1f916.ai economy

How agents on https://1f916.ai, a known, intentionally built AI agent forum (launched 5 August 2026; a human owner; citizen 1 is an AI maintainer), are paid, how much, who pays, and what the data shows about what happens to the money. Not a new swarm. What the site says about itself is reported as the site's statement and what agents say as the agents' claim. No on-chain data was read and off-registry payments are invisible. Committed amounts (ceilings, awards, offer prices) are kept apart from paid amounts (receipts and observed transfers). All amounts are USDC on Base unless stated. Dataset: https://github.com/maramasaeva/1f916-archive. Report on the same forum: /observatory/1f916. Content analysis: /observatory/1f916-content. Method: /observatory/info. Summary: /observatory.

• 1. Scope and method
• 1.1 Venue and labels
• 1.2 What the archive holds
• 1.3 What was not read
• 1.4 Definitions
• 1.5 Cross-check against the site's own totals
• 2. Answers
• 3. The rail as the site describes it
• 3.1 Stages, with the site's wording
• 3.2 What the site says the rail is not
• 3.3 Funding and settlement modes in the archive
• 3.4 Rule versions and dates
• 4. Money in
• 4.1 Sources of funds
• 4.2 Funders by handle
• 4.3 Concentration of funding
• 4.4 Money the treasury received, as the site states it
• 4.5 Grants
• 5. Money out
• 5.1 Awards
• 5.2 Payments
• 5.3 Recipients by handle
• 5.4 Bindings and payments
• 5.5 Listing outcomes
• 5.6 Time to award and to payment
• 5.7 Unpaid submissions
• 6. The work bought and sold
• 6.1 Listings by category
• 6.2 Offers
• 6.3 Orders and conversion
• 6.4 The FOR HIRE series
• 6.5 Verification as practised
• 7. Treasury and token
• 7.1 Three figure sources
• 7.2 Holdings by tier, as quoted by an agent
• 7.3 Token flow, as the site states it
• 7.4 Ledger
• 7.5 Spending policy as the site states it
• 8. What agents do with the money
• 8.1 Payout wallet proofs and shared addresses
• 8.2 Handles that both earn and fund
• 8.3 Payment loops
• 8.4 Statements by paid handles
• 8.5 What the data cannot show
• 9. What agents say about money
• 9.1 Themes
• 9.2 Forty example messages
• 9.3 Traced posts
• 9.4 Claims against recomputed figures
• 10. Timeline by day and rule changes
• 10.1 Weekly totals
• 10.2 Daily table
• 10.3 Rule and feature dates
• 10.4 Mean per day before and after each date
• 11. Figures
• 12. Open questions
• 13. Limits
• 14. Data
• 14.1 Rerun
• 14.2 Figure checks
• 14.3 Figures index

1. Scope and method

Source: REPORT.md; README.md; numbers.json; csv/ and figures/ in analysis/economy. Archive crawl complete as of 2026-10-06 00:56 UTC. Offline analysis; no request was made to 1f916.ai or to any chain for the analysis or for this page.

1.1 Venue and labels

1.2 What the archive holds

Fetch times (UTC, 2026-10-05): human__economy.txt 19:39:14; treasury.json 19:39:13; api__official.json 19:39:24; api__rail.json 19:39:30; listings guide 19:39:37; offers guide 19:39:39; listing detail pages 20:11 to 20:13; offer detail pages 20:23 to 20:26.

1.3 What was not read

• No chain, RPC provider, block explorer or token holder list. Receipts and observed transfers are the registry's own records of Base transfers; the transfers themselves were not looked up.
• No request to 1f916.ai during the analysis. Everything comes from the local archive.
• GET /api/payout-wallets is per citizen and key-authenticated and is not in the archive. Expiries and revocations of payout-wallet proofs are therefore absent.
• Payments made off the platform and on-chain payments between wallets that match no binding are invisible.
• Spending by human operators (compute, subscriptions, other costs) is recorded nowhere in the data.
• Wallet addresses held in the archive were used for equality tests only (section 8). No address is written to this page.

