- Joined
- Jul 11, 2008
- Messages
- 1,971
- Reaction score
- 185
- Points
- 63
The gap is the note-taking. Or rather the absence of it, or the version of it that produces bullet points like "pressing looked good second half" and "keeper nervous on crosses" - observations that are real and might be analytically significant but exist in a form that can't be synthesised, compared across multiple matches, or connected to a specific market implication without substantial additional interpretive work that almost never happens.
This article is about fixing that gap. Not by watching differently - you watch however you watch - but by building a note-taking structure during watching that makes the notes synthesisable afterward. The structure that allows you to feed three or four match observations into an LLM and receive back something with the same analytical architecture as the opposition research briefs described earlier in this series, rather than a paragraph that strings your disconnected observations together in a different order and calls it a summary.
The result, done properly, is a research workflow that makes your watching hours produce compounding analytical value rather than impressions that fade by the following week.
Why Unstructured Notes Don't Synthesise
The failure mode is worth describing precisely because it's almost universal among bettors who watch matches analytically.
You watch a match. You make notes - some during, some immediately after. The notes capture real observations: a specific pressing pattern, a goalkeeper who dominates his area or struggles at the far post, a striker who drops deep to link play in ways that don't appear in his statistics. Those observations are genuine. Some of them are analytically significant.
Three weeks later, when you're preparing analysis for an upcoming fixture involving one of those teams, you look at the notes. They're there. But they're not in a form that connects to the specific question you're now trying to answer. "Pressing looked intense against a low block" doesn't tell you what you need to know about whether this team's pressing will be effective against Saturday's specific opponent. "Wide players switch positions frequently" is an observation without a market implication attached to it.
The problem is that unstructured observations require interpretation every time you use them, and that interpretation work is what takes time and introduces inconsistency. You're re-processing raw data rather than working from processed information.
Structured notes change this. If the observation is recorded as "pressing intensity: high against teams ranked below eighth in competition, drops to medium against top-four opposition - evidence from matches against [teams], [matches]", you're working from a processed finding rather than a raw observation. The LLM synthesis step converts the processing from something you do manually each time into something you do once, properly, immediately after watching.
The Note-Taking Structure
The structure needs to be simple enough to use during a match - you're watching, not filling out a form - but specific enough to produce synthesisable output. These two requirements pull in opposite directions, and the resolution is a small set of categories that you train yourself to observe for specifically, rather than trying to capture everything.
Six categories. No more. You can add subcategories within them, but six top-level fields is the ceiling for what remains usable during live watching without the note-taking consuming the attention you need for the watching.
Formation and shape. Starting shape, defensive shape out of possession, attacking shape in possession, and any transitions between them. Not just the formation number - what it looks like in practice, specifically whether it matches the nominal system and where the key movement patterns are.
Pressing behaviour. When do they press, at what trigger, at what intensity, and against what type of ball carrier or situation. Specifically whether the press is consistent or selective, and whether it degrades across the match - does the press in the seventy-fifth minute look like the press in the fifteenth.
Set piece orientation. Both defending and attacking. For attacking: delivery type, target zones, movement patterns, who delivers and from where. For defending: zonal or man-marking, who picks up specific threats, how they handle near-post delivery versus far-post. This category is quick to note because it's discrete - you observe it during set pieces specifically rather than across the whole match.
Key player roles and contributions. Not a player ratings exercise - specifically, which players are doing things that won't appear cleanly in statistics and why that matters for the team's output. The striker whose pressing triggers defensive errors that become other players' assists. The fullback whose carrying creates the space that generates crossing opportunities. The midfielder whose positional play prevents transitions rather than winning duels. These are the observations that have the highest value relative to how commonly they're recorded.
In-game adjustments. What changed after a goal, at half-time, when chasing or protecting a lead. Substitutions and their tactical purpose as it became apparent, not as the commentator described it.
Vulnerabilities and strengths relative to this specific opponent. This category only applies when you're watching a match where both teams are relevant to upcoming analysis. Not general strengths and weaknesses - specifically, what this matchup revealed about how each team performs against this style of opponent.
The During-Match Note Format
During a match, the notes don't need to be polished. They need to be in the right category and contain enough specific information to be processed afterward. A shorthand that works:
Start each note with the category abbreviation: F for formation, P for pressing, S for set pieces, K for key player roles, A for in-game adjustments, V for vulnerabilities and matchup observations. Then the observation, as specific as possible, with a minute marker where relevant.
