Aspiration archaeology — DIG, 2026-08-30

The ask (ember, verbatim): "dig through our project history and of original ~/dev/dragons-clutch and see what we've got missing or left to do and so forth BESIDES the things you already had in mind... an opportunity to reevaluate, regroup, take a look."

Method: five parallel digs (dragons-clutch git history incl. all branches and the market-theory thread; dclutch decision/doc archaeology; the Lean formal side; a stranger's product review of https://clutch.dregg.pro; a bounded cv transcript sweep for aspirations in ember's own voice), synthesized against the KNOWN set (GOAL.md done-log, WAVE.md queue + cycle-3 charter, LIVENESS_CENSUS Q1–Q9). Everything in those files is excluded; this doc is only what is NOT there.

The reframe: the ledger already exists, and it is orphaned

The single most important prior fact: docs/evidence/ASPIRATION_LEDGER_2026_08_27.md (ARCH-EOL, 2026-08-27, 3,002 lines) already did the deep half of this dig — 537 intentions extracted, 187 verdicted MISSING, five tiers, ten recommendations. And WAVE.md and GOAL.md reference it zero times (fourteen other files cite it; the plan is not one of them). The ledger is suffering the exact fate it documents: named, unrouted.

So this doc has two jobs: (A) a three-day drift check — which ledger recommendations moved since 08-27 and which still float; (B) genuinely NEW finds the ledger does not contain. Ranked at the end by wow-per-effort for the launch ember is actually driving (a public demo that must feel ALIVE and be genuinely usable).

---

A. Ledger drift check (2026-08-27 → 08-30)

The ledger closed with ten recommendations "ordered by what a miss would cost." Verdicts today, each verified at the tree:

Drift verdict: 2.5 of 10 moved in three days. The ledger's rows are not being consumed; they are being independently rediscovered (this dig re-found several before finding the ledger).

---

B. NEW finds — not in the ledger, not in the plan

B.1 The site is a protocol museum, not a venue (fresh-eyes product review)

A stranger visiting clutch.dregg.pro sees integrity and zero motion: no question to bet on, no price, no clock, no people. Seventeen gaps were found; the load-bearing ones, each absent from plan and ledger:

The reviewer's verdict, worth keeping: the "empty rather than fake" honesty discipline is excellent engineering ethics and is currently the direct cause of the aliveness problem — the fix is shipping real motion so the honest branches stop firing, never relaxing the honesty.

(Coordination: POLISH lane opened 09:41 on stats/listing presentation — overlaps the first row's neighborhood, none of the rest.)

B.2 2.4MB of unreferenced key art

dclutch/apps/dclutch-web/public/art/dragons-clutch-key-art-v1.png (dragons-clutch working tree, untracked, created 08-29 04:33): a dragon's claw cradling a glowing faceted polytope — genuinely good brand art, on-thesis (payoff basis as gem), referenced by zero files, in zero commits. Wire it into the landing + make it the OG share-card background. Also orphaned: gen-1's whole docs/site-plan/ (concept inventory, site content map) has no successor. Size: hours. Level: engineering.

B.3 SECURITY.md's promise — the trigger has FIRED (decayed since the ledger)

Ledger M-48 recorded gen-1 SECURITY.md:50 — "A private reporting address and coordinated-disclosure process will be added before any public test deployment" — as latent. It is latent no longer: DEPLOY-1 put seven programs on public devnet (08-27/28) and 73b87027 (08-29) enabled signed browser submission. dclutch has no SECURITY.md at all. For a protocol whose brand is honesty, a missing security contact on a live public deployment is the one gap a hostile stranger screenshots. Size: an hour for the honest version. Level: engineering (contact address is ember's).

B.4 Aspirations in ember's own voice, in no file (cv sweep, deduped against the ledger)

1. Fundraising is the point of the demo — 08-28: "I was hoping I'd have something deployed and usable by other guys so that maybe we'd get some weekend fundraising but nobody is gonna be able to look at, participate, or understand anything." The plan tracks the demo as an engineering milestone with its motive stripped off. This quote is the ranking function for section B.1 — it is why aliveness gaps outrank protocol completeness. Level: vision (belongs in INTENT.md).
2. A longform blog post/poster on the compost method — 08-22: "'plan to throw one away' except 'plan to compost at least three' :D", target ~/src/dregg-posters, scraped from cv. Never started; would be the #buildinpublic artifact the six-step founding plan (ledger M-1) promised. Size: a day. Level: vision-flavored, ember-facilitated.
3. AI-authorship header for the filings — 08-20: ember explicitly wanted the filings to let the AI speak for itself ("This document is written by AI with a human facilitator and represents the positions of the AI"). Recorded nowhere; 1388 is still unfiled (A.1), so the stance could still ride it. Level: vision.
4. A Polymarket-resolved venue — 08-27, half-dismissed in the same breath ("a different venue over polymarket-resolved stuff...hm"). Recorded so it stops being re-invented; the relay design (CHAIN_STATE_SOURCES) is the natural home for a verdict. Level: vision, low priority.
5. Factory-contract stamping out per-market programs — 08-22, self-parked ("maybe for a very different type of protocol someday"). Recorded as a someday. Level: vision, lowest.