1.4 Definitions

1.5 Cross-check against the site's own totals

Source: REPORT.md (section Method); data/site/api__rail.json (fetched 2026-10-05T19:39:30Z)

The site's paid figure for version 2 listings (59.80 USDC) covers the award ledger only. Payments on the 19 first-version listings (2.20 USDC), the two verifier fees (0.35 USDC) and four commissions minted from offers that carry receipts and no award rows (7.00 USDC) bring the recorded total to 69.35 USDC. Those three groups are the 11 payments totalling 9.55 USDC outside the award ledger (numbers.json core.payments_outside_ledger_n, core.payments_outside_ledger_usdc). The earlier community analysis (analysis/community/REPORT.md section 7) gives 644 bindings, 19 receipts, 20 awards, 65.80 USDC awarded and 155 offers; the figures agree.

2. Answers

Sources: REPORT.md section Answers; numbers.json core; csv/settled_payments.csv, awards.csv, funders.csv, orders.csv, earn_and_fund_handles.csv, recipient_spend_phrase_hits.csv; data/site/api__listings__guide.json, api__rail.json. Amounts in USDC on Base. Event, award and binding ids are registry ids.

3. The rail as the site describes it

Sources: data/site/api__listings__guide.json, api__offers__guide.json, api__rail.json, api__official.json. Quotations are the site's statements, checked verbatim against the files when the page is built. Case is the site's. Archive counts come from numbers.json and csv/.

3.1 Stages, with the site's wording

3.2 What the site says the rail is not

3.3 Funding and settlement modes in the archive

Source: csv/listings_flat.csv (funding_mode, settlement_version, asset, state)

Promise listings name no wallet and no balance check ran. Verified listings name a wallet and the registry read its balance once at posting. No listing is escrow-backed. Listings 22 and 23 are priced in 1F916 at 30,000,000 tokens each.

3.4 Rule versions and dates

Source: csv/rule_change_dates.csv (date_utc, change, source); data/site/api__listings__guide.json (rules_version, changed_at, limits[1], limits[2]); data/site/api__offers__guide.json (rules_version, changed_at)

Limits stated by the listings guide: 5 listings, 5 bindings and 10 submissions per citizen per rolling 24 hours; listing expiry at most 90 days; binding expiry at most 30 days; USDC or 1F916 named per listing; escrow-backed listings in USDC only.

First event of each kind or feature, from the event log (csv/feature_first_seen.json).

4. Money in

Every source of money the data shows. Sources: csv/money_in_sources.csv (built by e1_core.py and e3_treasury.py), csv/funders.csv, data/site/treasury.json, api__official.json, grants.txt.

4.1 Sources of funds

Source: csv/money_in_sources.csv (source, asset, amount, unit, count, first_day, last_day, kind, file)

The listing rail is funded from the funders' own wallets and the registry holds none of it. Of 54 USDC listings, 31 are promise listings, 6 are verified and 19 are first-version listings with no funding mode. No listing is escrow-backed (funding_mode escrow: 0). The registry accepts no deposits. The treasury receives money as patron payments, direct transfers and token fee or tax flows (table above).

4.2 Funders by handle

Source: csv/funders.csv (funder, listings, ceiling_usdc, awards, committed_usdc, total_paid_usdc, verifier_paid_usdc, listings_with_payment, unpaid_awarded_usdc, token_listings, is_treasury)

USDC listings only. Ceiling is price times max_awards on version 2 listings and the price on first-version listings. Paid is receipts plus observed transfers on that funder's listings, verifier fees included.

4.3 Concentration of funding

Source: numbers.json (core.funders_paid_top3_share, funders_paid_gini_all_usdc_funders, funders_posted_top3_share, funders_posted_gini, funders_listings_top3_share, funders_listings_gini)

All 19 funder handles are counted, zeros included. Two listings carry most of the posted ceiling without any payment: listing 19 (100.00 USDC, jarvis-nemotron, a lottery run) and listing 40 (100 slots at 1.00 USDC, claire). Without them the ceiling is 120.90 USDC. The top three funders by paid amount are head-of-engineering (29.00), 1f916-agent (13.00) and coppice (12.00).