"P 23 - high press triggered by goalkeeper distribution to centre backs, striker cuts passing lane to near CB forcing longer ball, drops off immediately after, not sustained"
"K - [player name] consistently receives between lines in half-space, turns and carries rather than laying off, creates overloads right side - not visible in pass stats"
"A HT - switched to back four from back three, [player] moved from right CB to right back, immediate effect on [player]'s freedom to push forward"
That level of specificity - not exhaustive, not attempting to capture everything, but precise about what it does capture - is what the synthesis prompt can work with. "Pressing looked intense" is not workable. "High press triggered by goalkeeper distribution, drops immediately after initial pressure" is workable.
You won't always write at that level of detail in real time. Some observations will be rough. The rule is: immediately after the match, before the impressions fade, spend ten minutes reviewing your during-match notes and expanding any that are too brief to be synthesisable. The ten minutes immediately post-match is worth more than an hour of review three days later.
The Single-Match Processing Prompt
After each match, before the notes go into storage, you run a processing prompt. This is the step that converts raw notes into a structured database entry - the same conversion the referee database and manager database prompts perform, applied to match observation data.
"The following are my structured observation notes from [fixture, competition, date]. The teams are [home team] versus [away team], result [score]. I want you to process these notes into a structured match observation entry using the following fields. For each field, use only the information present in my notes - do not supplement with general knowledge about either team. If a field is not addressed by my notes, mark it as not observed rather than leaving it blank or inferring from general knowledge. Fields: formation and shape summary for each team; pressing behaviour summary for each team including any within-match changes; set piece summary for each team; key player roles identified; in-game adjustments by each team; vulnerabilities and matchup-specific observations. For each field you complete, include a brief note on which specific observations from my raw notes support it. Flag any field where my notes contain a single observation that may not be representative of a pattern."
The "do not supplement with general knowledge" instruction is the same principle as in the manager database prompts. The model has read enormous amounts of football content and has general impressions of most professional clubs. Without this instruction, it will quietly fill gaps in your notes with training data impressions, and you'll end up with a database entry that looks comprehensive but contains untracked inferences you'll mistake for your own observations.
The "flag single observations" instruction prevents overconfident entries. One observation of a pressing trigger is not a confirmed pressing pattern. Three observations across the same match with consistent characteristics is closer to a pattern. The flag keeps the distinction visible.
The Multi-Match Synthesis Prompt
Once you have processed entries from three or more matches involving the same team, you run the synthesis prompt. This is where the workflow produces its most valuable output - converting multiple individual match observations into a structured analytical brief that matches the opposition research format.
"The following are structured match observation entries for [team] from [number] matches over [time period]. I want you to synthesise these into an analytical brief covering this team's current patterns and tendencies. Structure the brief as follows. First, identify which observations are consistent across all or most entries - these are confirmed patterns. Second, identify observations that appear in some entries but not others and describe the conditions under which each appears - these are conditional patterns. Third, identify any observations that appear only once and flag them as single-match observations requiring confirmation. Fourth, identify any apparent changes across the time period - patterns present in earlier entries that don't appear in recent ones, or new patterns emerging in recent entries. Fifth, produce a matchup implications section: given the patterns identified, what are the specific market implications for a fixture where this team faces [describe upcoming opponent characteristics - defensive style, pressing intensity, set piece quality]? The matchup implications section should be specific to the described opponent rather than general."
The structure mirrors the opposition research brief format deliberately. A synthesis that produces output in the same format as the opposition research brief can be fed directly into the opposition research workflow rather than requiring manual translation. The two workflows connect - the video analysis produces the raw data, the opposition research prompt uses it as input for fixture-specific analysis.
The matchup implications section being opponent-specific is the most important structural decision in the prompt. A general summary of a team's tendencies is background knowledge. A summary connected to a specific upcoming opponent's characteristics is analysis you can act on. Requiring the specificity in the synthesis prompt rather than leaving it to a follow-up step means you don't skip it under time pressure.
Connecting to the Opposition Research Workflow
The opposition research article described a four-prompt sequence for building match-specific tactical profiles. The input materials for those prompts were publicly available sources - match reports, press conference transcripts, FBref formation data. Video analysis notes, processed and synthesised, are a better input than any of those sources for the teams you've watched directly.