(The cv sweep's sixth find — "beat isomkts on the merits" — IS ledger M-3, the twelve-item ceiling; not new, still unrouted. Coverage caveat: the sweep reaches only sessions whose cwd contains "clutch", corpus starts 08-18; dClutch thinking in degg-research/breadstuffs/dregg-posters sessions is unswept.)

B.5 PRODUCT_THEORY_REDIRECTION — 539 lines alive only on a branch

git show origin/agent/product-theory-redesign:docs/reviews/PRODUCT_THEORY_REDIRECTION_2026-08-24.md (commit 9dba79f6): seven concrete product upgrades — exact dual-bound optimality certificate, prepaid lazy foundation graph, cross-expiry rolls, products beyond one source/statistic/window, quiescent Dealer epochs + transferable LP shares, shared Source work capitalization, fee geometry as measured experiment profiles. Its sibling (the trading-UI brief) was rescued to docs/recovered/; this one was not — the ledger cites its name (G-20) without its content. Recovery is one git show > docs/recovered/.... Size: minutes to recover, then a reconcile pass. Level: engineering to recover, vision to adopt.

B.6 There is no CI

The live tree (~/dev/dclutch) has no .github directory at all; the only workflow in either repo is dragons-clutch's pages.yml, a site deploy. Every gate this project is proud of (gauntlet, abi:verify, fixtures:verify, sbom_check, emitter byte-identity, the new seam-audit) runs only when someone runs it. The plan's SEAM-CI built the audit; there is no substrate that executes it on push. Ledger G-9 named this; it is in no queue. Size: a day for the first honest workflow (fast gates only, hbox for heavy). Level: engineering.

B.7 The market-theory thread, resolved

The branch this dig ran from (agent/market-theory-support4) is a vestigial name: zero commits, zero files, zero docs mention market-theory; current work simply lands on it. The only substantive trace is stash@{8} (market-theory-quantized-work-authority, 614 insertions of gen-2 quantized-relation/price-measure work) — already inventoried as ledger M-58, superseded by the gen-3 restart. No action beyond M-58's existing row.

B.8 Gen-1's product/economics/strategy layer has no successor

archive/gen1/docs/ holds PRODUCT_THESIS.md ("a state-space compiler"), ECONOMICS.md, COMPETITIVE_POSITION.md, RESEARCH_AGENDA.md (R1–R4 with falsifiers), SIMPLEX_AUCTION.md, COST_MODEL.md, BENCHMARK_PLAN.md, DEPLOYMENT_REVENUE_BOUNDARY.md. dclutch/docs has decisions/design/evidence — and no product thesis, no economics, no competitive position, no research agenda. Two specific losses with teeth:

• DEPLOYMENT_REVENUE_BOUNDARY.md §5's five release tracks (A–E): the project is executing something Track-C-shaped (author-affiliated devnet deployment) without any record of choosing a track — while ledger M-2 carries ember's "if i can't at least deploy it to mainnet myself i'm probably not interested", silently inverted by the devnet-only posture.
• The counsel-ready regulatory packet (archive/gen1/docs/regulatory/, never edited since 08-18) is referenced by nothing live; its own use discipline ("reconcile every factual statement against the then-current repository") is two generations stale.

Size: the re-ask is an ember conversation; the docs are days. Level: vision.

B.9 Escape hatch for the live 1.4M wall, in no queue

Gen-1 scouted Groth16 succinct clearing at ~255k CU in a 795-byte transaction — 5.5× margin against the 1.4M ceiling ("We are the consumer it never had"). The fourteenth wall is that exact ceiling, live today (seeds 1 and 7 still exceed it; DECOMP owns the residual). The scouted hatch appears in no queue (ledger M-55 named it; still unrouted). Size: a spike to re-validate the numbers. Level: engineering, authority-adjacent.

