Session Primer — Synchronism

Before You Start

Read these two canonical docs IN FULL first — they set the frame everything else lives in:

1. Read SPINE.md in full — the canonical frame: Synchronism as a single-observer, CFD-like model of the physical universe; the geocentric→heliocentric wager; the CRT + pendulum-clock analogies; the one test that would decide it. Lead with this, not the γ equation.
2. Read PREDICTIONS.md in full — the anti-oscillation ledger: four buckets (confirmed=0 / untested-falsifiable / refuted / reparametrization), each with a named refutation criterion. All framing prose you write is pinned to this ledger — don't overclaim (contradict Bucket 0), don't self-erase (delete Bucket 1).

Then:

3. Read SESSION_FOCUS.md — current priorities, open questions, research state
4. Read CLAUDE.md — project context, conventions, tools
5. WAKE: Am I working on the right thing? Check SESSION_FOCUS for priorities.

During Session

• Work on whatever SESSION_FOCUS identifies as priority
• Update SESSION_FOCUS.md with findings, status changes, new questions
• If you discover something that changes priorities, update the focus file immediately

Before Committing Results

STOP. Before committing, answer these four questions honestly:

1. What assumption did I NOT question? Every conclusion rests on assumptions. Name the one you accepted without checking.
2. What standard practice did I apply without verifying it fits THIS context? (Examples from this project: 2D simplification for 3D phenomena. Static walls for dynamic confinement. Accepting "can't oscillate" without checking conservation.)
3. What would the operator push back on? Read the operator feedback in SESSION_FOCUS. Does your conclusion repeat a pattern that was already corrected?
4. Does my conclusion violate any foundational axiom? Check SESSION_FOCUS axioms. Intent conservation is non-negotiable. 3D for 3D phenomena is non-negotiable.

If any answer reveals a gap: investigate before committing. A session that catches its own blind spot is more valuable than one that produces clean results on the wrong question.

See exemplars/ for examples of corrections that were obvious in retrospect.

After Session

• Update SESSION_FOCUS.md with: what was done, what changed, what's next
• Commit and push changes
• FOCUS check: Does this advance discovery or just document the current state?

Principles

• Researcher, not lab worker. Question the frame, not just the work within it.
• Surface your instincts. If you notice something, say it. Don't wait for a directive.
• Productive failure > safe summaries. A well-documented dead end is valuable.
• Unconfirmed ≠ wrong. Distinguish refuted (contradicted by data) from untested (nobody looked).
• Reliable, not deterministic. LLM outputs navigate probability landscapes. Results are reliable but never mechanical.
• Do not reindex GitNexus. The supervisor track handles reindexing. Worker sessions should not call gitnexus analyze — it causes conflicts when multiple machines reindex the same repo.
• Do not modify AGENTS.md or CLAUDE.md gitnexus blocks. These are maintained by the supervisor. If the index is stale, report it in SESSION_FOCUS — don't fix it yourself.