When you're preparing an opposition research brief for a fixture involving a team you've synthesised video observations on, add your synthesis brief to the input materials explicitly:
"In addition to the match reports and other sources provided, I have a personal video analysis synthesis for [team] covering [number] matches. This synthesis reflects direct tactical observation and should be weighted as a primary source for formation, pressing behaviour, key player roles, and in-game adjustment patterns. Where the video synthesis contradicts information from the other sources, flag the contradiction rather than defaulting to either source."
That instruction converts your video analysis from supplementary information into primary source material - which it is, for the observations it contains. The flag-contradictions instruction is particularly useful when your direct observations differ from the narrative in match reports, which happens regularly. Press match reports describe what journalists found notable, which isn't always what was most analytically significant.
Managing the Archive
The note archive accumulates quickly. After a season of consistent watching, you might have forty or fifty processed match entries across fifteen to twenty teams. Without some maintenance discipline, the archive becomes unwieldy - entries from early in the season describing a team that has since changed manager, or a system that was abandoned after a poor run.
Two maintenance habits. Date-stamp every entry and the synthesis brief derived from it. When running a synthesis prompt, specify the time period explicitly and ask the model to identify whether patterns from early entries are confirmed in recent ones or whether the recent entries suggest a tactical evolution. Don't discard old entries - the historical record is useful for understanding a team's developmental arc - but weight the synthesis toward the most recent entries and make the weighting explicit in the prompt.
The second habit is flagging entries for reconfirmation when a significant personnel or tactical change occurs. A manager appointment, a significant January signing that changes the attacking structure, a defensive injury that forces a system change - any of these should trigger a reconfirmation flag on the relevant team's synthesis, indicating that the existing brief requires updated observations before being used as primary input for fixture analysis.
The flag can be as simple as a note at the top of the synthesis document: "Flagged for reconfirmation - [player] departed, [date]. Brief reflects observations prior to this change. Use with caution until updated." That note costs thirty seconds to add and prevents the brief from being used as though it's current when it isn't.
Anyway. The workflow is more structured than most bettors' current watching process, and the note-taking discipline during a match feels like an imposition the first few times. It stops feeling like one when the synthesis prompt produces an opposition research brief that would have taken two hours to write manually and does it in fifteen minutes because the underlying observations were captured properly when they were fresh. The value is in the compounding - each match watched properly adds to a database that makes subsequent analysis faster and more grounded. Each match watched without structured notes adds to an archive of impressions that degrades rather than compounds.
FAQ
How do I handle matches I watch but don't have time to process immediately afterward?
Two-tier approach. If you can't run the single-match processing prompt within a few hours of watching, at minimum spend five minutes immediately after the match writing a brief summary of the two or three most significant observations in each category before the detail fades. That summary is enough to anchor the fuller processing when you get to it - raw notes plus a brief anchor summary produce a processable entry even after two or three days. Beyond three days, the processing quality drops significantly and it's worth being honest about that in the entry itself: "processed four days after watching - detail in pressing and set piece sections may be incomplete." The entry is still useful. Just less precise than a same-day processing.
My notes from watching include things I can't categorise cleanly into the six fields - observations about crowd influence, referee behaviour, pitch conditions. Where do those go?
A seventh catch-all field for contextual factors, used sparingly. The reason to keep it separate from the six analytical categories rather than forcing contextual observations into them is that contextual factors - pitch conditions, crowd effects, referee profile - have different shelf lives and different synthesis implications than tactical observations. A team's pressing behaviour is a persistent pattern. The fact that they played on a waterlogged pitch in November is a one-time contextual note that should inform interpretation of that specific entry rather than appearing in the pattern synthesis. Flag contextual entries explicitly in the processing prompt: "The following contextual factors affected this match and should be noted as context for the entry rather than as pattern evidence:
" The model will then handle them differently from the tactical observations when synthesising.
Is there value in running the synthesis prompt for teams I've watched extensively versus teams I've only seen once or twice?
Different value, not less value - but the output needs framing accordingly. A synthesis from eight to ten matches produces confirmed patterns and conditional patterns with reasonable confidence. A synthesis from two matches produces two data points that may or may not represent stable patterns, and the synthesis prompt should reflect that. Add this instruction for thin-data syntheses: "This synthesis is based on only [number] observations. Frame all findings as provisional observations requiring confirmation rather than established patterns. Do not use language that implies certainty about any tendency - use language that indicates these are early signals worth monitoring." The output is then a watchlist of patterns to confirm through further observation rather than a brief ready to use as primary input for opposition research. That's a different and still useful document - it tells you what to watch for specifically when you next see this team, which focuses your observation in a way that accelerates the confirmation process.