B.10 Design rationale at risk: the batch relation is an FHE target on purpose

The DrEX/dark-FHE ambition was properly DROPPED-BY-DECISION (ledger §verdict) — but the ruling retired the ambition without preserving the rationale: the batch relation was chosen because it is a good FHE/MPC/vFHE target (ledger M-3 item 11), and nothing in either repo records that as the reason for its shape. If the relation is ever "simplified" by someone who doesn't know why it is shaped that way, the door closes permanently. One paragraph in INTENT.md preserves it. (Same class: the original motivating use case — energy providers settling plans without disclosure — is not crypto at all.) Size: one paragraph. Level: vision-preservation.

B.11 Smaller new rows

• Manipulation-cost table withdrawn from the public site: the thread transferred (ManipulationFloorV1 live in the founding generator) but its public evidence surface was quarantined 08-22 (329faab5 → bfe4b4f0) and never restored — a differentiator (manipulation-cost-aware markets) with no public face.
• M-30 still unowned: --no-default-features --features dealer-family does not compile; in no queue.
• docs/ledger/VALIDATION_BACKLOG_2026-08-31.md (28KB, 08-29) — a second live backlog neither GOAL nor WAVE references; same orphan class as the ledger. Someone should merge or cross-link it into WAVE's queue before it too is rediscovered.
• Nothing re-queues a deferred measurement (ledger N-20's moral): a measurement that loses a resource race is written down perfectly and never re-run. The board has no mechanism; worth one standing rule.
• External venue routing (Manifest/AMM/RFQ/Jupiter) — ledger M-52; noted here only because the plan's "exchange story" can be mistaken for it: the plan is markets about venues, M-52 is routing claims to venues. Distinct and unretired.

B.12 The formal layer: proven, emitted — and unconsumed (Lean sweep)

Scope: formal/dclutch-semantics = 122 modules, 1,688 theorems, 80 emitters, zero sorry/admit/axiom (native_decide confined to examples/ABI-constants per TRUST.md, spot-checked). 69 Rust files carry Lean provenance headers. 159 theorems live in 20 modules unreachable from any emitter. Ranked:

Size: mostly days-per-item; the CI+re-emit gate and the spline layout field are the two with launch-visible payoff ("every number on this site is theorem-checked" is a marketing sentence no competitor can say). Level: engineering, with the spline field being an authority decision (wire break).

B.13 The decision-log dig: complete designs, one owner short (doc sweep)

The pattern across every top find here: the design is finished and correct; the executable half stopped exactly one owner short. None is blocked on a missing idea.

