{ "_about": "A learning INFERS: it says something the transcript does not literally say - a generalisation, a causal reading, a ranking, a comparison, an application to a question. Learnings are NOT produced at ingestion. They are produced when someone asks a question, and stored together with that question, so every learning answers 'why does this exist and what was it for'. Evidence that supports, challenges, or merely contextualises the learning is listed separately - never merged into a single 'sources' list.", "_field_guide": { "kind": "interpretation | ranking | risk | gate_analysis | change_over_time | hypothesis - what sort of inference this is", "confidence": "Extractor's own confidence in the inference GIVEN the evidence. Not a claim about the world.", "supported_by / challenged_by / context_from": "Evidence IDs. A learning with nothing in challenged_by should be viewed with suspicion, not relief.", "caveats": "Everything that would make a careful reader trust this less.", "generalises_beyond_transcript": "false unless there is evidence from more than one source. One interview never licenses a population claim." }, "questions": [ { "id": "q_001", "asked_on": "2026-09-18", "asked_by": "researcher", "question": "Is manual column remapping a problem worth solving for this team?", "scope": ["tx_001"], "learnings": ["lr_001", "lr_002"] }, { "id": "q_002", "asked_on": "2026-09-18", "asked_by": "researcher", "question": "What would it actually take for this team to adopt a remapping tool?", "scope": ["tx_001"], "learnings": ["lr_003"] }, { "id": "q_003", "asked_on": "2026-09-19", "asked_by": "engineering", "question": "Did the previous automation attempt fail for a reason a new tool would share?", "scope": ["tx_001"], "learnings": ["lr_004"] }, { "id": "q_004", "asked_on": "2026-09-19", "asked_by": "researcher", "question": "Has the report timeliness problem changed over time?", "scope": ["tx_001"], "learnings": ["lr_005"] } ], "learnings": [ { "id": "lr_001", "question_id": "q_001", "kind": "interpretation", "statement": "Remapping is a real, recurring cost (roughly 6-10 team-hours per week, every week), but the speaker does not experience it as the main problem. The felt problem is downstream: the report reaching directors late or incomplete.", "what_this_adds": "The transcript never ranks 'effort' against 'timeliness' in these terms. This learning combines the effort figures with the speaker's explicit re-ranking in turn 14 and the correction in turn 16.", "supported_by": ["ev_005", "ev_006", "ev_011", "ev_015", "ev_016", "ev_018"], "challenged_by": ["ev_011"], "challenged_note": "ev_011 flag: the speaker never says remapping accounts for ALL six hours. The effort attributable to remapping specifically is unknown.", "context_from": ["ev_002", "ev_004"], "confidence": "medium", "caveats": [ "Single interview, single speaker.", "Severity answer (ev_015) was given in response to a leading question; the pushback strengthens it but the framing was still supplied by the interviewer.", "ev_016 (directors deciding on stale numbers) is the speaker's account of third parties; basis unknown." ], "generalises_beyond_transcript": false }, { "id": "lr_002", "question_id": "q_001", "kind": "hypothesis", "statement": "A remapping tool would be valued by this team for its effect on report completeness and timeliness more than for hours saved. Pitching it as 'save six hours a week' would target the thing the speaker said matters less.", "what_this_adds": "This is a recommendation-shaped inference. Nothing in the transcript is about how to pitch anything.", "supported_by": ["ev_015", "ev_016", "ev_018", "ev_028"], "challenged_by": [], "challenged_note": "No evidence against it in this transcript - which is itself a caveat. The interviewer never asked 'what would you want a tool to do for you'.", "context_from": ["ev_020"], "confidence": "low", "caveats": [ "Depends on lr_001 holding.", "Whether a remap tool would actually reduce 'gaps' (ev_018) is unknown, because what the gaps are is unknown (see probes.json topics_not_probed)." ], "generalises_beyond_transcript": false }, { "id": "lr_003", "question_id": "q_002", "kind": "gate_analysis", "statement": "Adoption would have to pass through at least three separate gates that the speaker described independently: (1) price - under ~5k/yr the speaker can approve alone, above it the director must; (2) money - no budget line is known to exist this year regardless of who approves; (3) security - if the tool touches the warehouse system, a review of unknown but reportedly multi-week duration. The speaker has expressed interest and conditional openness to a pilot, but no commitment at any gate.", "what_this_adds": "The transcript gives these facts across turns 18-22 without connecting them. The learning is the assembly into a dependency chain and the explicit observation that none of them is satisfied.", "supported_by": ["ev_020", "ev_021", "ev_022", "ev_023", "ev_024", "ev_025", "ev_026", "ev_027"], "challenged_by": ["ev_024"], "challenged_note": "Gate (3) is conditional: ev_024 only applies if the tool connects to the warehouse system. A tool operating on already-exported files might not trigger it. This is a design question, not a settled constraint.", "context_from": [], "confidence": "medium", "caveats": [ "ev_023 is hedged ('that I know of'). Budget may exist outside the speaker's visibility.", "ev_025 is secondhand and a single data point.", "The sample-exports date (ev_027) carries three stacked hedges and must not be treated as a milestone." ], "generalises_beyond_transcript": false }, { "id": "lr_004", "question_id": "q_003", "kind": "risk", "statement": "The previous script failed on input size, specifically on the fleet export. Any new tool has the same failure mode as an open risk until the fleet export's actual size is known and tested. The threshold is unknown; the speaker did not say and was not asked.", "what_this_adds": "Reframes a past event as a forward-looking test criterion. The transcript describes what happened; it does not say 'test this'.", "supported_by": ["ev_012", "ev_013", "ev_014"], "challenged_by": [], "challenged_note": "Nothing in the transcript says the failure was inherent to file size rather than to that script's implementation. 'Choked' is the speaker's word; the actual failure (memory? timeout? crash?) is unknown.", "context_from": ["ev_007"], "confidence": "medium", "caveats": [ "One attempt, one script, described in one sentence.", "The team may have dropped it for reasons beyond the size failure that were not mentioned." ], "generalises_beyond_transcript": false, "open_questions_raised": [ "What is the typical and maximum size of the fleet export?", "What did the script actually do when it 'choked'?" ] }, { "id": "lr_005", "question_id": "q_004", "kind": "change_over_time", "statement": "Before roughly spring 2026 the report was late about once a month. A hard Friday 3 p.m. cutoff policy then converted lateness into incompleteness: the report now always ships on time but sometimes with gaps. The underlying cause (manual, effort-heavy preparation) was not addressed by the policy; its symptom was moved.", "what_this_adds": "The 'symptom moved, cause unchanged' reading is an inference. The speaker described the before and after; they did not characterise the policy as a non-solution.", "supported_by": ["ev_017", "ev_018", "ev_019"], "challenged_by": [], "challenged_note": "The speaker may regard the policy as an improvement; they did not say either way.", "context_from": ["ev_016", "ev_005", "ev_011"], "confidence": "medium", "caveats": [ "'Gaps' is undefined. If gaps are trivial, the policy may have genuinely solved the problem the directors cared about.", "Timeline for the policy change is approximate (ev_018 flag).", "ev_017 is superseded and is cited here only as the origin of the corrected figure in ev_019." ], "generalises_beyond_transcript": false, "tracking_note": "If this person or team is interviewed again, ev_018 is the baseline to compare against: has the cutoff held, and what do gaps consist of now?" } ], "_cross_transcript_placeholder": { "note": "When more transcripts exist, a question's scope lists several transcript IDs and a learning may cite evidence from each. Only then can generalises_beyond_transcript become true, and only if the supporting evidence comes from independent sources - the same person saying the same thing twice is one source.", "example_shape": { "id": "q_0xx", "question": "Do teams that produce recurring reports commonly have a manual reconciliation step?", "scope": ["tx_001", "tx_002", "tx_003"], "learnings": ["lr_0xx"] } } }