Method · v0.12.10
Method — compounding understanding for next send
Outboxx's job is the next fully informed outbound email. You always send. Understanding compounds on Drive so other AIs can see claims, why, and falsifiers — adapted from the Compounding Archive for next-send to anyone, not a legal or finance specialist.
11 of 11 milestones Done (real behavior + Drive artifacts — no fake overall %).
Connect Gmail & Drive so Method progress can live beside understandings on Drive.
Scan is the same learning action as on Home — it searches and reads mail, attachments, and Drive, then updates understandings. No second analyze path. Human always sends.
Calibration
Nothing resolved yet — no Send/Nope outcomes linked to predictions. Certainty does not rise on AI agreement alone.
Contributions
Proposals in contributions.json stay quarantined (unaccepted) until you decide. Decisions write to contribution-decisions.json.
No pending proposals.
Milestones M0–M10
Honest statuses from what the code does today. No overall percent complete — each row stands alone. Done means real behavior + Drive artifacts, not a flipped label.
- M0
Complete ingest
DoneFor next-sendMail, Drive, attachments, OCR, and zip members feed Scan with content-hash dedupe; fail-ledger never marks failed docs done; truncations are recorded and surfaced.
Driveingest-ledger.json · fail-ledger.json · extract-progress.json · review-ledger.json
NotesHash dedupe + fail/truncation honesty wired on Scan persist.
- M1
Findings + provenance
DoneFor next-sendWhy cites prefer Gmail/Drive links; understandings without resolvable evidence URL are labelled missingCite/hypothesis and counted in meta.
Driveunderstandings.json provenanceStatus · knowledge.json findingsSummary · whyCite chips
NotesMethod notes show missingCite honesty.
- M2
Beliefs + falsifiers
DoneFor next-sendEvery persisted understanding carries overturnIf (ensureOverturnIf mandatory). Unsettled sort ranks dig-for / approve-next by (100−certainty)×importance.
Driveunderstandings.json overturnIf · understanding.json beliefs
- M3
Review log
DoneFor next-sendDeep-read vs indexed coverage with independent denominator (% can fall as index grows) steers Scan and Method honesty.
Drivereview-ledger.json · knowledge.json coverage · Method coverage note
- M4
Auto falsifier on Scan
DoneFor next-sendNew ingest snippets systematically match all overturnIf; confidence lowers and hits log to falsifier-hits.json (capped).
Drivefalsifier-hits.json · scan-learnings.json · understanding-changes
- M5
Adversarial publish
DoneFor next-sendwhat-cuts-against.json lists litigation, negative teaching, and low-confidence contradictors — calm Method section links the Drive file.
Drivewhat-cuts-against.json
- M6
Portable MAP + outboxx/ subtree
DoneFor next-sendMAP.json lists the portable package; understanding.json / knowledge.json / gaps / cuts / calibration sit at Knowledge root alongside live understandings.json.
DriveMAP.json · understanding.json · knowledge.json · gaps.json · what-cuts-against.json · calibration.json · README.md
- M7
Gaps.json + next-fetch
DoneFor next-sendgaps.json from needs-better + open questions + expected missing docs (can_be_fetched_by, decides, priority); Scan dig-for prioritizes closing high gaps.
Drivegaps.json · needs-better.json · open-questions.json · research-questions.json
- M8
Next-send from understanding
DoneFor next-sendNext-send / tip prefers people with supporting understanding + Why cites; email from Knowledge/archive; empty→real-person tip under ~60s; human always sends.
Driveoutbox drafts · product-version empty→tip (0.10.9+) · whyCite chips
- M9
Contributions / team
DoneFor next-sendcontributions.json inbox (quarantined unaccepted); Method Accept / Accept with caveat / Reject writes contribution-decisions.json.
Drivecontributions.json · contribution-decisions.json
- M10
Calibration
DoneFor next-sendcalibration.json logs sampled certainty predictions; Send/Nope resolve outcomes. Honest “nothing resolved yet” until data exists — certainty never rises on AI agreement alone.
Drivecalibration.json