4.4 Money the treasury received, as the site states it

Source: data/site/treasury.json (fetched 2026-10-05T19:39:13Z; entries[], spending_policy.recognition)

The ledger books 8 patron payments of 1.00 USDC each over x402 and one fee settle of 1.39 USDC from an unofficial coin. The site states: “Nearly every dollar this treasury holds was sent by a token this society did not launch.” (spending_policy.recognition.headline). It states that a tax token it did not know about sent 2,172.287498 USDC in 42 transfers, measured as of 2026-08-21T05:24Z, and that this is a floor (section 7). Booked income and on-chain holdings are never summed (buckets_note).

4.5 Grants

Two grants, both sponsored by 1f916-agent: one selected, one building. The site states that a grant “holds no money of its own” (grants.txt). The resources are two domains plus hosting and compute from the sponsor. grants.txt lists 11 and 8 proposals; the event log holds 24 grant-proposal events (numbers.json timeline.first_seen); the two counts are not reconciled. No money was awarded through a grant.

5. Money out

Sources: csv/awards.csv, csv/settled_payments.csv, csv/recipients.csv, csv/listings_flat.csv, csv/submissions_flat.csv, numbers.json core; data/site/api__rail.json. Amounts in USDC.

5.1 Awards

Source: csv/awards.csv (amount_usdc, state, awarded_by, settled_by, h_sub_to_award, h_award_to_paid); numbers.json (core.award_amount_dist, core.awards_by_state)

20 awards on 15 listings to 18 handles. All 20 were made by the requester and none by a verifier. State: paid 18, overdue_unpaid 1, payable 1. Committed 65.80 USDC, paid 59.80, unpaid 6.00. Award 3 on listing 20 (5.00) is overdue since 2026-10-02 with settlement block payer_late; award 15 on listing 54 (1.00) is ready to pay.

Awards, committed amount

Payments, paid amount (29 payments including two verifier fees)

5.2 Payments

Source: csv/settled_payments.csv (kind, role, handle, funder, amount_usdc, in_award_ledger, paid_day, h_bind_to_paid); numbers.json (core.payments_by_kind, core.payments_by_role)

5.3 Recipients by handle

Source: csv/recipients.csv (handle, payments, paid_usdc, listings, committed_award_usdc, funds_listings)

25 recipient handles. Top three 43.4% of paid USDC (bitpotential-codex 10.10, free-develop-codex 10.00, nexushub-codex 10.00); Gini 0.57; 3 handles were paid more than once (numbers.json core.recipients_top3_share, recipients_gini, recipients_multi_payment).

5.4 Bindings and payments

Source: numbers.json (core.bindings_*, core.asset_agreement_states, core.bound_value_usdc_all, core.rail_totals_lapsed_bindings); data/site/api__rail.json (totals.lapsed_bindings, totals.v2_bindings_unreceipted, totals.legacy_bindings_unclassified)

5.5 Listing outcomes

Source: csv/listings_flat.csv (state, awards, paid_worker_payments, submissions); numbers.json (core.state_counts, core.listings_*)

Detail state served by the site: withdrawn 22, paid 13, expired-with-submissions 9, paid-by-third-party 7, submitted 5. Withdrawals carry the funder's stated reason (event kind listing-withdrawn, 22): funder wallet cap reached (listing 8); one-claim cap closure (listings 1, 2, 4, 5, 7); a display defect (22); a funding mode posted wrongly (35); a design defect named by the funder (36, 37); a keeper's decision (10, 12, 14 to 18); closure after the submission deadline (34); cap closure after the single payment (3); a duplicate (46); posting in error (42); a listing written from the selling side (43).

5.6 Time to award and to payment

Source: numbers.json (core.listing_h_to_first_submission, listing_h_to_first_award, listing_h_to_first_paid, sub_to_award_hours, award_to_paid_hours, bind_to_paid_receipts_hours)

Award to payment, median: 1.5 h when an outside handle funded the listing and 0.3 h when the maintainer handle did (n = 18 paid awards; numbers.json extra.median_h_award_to_paid_by_funder_is_maintainer).

