# Trace patterns and query templates Tribal knowledge leaves residue, not statements. These are the seven residues worth hunting, with phrasings that actually surface them. A note on how to use these: the templates are conversational fragments, not keywords. People don't write "exception to the standard invoicing policy" — they write "we usually don't do this but." Search the way people type, including the typos and the swearing. Most search backends support quoted phrases and OR; where semantic search is available, a natural-language question often beats a keyword string. Try both. Substitute your own domain nouns for `` throughout. --- ## 1. The repeated question The same question asked by different people, months apart. Means no canonical answer exists — and the answers given each time may not even match. ``` "how do we" "how does" "work" "who owns" "do we have a" "is there a doc" OR "is this documented" OR "where do I find" "dumb question" OR "quick question" OR "sorry if this is obvious" "asked this before" OR "asking again" OR "reposting" ``` The tell is repetition across time and people. One person asking is a question; four people asking over eight months is a documentation gap with a dollar value. Note every distinct asker and date — that count is the finding. Follow-up: for each instance, did anyone answer? Compare the answers. Divergent answers to the same question is the strongest possible evidence that the org has no shared understanding. ## 2. The exception The written rule is a fiction; the exceptions are the real policy. ``` "we usually" OR "normally we" OR "typically we" "but for" "we" "one-off" OR "special case" OR "carve out" OR "carveout" "except" "make an exception" "this time only" OR "just this once" "grandfathered" ``` When you find one, immediately search for the entity it applies to (the client name, the team, the product) to see whether the "one-off" happened repeatedly. A one-off that recurs is an unwritten rule. ## 3. The reversal A decision made, then undone. The reasoning behind the undo is almost never recorded anywhere official, and it's usually the thing worth knowing. ``` "we decided" "actually let's" OR "scratch that" OR "change of plans" "reverting" OR "rolled back" OR "we're going back to" "we tried that" OR "we did that before" "that didn't work" OR "didn't pan out" "no longer" OR "we stopped" OR "we killed" "as of" "we're" ``` Search the original decision and the reversal separately, then reconstruct the sequence with dates. The gap between them — how long the decision survived — is itself informative. ## 4. The name Knowledge concentrated in one human. This is what a bus factor looks like before it becomes a crisis. ``` "ask" OR "check with" "would know" OR "knows this" "only" "has" OR "only" "can" "waiting on" "who owns" OR "who's the owner of" OR "who handles" "@" — high mention volume on one topic is itself the signal ``` Count mentions per person per topic. One name appearing across many otherwise-unrelated threads on the same subject is a single point of failure. Worth flagging even when it wasn't what you were asked to find — especially if that person is leaving, on leave, or was recently reorganized. ## 5. The workaround Undocumented process doing real load-bearing work. ``` "workaround" OR "hack" OR "band-aid" OR "bandaid" "manually" OR "by hand" OR "manual process" "I have a script" OR "there's a script" OR "run the script" "my spreadsheet" OR "the tracker" OR "the sheet" "temporary" OR "for now" OR "until we fix" "copy paste" OR "just copy" "don't ask" OR "long story" ``` "Temporary" is the highest-yield word in this list. Search it, then check the date — a temporary fix from 14 months ago is permanent infrastructure that nobody has documented and everybody depends on. ## 6. The warning Scar tissue. Points at an incident whose lesson never made it into a doc. ``` "careful" OR "be careful" OR "heads up" "last time we" OR "the last time this happened" "don't" OR "never" "learned that the hard way" "this bit us" OR "burned us" OR "bit us last" "make sure you" OR "whatever you do" "remember when" ``` Every warning implies a prior incident. Search for the incident itself once you have the shape of it — the postmortem may exist even where the lesson never propagated. If it doesn't exist, that absence is a finding. ## 7. The stale artifact A doc everyone links to that the live record quietly contradicts. Approach differs from the others — it's a comparison, not a search: 1. Find the canonical doc (wiki page, README, policy PDF, pinned message). Note its last-updated date. 2. Search the chat/ticket record for the same subject *after* that date. 3. Compare. Anything the record says that the doc doesn't — or contradicts — is the finding. ``` "the doc says" OR "according to the doc" OR "per the wiki" "that's out of date" OR "outdated" OR "stale" OR "no longer accurate" "ignore the doc" OR "the doc is wrong" "we should update" OR "someone should update" ``` "The doc is wrong" said out loud in a channel, with nobody fixing the doc, is a perfect specimen: it proves the gap, dates it, and names someone who knew. --- ## Cross-source technique **Follow vocabulary.** The first pass teaches you the internal language — project codenames, client nicknames, the acronym for the thing. Note every unfamiliar term and re-search with it. You cannot write these queries at the start; they're the main reason a second pass beats a longer first pass. **Follow people.** Once a name recurs, search that person's messages on the topic directly. People are more consistent than channels. **Follow IDs.** Ticket numbers, PR numbers, invoice numbers, deal IDs, incident numbers. These are exact-match handles that cross system boundaries — a ticket ID mentioned in chat links the conversation to the record, and each usually has content the other lacks. **Follow the calendar.** Recurring events, quarter boundaries, launch dates, and someone's last day cluster decisions around them. If a change happened, something on the calendar probably prompted it. **Widen before you conclude.** A finding based on one thread is a hypothesis. Search the same claim two more ways before you write it down — different phrasing, different source. If you can't corroborate, ship it as low-confidence with the single source shown, not as fact.