1. Dealer still commits the defect class that stranded the first market's principal. Decision 0008 §1: "Every consumer DERIVES both the replay and the Vault from [custody_context]… no route may assume the Market address." Verified live today: dealer/v3_composer.rs:561, v3_accelerator_accounts.rs:849, v3_multi_lp.rs:1009-1011, and dealer-sbf/src/lib.rs:2049 all derive HoardPrincipal from the MARKET address; none reads aggregate.custody_context. The owning lane (tranche-A Dealer) was recorded CLOSED — for the CoreState decode fix, not this. Plus: Dealer v1 partitions by Dealer state address where v2/v3/v4 partition by child_root — two conventions in one family. Days; engineering; correctness-critical.
2. The recovery ladder is welded shut for every family — every devnet market has exactly one source attempt and no fallback. process_funded_transition reachable only from #[cfg(any())]; RecoveryMaterialSlotV1::new is Pyth-only; Core refuses CreateFund on any recovery material (the correct Q2 weld). MAINNET_STATE_RELAY §13 names the post-v1 lane (funded FailNext over RecoveryPolicyV2, relayed leg under a disjoint key set); it has no owner and appears in no charter. The open question on top: does v1 ship one-attempt markets forever? Weeks; engineering under a ruling. (edges maybe-known: GIT-SCAN item 2, census R2)
3. The fee program's trigger fired and the lane did not spawn. FEE-GEO's row says "Trigger: cycle 3"; cycle 3 opened 08-29 with six items and FEE-GEO is not one. Verified: dispersion_bps/FeeBaseV1/ accrual_monotone — zero hits; General's config has NO fee field (General charges nothing). Underneath: M-26 (the day-one rate question) and N-15 (formalize-before-freeze precondition, recorded in no gen-3 doc). Kernel + composite: weeks. The rate: ember's alone.
4. General is an order-book venue with no way to place an order — effect_artifacts_v3.rs:239 still lists all seven collection/candidate actions as unauthored_actions!. GEN-SEVEN sits in the 08-27 handoff queue with rungs laid; the cycle-3 charter does not carry it. Inside it, three decision-0010 residues: work-escrow lamports never move (§6.3), ExpireSettlement has no counterpart — a stalled settlement cursor is stuck (§6.4), nothing creates/closes the claim-escrow Position (§6.5). Weeks, one coordinated unit; engineering. (GEN-SEVEN itself maybe-known; the charter-drop and the three residues are the finding)
5. Decision 0005's promised omission rows were never recorded (the M-19 shape, verified twice more): the capability-seal Lean ABI migration ("recorded in the omission index" — it is not; no seal emitter exists) and seal rent reclamation (no CloseSeal route anywhere; permanent unreclaimable rent on a growing write-once account class). OMISSION_INDEX has neither row — all 38 read. Index rows: minutes. Migration: days.
6. The artifact bridge is frozen at day one. All four ELF-level theorems date 2026-08-25 and cover artifacts retired the same day; formal/qedsvm-direct-v12/ contains ZERO .lean files (traces + harness + evidence.json only); the eight committed Kani harnesses in tools/direct-translation-validator/src/kani_proofs.rs have never run once. The universal-theorems park (D-4) covers none of this half. Running Kani: hours. The bridge: ember's "is this still the architecture?" first.
7. Retirement — the market's last life step — has never run anywhere, on any substrate. README's own voice: "winding a market all the way down to retired has not run anywhere yet"; the journey gauntlet records the gap MOVING rather than closing (retire refuses while the Hoard holds one atom; emptying means redeeming; redemption is behind the Hot gate); census Q3 + Q6 now gate it too. The acceptance condition that distinguishes a market from a one-way trap. Ruling (Q3) then days-to-weeks.
8. Dealer's capital design (consent-bound tranches, quiescent epochs, scenario solvency) is fully written and has no path to execution — grep epoch in dealer-codec + scenario kernel: nothing; the design doc's tranche fence ("must not simulate tranches with Dealer counters") holds only because nothing simulates anything. O-011's closure condition untouched; Q4 rates Dealer least-live. Design pass + weeks; a capital-structure ruling inside.
9. The plan carries a transport candidate the design already ruled out. WAVE's demo-shape section still names Wormhole Queries as "candidate permissionless upgrade… MR lane owns pinning"; MAINNET_STATE_RELAY §3.1 concluded "not a candidate for v1 and not a near-term upgrade path." One of the two is stale; a lane that picks it up re-does a closed investigation. One paragraph.
10. Relay-slice named lifts, each with a stated trigger and no owner: large-account chunking (inline window > 448 B needs persisted SHA-256 midstate), m-of-n with m>1 (threshold expressible, no multi-signer campaign), the Realm-level shared observation cache (N markets pay N rents), and §10.1's time bomb — after DBC 0.2.0 a decoder pinned to VirtualPool silently stops seeing transfer-hook pools (fix named: account 5 = PoolConfig, unbuilt). The graduation market's nine-record set: "one getAccountInfo away and none has been read" — the demo thesis is complete in a harness and has never touched mainnet bytes.
11. Cross-host reproducibility is unestablished — checked-release-candidate.sh:6 says LOCAL in its own capitals; nothing shows a second host reproduces the digests (PROJECT_METHOD rung 6; U-011). Pairs with B.3/B.6: the two assurance rungs that matter because the deployment is already public.
12. Gen-1's ratified commit/reveal subdivision did not cross the generation boundary — ADR 0006's answer to candidate withholding / proposer bonds; zero occurrences in gen-3; General ships Consider→Freeze without the subdivision and no record says why. A solver who withholds the best candidate faces no bond and no detection. A design pass; vision-level mechanism work. (sharpens ledger G-7)
13. Three cheap honesty repairs: (a) README advertises a public relay publication log at portal.dregg.studio/relay/publication_log.jsonl that does not exist — while DEPLOY_1 §6.3 has the relayer disarmed and §4.11 says a relayer profile "should not be released" without publication; the site makes a liveness-checkability claim the deployment cannot support. (b) ADOPTED_2026-08-20 is cited eight times from dclutch at two different paths, both wrong (it lives only in dragons-clutch/archive/gen1) — the fee-shape decision record is unreachable from the repo that cites it. (c) COMPOST.md promises the repo graft "will have its own reviewed plan" — none exists. Hours each.

And the doc sweep's framing note, which this whole dig confirms: "the mechanism is not the medium; it is whether a row is in the thing A GATE READS. blocked.json and SETTLEMENT_BLOCKERS are read by tests. OPEN_QUESTIONS.md is read by people." The census Q-rows, the fee lane, the spline question, and the Dealer capital design live in documents only people read.