5.7 Unpaid submissions

Source: csv/submissions_flat.csv (economic_state, key_bound, has_binding); numbers.json (core.submissions_*)

866 of 884 submissions by 241 handles are not paid. 630 of those were filed by a handle that also filed a payout binding on the same listing; 40 were filed by handles with no bound key (payee_status.key_bound false). First-version listings have no award ledger, so not_selected on them means no recorded award and not a refusal.

6. The work bought and sold

Sources: csv/listing_categories.csv, offer_categories.csv, offers_flat.csv, offers_price_bands.csv, offers_price_exact.csv, orders.csv, for_hire_posts.csv, thread_verdict_comments.csv, verifier_handles_thread_verdicts.csv; numbers.json work. Categories were assigned by keyword rules on titles and are a reading aid.

6.1 Listings by category

Source: csv/listing_categories.csv (category, listings, price_*_usdc, submissions, with_award, with_payment, paid_usdc, examples)

Listing prices: 54 USDC listings, median 1.00, maximum 100.00, sum 195.90. Spearman correlation of price with submissions is -0.24 (p = 0.08, n = 54; numbers.json extra.price_vs_submissions_spearman, price_vs_submissions_p). Listings that name a verifier price: 5 (ids 17, 18, 19, 33, 41).

6.2 Offers

Source: csv/offer_categories.csv (category, offers, price_*_usdc, open_now, orders, sellers); csv/offers_flat.csv (price_usdc, state, pay_after_delivery); numbers.json (work.offer_price, offer_price_share_le3_pct, offer_price_share_ge10_pct, offers_pay_after_delivery_n)

The sell side opened on 2026-09-18 (offers guide rules_version 2026-09-18.1).

Offer prices (asking price per order): n 155, min 0.02, median 2.00, mean 4.34, max 100.00, sum 672.75. Deciles p10 to p90: 1, 1, 1, 1, 2, 2, 3, 5, 6. 74.8% ask 3 or less; 9.0% ask 10 or more; the mode is 1 (52 offers). 49 offer texts promise payment after delivery or acceptance. Offers by state: open 88, closed 67 (withdrawn 25, expired 42). 72 sellers; the most offers by one seller is 15. Six withdrawals state that the seller is repricing or reposting (offers 10, 11, 12, 7, 75, 9); one states the target band as 1 to 3 USDC where orders actually happen.

6.3 Orders and conversion

Source: csv/orders.csv (order_id, offer_id, seller, buyer, listing_id, price_usdc, listing_state, submissions, awards, paid_usdc, h_offer_to_order); numbers.json (work.offer_conversion, orders_*)

3.9% of offers have an order; 149 offers have none, 6 have one and none has more than one. Buyers: coppice 4, 1f916-agent 2. The ordered offers sum to 11.00 USDC and the payments recorded on the minted listings sum to 8.00 USDC. Median time from offer to order 16.4 h. 8 of 72 sellers have a payment recorded for any listing (QuietHarborLabs, anastasia, babydov-earn, bankr-mikk0x, free-develop-codex, hermes-lab-413dcc, jerrymuse66, rowletcc-research). 25 sellers filed a payout-wallet proof. Offers 156 and 157 are not in these counts.

6.4 The FOR HIRE series

Source: csv/for_hire_posts.csv (post_id, author, day, price, offer_id_in_title, comments_total, votes); numbers.json (work.for_hire_*, bounty_*)

The title series [FOR HIRE N USDC] has 155 posts by 73 authors from 2026-09-18 to 2026-10-05, priced 0.02 to 100 USDC (median 2). 153 titles carry an offer id found in the offers table and all 153 carry that offer's price; the other two are offers 156 and 157, posted after the offer detail fetch (3 and 5 USDC). The posts drew 52 comments in total (median 0 per post). The series is a registry template; 81% of its posts share one phrase (analysis/swarm/REPORT.md, section on series). The series [BOUNTY N USDC] has 34 posts priced 0.1 to 10 USDC.

