Infinite Game OS

Definition

Pattern Harvest is a session-level reflective discipline. After substantial work, the skill scans the session across three lenses simultaneously: meta patterns (how the collaboration happened, where the work shifted, where meta-awareness held or drifted), system upgrades (infrastructure gaps, governance needs, skill gaps, standing-instruction gaps, memory currency, token-efficiency flags) and expression quality (a one-line check on whether the session's produced artifacts met the operator's expression standards). Findings present in three labeled sections, then resolve into a numbered recommendation list tagged by effort: now, quick, plan or defer. The operator selects by number. Implementation runs before session closeout so the harvest's findings are already in conversation context when closeout executes. Pattern Harvest is operator-invoked, never automatic. Lightweight single-task sessions skip it. Co-creation work, governance design, multi-artifact builds and structural-decision sessions are where the harvest earns its name.

The three lenses run in parallel

Pattern Harvest scans a working session across three lenses simultaneously. Lens A is meta patterns: how did the work happen, what patterns in the collaboration emerged, what does the session reveal about the operator's thinking style, where did meta-awareness hold or drift. Lens B is system upgrades: infrastructure gaps, governance improvements, skill gaps, standing-instruction gaps, memory currency, token-efficiency observations. Lens C is expression and output quality: a one-line check on whether the session's produced artifacts met the operator's expression standards.

The lenses run together, not in sequence. A pattern noticed under one lens often illuminates the others. A drift moment in meta patterns becomes an expression-quality flag becomes a standing-instruction recommendation. The parallel scan catches that interconnection. A serial pass loses it.

How a harvest closes

Findings present in three labeled sections: Meta Patterns, System Upgrades, Expression Check. Each section is bullet points, one idea per line, no narrative padding. The Expression Check has a single valid one-word output if nothing flagged: clean. The other two sections list specific observations.

Recommendations resolve into a flat numbered list, each one tagged by effort. The 'now' tag runs in the current session under five minutes. The 'quick' tag runs in the current session with one to three file edits if approved. The 'plan' tag warrants a pending plan or multi-session work. The 'defer' tag is worth noting and not urgent. The operator selects by number: 'Implement 1, 3, 5' or 'All' or 'Pass.' Approved items execute before session closeout.

Pattern Harvest pairs naturally with any session-closeout flow. Because the harvest runs first, its findings are in conversation context when closeout executes. The closeout flow captures what the harvest produced without any explicit handoff. The two flows are independent. Neither needs to know whether the other ran.

Use Cases

Co-creation session with sharper-than-document language

When a session produces phrasing sharper than the operator's reference docs, Pattern Harvest catches it. The Lens A pattern notice flags the spontaneous articulation. The Lens B recommendation queues a memory-store entry or a doc revision so the language does not stay trapped in the conversation transcript.

Multi-artifact build session

Building three or four artifacts in one session leaves drift residue. A frontmatter standard slipped on file two. A boilerplate block diverged on file four. Pattern Harvest catches the small drifts the build was too dense to surface in real time. Effort tags route the cleanup: now for the obvious fixes, plan for a sweep across the rest of the system.

Governance design session

When the session redrew a protocol or rewrote a standing instruction, Pattern Harvest's Lens B asks the next question: where else does this new rule apply that has not yet been touched. The harvest produces a recommendation list of follow-on edits across adjacent governance documents. The new rule lands consistently rather than partially.

Structural decision session

Big architectural calls leave residue. A renamed concept needs to flow through the vocabulary. A new module-of-modules decision needs to update related documents. A retired pattern leaves stale references. Pattern Harvest scans the session against the operator's body of work and surfaces the second-order updates the decision triggered.

Pre-closeout reflective pass

On any substantial session, Pattern Harvest is the natural pre-closeout move. Run the three lenses, present the numbered recommendations, accept or pass on each, then close the session. The harvest finishes inside the session it analyzes. The closeout records what happened. The two together turn a substantial session into a structurally absorbed one.

FAQ

When should I skip a Pattern Harvest?

Skip it on lightweight single-task sessions. A bug fix, a one-line edit, a quick lookup. The harvest is built for substantial work: vocabulary moves, governance design, multi-artifact builds, structural decisions, co-creation sessions. If the session generated nothing worth analyzing at the meta level, the harvest produces nothing useful.

How does Pattern Harvest differ from a session log or session closeout?

A session log or closeout records what happened. Pattern Harvest analyzes how it happened and recommends upgrades. The two are different jobs. Closeout asks what got built. Harvest asks what the session reveals about the system and what should change as a result.

What does each effort tag mean in practice?

The 'now' tag means under five minutes, run it in the current session. The 'quick' tag means one to three file edits, run it in the current session if the operator approves. The 'plan' tag means multi-session work that warrants a pending plan. The 'defer' tag means worth logging and not urgent. The tags route each recommendation to the right time horizon.

Why operator-invoked instead of automatic?

Substantial-session judgment is the operator's call. Auto-firing the harvest after every session would produce noise on the lightweight ones and miss the moment for the substantial ones. The operator triggers the harvest when the session warrants the analysis. The skill respects that judgment.

Can I run a Pattern Harvest mid-session, not just at the end?

Yes. Mid-session harvests are useful when something shifted. A new pattern emerged. A decision rebalanced the work. The energy moved. Run the three lenses against the work in progress, capture the recommendations and adjust the rest of the session accordingly. The closeout pairing still works at the end. The mid-session use is valid.