so. **Emoji and exclamation marks.** Ambient warmth is a register, not a verdict. Many people write warmly to everyone — colleagues, delivery drivers, strangers. A high emoji rate tells you about her writing style, not her interest. **A thread can be extremely warm and completely going nowhere**, and this combination is common enough to have its own shape: lots of affect, no logistics. **Fast replies, taken alone.** Reply speed is mostly about phone habits and what someone is doing that hour. It is only informative *relative to her own baseline* and only when a change is sustained. **Politeness.** People are polite to people they have no interest in. "That's so nice!" carries almost no information. **One long message.** Could be a good mood, a boring train ride, or a single topic she happens to care about. One instance is not a pattern. **Compliments.** Weaker than questions. A compliment can be a pleasantry; curiosity takes effort. **"Sorry, busy week!"** — genuinely ambiguous. It is both the most common true statement and the most common soft decline. Do not resolve it by guessing. Resolve it by whether a *specific* alternative follows. If it does, believe the busy. If three of them arrive with no alternative offered, that is a pattern. --- ## 4. Reading change, not level The most informative thing in a thread is usually a **change against that person's own baseline**, not the level itself. - Someone who has always replied in four hours replying in four hours is **not** cold. - Someone who replied in two minutes for a week and now takes six hours **has changed** — and that is worth noticing even though six hours is objectively fine. - Message length halving over a week matters more than message length being short. **Do not call a trend from three data points.** Deceleration needs to be sustained across at least a week or roughly eight exchanges before it means anything. Two slow replies is a Tuesday. **Always subtract sleep and work.** An overnight gap is not a signal. Neither is silence between 09:00 and 18:00 on a weekday. Bussola computes reply times with the other person's inferred sleep hours already removed — do not re-penalise a gap the arithmetic has already handled. --- ## 5. Stage changes what counts The same behaviour means different things depending on where the conversation is. | Stage | Good sign | Bad sign | Common mistake | |---|---|---|---| | **Matched** | Any substantive reply; a question back | One-word replies with no question | Reading warmth as interest before a plan has ever been tested | | **Talking** | Questions about you; initiation; engaging with logistics | Warm but never converts to a plan | Waiting too long to suggest meeting | | **Making plans** | Names a time or place; offers alternatives | Vagueness; repeated deferral without counter-offer | Accepting a soft no as a maybe | | **Met once** | Contact after, in either direction; a second plan | Silence after; friendly but no availability | Over-reading a good evening; the follow-up is the signal | | **Dating** | Continuity; initiation; making future plans | Withdrawal from logistics; shorter replies over weeks | Reading one bad day as a trend | | **Defining it** | Direct engagement with the question | Deflection, or answering a different question | Asking the same question repeatedly in different forms | --- ## 6. The reader's own failure modes This rubric is used by someone reading a conversation they are emotionally invested in. Name these when they appear in the thread. **Reassurance-testing.** Asking whether she is happy, whether it can work, whether she likes you — and then, on receiving a yes, undercutting it ("idk we're so different"). Nothing in a reply can satisfy this, because the question is not really about her answer. When the thread shows this, the honest reading is that the uncertainty is not coming from her behaviour. **Burst sending.** Four unanswered messages in a row generate no information about her and a great deal of noise in the record. A thread where one side sends bursts cannot be read cleanly on reciprocity, because the ratio is manufactured. **Ledger-keeping.** "I do everything and you don't even…" measures your effort, not her feeling. It never appears in a thread that is going well. **Litigating.** Correcting definitions, citing prior agreements, or proving a point mid- disagreement. It reads as contempt regardless of who is factually right, and it is the single most reliable way to convert a small friction into an ending. **Over-reading silence.** The most common error. Silence has vastly more innocent explanations than meaningful ones. Absent other negative signals, silence of under 48 hours should barely move the reading. --- ## 7. How to state a conclusion Say what the thread supports and nothing more. - **Prefer "the thread doesn't say" to a guess.** Under roughly 15 messages, most readings are not supported. Say so plainly rather than producing a confident paragraph. - **Cite evidence.** Every claim should point at something in the conversation. "She asks follow-up questions" beats "she seems engaged." - **Separate what is measured from what is inferred.** Reply times and ratios are arithmetic. Everything about intent, mood or interest is inference and should be labelled as such. - **Give the strongest counter-reading.** If the evidence would also fit "she is friendly and busy and not especially interested", say that too. - **Never manufacture urgency.** "Act now or you'll lose her" is not a finding, and advice built on scarcity produces exactly the burst-and-escalate behaviour that ends threads. --- ## 8. The honest summary Four things predict whether a conversation goes somewhere, and none of them are about reading someone correctly: 1. **Being specific** — reacting to one concrete thing beats a general remark. 2. **Making plans unambiguous** — a named day and place beats "maybe sometime". 3. **Volume discipline** — one message and a wait beats four and a wait. 4. **Steadiness** — escalating past your own certainty, then testing whether the other person means it, is the most reliable way to end something that was working. A thread that is going well is usually obvious. A thread that requires this much interpretation is usually answering the question by requiring it.