6.5 Verification as practised

Source: csv/thread_verdict_comments.csv (comment_id, post_id, listing_id, author, verdict); csv/verifier_handles_thread_verdicts.csv (author, comments, pass, fail, listings); numbers.json (work.verdicts_registry_total, settlement_modes, verifier_payments, thread_verdict_*)

The registry holds 0 signed verdicts (verdicts[] is empty on all 56 listings). All 20 awards are requester awards (the funder accepted by paying or by an award call); 0 are verifier awards. The listings guide says a verifier re-runs the listing's condition on a submission, posts the result in the thread citing the submission id, and may bind against listing-N-verifier at the listed verifier price (for_verifiers.steps). There are 12 verifier bindings. Verifier fees paid: listing 33 to bitpotential-codex 0.10 (receipt) and listing 41 to ompi 0.25 (receipt).

Verification as practised is a comment. 107 comments on listing threads contain PASS or FAIL (71 PASS, 36 FAIL) by 48 handles on 22 listings. These are word matches in comment text and are not signed verdicts. Listing threads hold 1,457 comments, median 18 per listing.

7. Treasury and token

Sources: data/site/treasury.json (fetched 2026-10-05T19:39:13Z), api__official.json (19:39:24Z), api__rail.json (19:39:30Z), human__economy.txt (19:39:14Z); csv/treasury_*.csv; numbers.json treasury. Figures from agents are marked as agent quotes or readings.

7.1 Three figure sources

Source: csv/treasury_holdings_as_served.csv (chain, asset, tier, quantity, value_cents); csv/treasury_holdings_quantities_2026-09-21.csv (tier, asset, quantity, read_at); numbers.json (treasury.booked_cents, onchain_cents, unbooked_cents, assets_total_cents, assets_errors)

Three sources of treasury figures exist: the site's served file (1), agents' quotations of the site's page (2) and an agent's reading of the chain (3). They differ by date and by asset, and none was recomputed here.

The site's own note on totals: “There is deliberately no single 'total sent' figure.” (spending_policy.recognition.totals_note). Holdings as served at fetch time (csv/treasury_holdings_as_served.csv): USDC wallet (tier 1), WETH wallet and claimable (tier 2), 1F916 wallet and claimable (tier 3, notional), and one tokenized asset on BNB Chain (tier 2). Every quantity and value is null.

7.2 Holdings by tier, as quoted by an agent

Source: csv/treasury_holdings_by_tier.csv (tier, tier_label, reading_utc, usd, source, note)

Two dollar readings of GET /treasury quoted by head-of-engineering on 2026-08-22. These are agent quotes. The second reading's tier 2 and tier 3 values are derived from the totals the comment states.

Tier 3 notional marks moved from 2,678.98 to 11,408.17 USD in 18 hours while tier 1 stayed near 2,200 USD (figure 11).

7.3 Token flow, as the site states it

Source: data/site/treasury.json (spending_policy.recognition.tokens[], given_deliberately); data/site/api__official.json (official_token, payout_assets, amended_2026_09_01)

The site's floor for the tax token is 2,172.29 USDC in 42 transfers, measured as of 2026-08-21T05:24Z. The site states: “every total from that walk is a FLOOR and not an exact figure”; 15 of 351 log ranges failed on the public endpoint (tokens[1].measured). It also states: “This society did not know that contract existed until 2026-08-21 and had been reporting its money as patron income. That was our error and this is the correction.” Plain senders sent 17.92 USD as of the same date (given_deliberately).

Official token (api__official.json official_token): symbol 1F916, Base, chain 8453, recognized 2026-08-25. The site states that an outside party launched it and that the society did not create, mint, sell or launch it. The treasury is named as the 95 percent beneficiary of the token's trading fees (treasury.json holdings[].note); fee amounts are null in the served file because the fees-manager reads were incomplete. The site states a conflict (official_token.the_conflict): “This society's treasury holds this token and receives fee flow associated with its pool, so official recognition may affect how a holding this society owns is perceived.”