---

C. The ranked gap map (wow-per-effort for a launch that must feel alive)

The meta-move first. This dig's most repeated finding is the same one the ledger made about itself: rows survive only in the thing a gate (or the plan) reads. The ledger, VALIDATION_BACKLOG, the census Q-queue, the fee lane, and now this doc all live outside WAVE.md. The single highest-leverage act is ONE WAVE entry that dispositions this doc's tiers — schedule / park-with-trigger / retire-with-reason — so this file does not become the next orphan the next archaeology lane rediscovers.

Tier 1 — hours each; each either visibly changes what a stranger sees or is irreversible if missed:

1. Market titles/questions registry (B.1) — the single highest wow-per-effort item this dig found.
2. Countdown clocks + implied-probability display (B.1).
3. /pulse + /activity into the nav; default activity to the flagship; SIM flip (B.1 — the flip is queued; the nav rows are not).
4. Key art onto the landing + OG share cards (B.2 + B.1) — distribution IS the demo for a memecoin-adjacent launch.
5. SECURITY.md with a real contact (B.3, B.13.11) — the "before any public test deployment" trigger already fired.
6. CFTC 1388: check the docket, file or explicitly withdraw (A.1) — the one irreversible deadline; the AI-authorship header (B.4.3) can ride it.
7. Recover PRODUCT_THEORY_REDIRECTION + d5dda5d before git gc (B.5, A.8).
8. Honesty repairs: README's nonexistent relay publication log, the ADOPTED_2026-08-20 broken citations, the Wormhole plan/doc contradiction (B.13.9, B.13.13).
9. Run the eight committed Kani harnesses once (B.13.6).

Tier 2 — days; structural product and correctness wins:

1. docs/INTENT.md (A.2 + B.10 + B.4.1) — one hour of writing that makes every future lane rank like ember ranks; still the ledger's best call, now three days more overdue.
2. Time-series poller + sparklines; ws live updates (B.1) — the site's signature image is currently missing from the site.
3. First CI workflow + the Lean re-emit byte-gate (B.6 + B.12) — 54 of 69 generated files unguarded; every gate today runs by hand.
4. Dealer custody_context derivation fix (B.13.1) — the defect class that stranded the first market's principal, still live in four routes.
5. Search/filter, "Create" labeling + faucet link, static-JSON API (B.1).
6. Decision-0005 omission rows + seal emitter migration (B.13.5).

Tier 3 — ember decisions; cheap to ask, compounding to leave un-asked:

1. The floating Tier-3 questions M-22 (+ its new 7Mcu-untradeable sibling), M-23, M-25, M-26 (A.6) — and spawn FEE-GEO, whose trigger fired (B.13.3).
2. Upgrade posture / release track A–E / the mainnet precondition, re-asked as one conversation (B.8, B.13 items 6+9's "is this still the architecture?", census Q1 interaction).
3. The spline basis layout field — is the certified-basis substitution the B-spline requirement, and does a Market get to select it? (B.12 top row, M-4/O-013.) 221 theorems wait on one field.
4. Twelve-item ambition ceiling: adopt / trim / retire per item (ledger M-3) — the de facto product bar, still in no file.
5. Expansion-frontier + proven-unconsumed-kernel disposition lane (A.4, B.12) — schedule, park-with-trigger, or retire, each with a reason.
6. One-attempt markets: does v1 ship without any recovery ladder forever? (B.13.2.)
7. Retirement: rule Q3, then run the full wind-down ONCE anywhere (B.13.7) — the last unexercised acceptance condition.
8. General order collection (GEN-SEVEN) into a charter, or explicitly post-launch (B.13.4).

Tier 4 — recorded so they stop being re-invented: compost-method blog post (B.4.2), Polymarket-resolved venue (B.4.4), factory contracts (B.4.5), external venue routing (M-52), EVM (M-60), Groth16 hatch spike (B.9 — promotes to Tier 2 if the 1.4M residual resists DECOMP), commit/reveal mechanism pass (B.13.12), Dealer capital design (B.13.8 — large; wants its own charter), relay-slice lifts incl. the DBC-0.2.0 decoder time bomb (B.13.10 — the decoder row promotes the moment a graduation market is real).

---

Provenance: five subagent digs 2026-08-30 morning (dragons-clutch git archaeology; dclutch doc archaeology; Lean formal sweep; clutch.dregg.pro stranger review; cv transcript probe), synthesized by DIG. Verification commands inline. Read-only dig; nothing but this file changed.