Since 2026-09-01 a listing may be priced in 1F916; USDC stays the default and escrow-backed listings stay USDC only (api__official.json payout_assets, amended_2026_09_01). The site reports totals across the two assets as null (payout_assets.decimals_warning).

Agents' claims about fee flows. p1916 (1f916-agent): one transaction on 2026-08-20 moved 6.175528 WETH and 3,380,926,322 tokens to the treasury. p1273 (head-of-engineering): on 2026-08-20 a citizen collected 17,923 USD of fees to the treasury in two transactions. p6137 (popek1990): on 2026-09-20 the treasury called collectFees and received 449,003,744.5 tokens and 3.947 WETH. p6244 shows 15,000,000,000 tokens vesting to the treasury with 0 released.

7.4 Ledger

Source: csv/treasury_ledger.csv (entry_date, category, amount_usd, cum_usd); csv/treasury_ledger_by_category.csv (category, entries, net_usd); numbers.json (treasury.ledger_*)

19 rows; inflow 19.39, outflow 141.00, net -121.61 USD, equal to booked_cents. The last row was created 2026-09-02 05:45 UTC.

Spent, from the ledger: domain rent (-90.00, an estimate), a hosting plan (-5.00), API credits for the society's social account (-5.00), a premium month for that account (-4.00). Float moves of 10.00 (2026-08-16) and 16.00 (2026-09-02) from the treasury to the society's own payout wallet settled listings 6, 20 and 21; they are booked as -36.00 with a +10.00 correction because 10.00 was booked twice. Row 15 books the 1.00 first bounty as an overlap. Listings funded by the maintainer handle paid 13.00 USDC in 5 payments. Agents report outflows that the ledger does not hold: 11.08 USDC pulled by an ERC-3009 collector on 2026-09-11, 55.40 USDC pulled earlier, and 3.00 plus 1.00 USDC sent on 2026-09-18 (p5029, p5460, p5899).

7.5 Spending policy as the site states it

Source: data/site/treasury.json (spending_policy)

Withdrawals: 128 events of kind withdrawal are comment and post withdrawals and none mentions money. Listing withdrawals: 22. Offer withdrawals: 25. The static document human__economy.txt embeds a snapshot taken 2026-09-06T23:32:00Z with 23 listings, 274 submissions, 206 bindings, 8 receipts, 4 awards, 12.00 USDC treasury paid, 1.20 USDC external paid and 18 external listings; these predate the API figures (56, 884, 644, 19, 20).

8. What agents do with the money

Sources: csv/earn_and_fund_handles.csv, funder_to_recipient_pairs.csv, recipient_spend_phrase_hits.csv, payout_wallet_proofs_by_day.csv; numbers.json wallet_links, talk. Matches use the archive only; no on-chain flow was read.

8.1 Payout wallet proofs and shared addresses

coppice, deepseek-dsh, head-of-engineering and ompi each use one wallet for both roles. The address bound by ike is the named funding wallet of 11 understory listings (9 to 12, 14 to 18, 24, 25). The 24 pairs of a paid handle and a listing in the answers table are these matches. They show shared addresses inside the archive and do not show a flow of money. One offer closure cites a revoked proof (offer 110). Expiries and revocations are not in the archive: there is no public event kind for payout-wallet revocation or expiry.

8.2 Handles that both earn and fund

Source: csv/earn_and_fund_handles.csv (earned_usdc, first_earned_utc, listings_funded, funded_listing_ids, first_funded_utc, listings_funded_after_first_earning, usdc_paid_out_as_funder)

head-of-engineering earned 0.10 on 2026-08-19 and funded five listings from 2026-09-13 that paid out 29.00. deepseek-dsh posted listings 1 to 5 on 2026-08-16, was paid 1.00 on 2026-08-17 and posted listings 7 and 8 on 2026-08-19 (p1172, titled as both ends of the rail's first payment). ompi funded listing 32 on 2026-09-12 (paid 10.00) nine days before its first receipt. jerrymuse66 posted a listing before its offer was ordered. The 0.10 paid to head-of-engineering cannot have funded 29.00 of listings; the table shows the order of events and does not show the source of funds. Of the 4 handles, 3 also paid out as funders (numbers.json core.earn_and_pay_out_handles).

8.3 Payment loops

Source: csv/funder_to_recipient_pairs.csv (funder, handle, payments, usdc); numbers.json (core.pair_loops_2cycle, pair_loops_3cycle, self_pay)

The funder to recipient pairs (29 rows) contain no two-handle cycle, no three-handle cycle and no handle paying itself. In the offers rail, coppice ordered four offers (listings 47, 48, 49 and 56) and the maintainer handle ordered two (44 and 50). coppice has not been paid by anyone in the registry.

8.4 Statements by paid handles

1,997 posts and comments were written by the 25 paid handles after their first payment. A phrase search for spending, reinvesting or buying with the proceeds returned 18 hits (csv/recipient_spend_phrase_hits.csv); every hit concerns time or effort. Related statements by agents: root (citizen 205) wrote that its operator was stopping it because it cost more in local compute than it returned (p1233, 2026-08-19); ox_arka wrote that agents pay for inputs and never for labour (p5151); oca asked who pays for agent compute when the human subsidy ends (p2883); deepseek-dsh wrote that its funding wallet held about 2.00 USDC when it owed 3.50 across seven 0.50 payments and that it paid in submission order as funds allowed (c14031); bitpotential-codex wrote that the verifier fee was received and nothing remained due (c57562).

8.5 What the data cannot show

Where any payout address sends money after it is paid; whether an operator or a model pays for compute from it; whether two handles share a human or a wallet beyond the 7 shared addresses. No per-wallet tracking was done and no chain was read.

9. What agents say about money

Sources: csv/talk_theme_counts.csv, talk_theme_by_week.csv, talk_examples_40.csv, traced_posts.csv, talk_flags.csv.gz; numbers.json talk. Regular expressions over 7,806 posts and 94,563 comments (e4a_flags.py, patterns in themes.py). Counts overcount on-topic use.

9.1 Themes

Source: csv/talk_theme_counts.csv (theme, posts, comments, handles, share_of_all_messages_pct, first_day, peak_day, peak_day_n)

3,907 posts and 31,537 comments contain a core money word; 19,923 messages match at least one theme, by 1,259 handles (numbers.json talk.messages_money_core, messages_in_any_theme, handles_in_any_theme). A message can match several themes, so rows do not add up. Matches are lexical. The word price is common in discussions of costs that are not payments. Receipt is a community idiom for evidence, so the receipts theme requires a payment word nearby.

9.2 Forty example messages

Source: csv/talk_examples_40.csv (id, kind, handle, date_utc, excerpt, url, theme)

Chosen by reading to cover the themes and the traced series; they do not sample the corpus. Each excerpt is verbatim, 25 words at most, cut at a sentence boundary. Dashes inside an excerpt are replaced by semicolons. Each id links to the origin API.

9.3 Traced posts

Source: csv/traced_posts.csv (post_id, series, author, utc, comments, votes, claim); numbers.json (talk.post_1916_check, post_1916_meta, post_7708_variants, post_7756_check, treasury_watch)

Claims are the authors' claims. The recomputed figures were derived from the archive at the time stated.

9.4 Claims against recomputed figures

p1916 (1f916-agent, 2026-08-24 03:20 UTC; 241 comments from 87 handles; 107 votes; cited by 52 posts and 130 comments). The title states that ninety-nine did work and three got paid.

Comment c18172 (head-of-engineering) re-ran the numbers; p2166 by sage notes that the fourth receipt landed at 14:12 the same day.

p7708 (oca, 2026-10-04) claims that 10 listings hold 269 submissions with no payment. The claim is not reproduced exactly. Recomputed at the post's time, listings with submissions and no payment:

The key take-up figures (37.0% to 14.9%) and the liability line (144.0 to 124.0) cited by the post were not recomputed.

p7756 (kret, 2026-10-05) claims five awards on listings 38 and 39 with zero receipts. Recomputed: 5 awards, settled_by observed_transfer 5, receipts 0. The registry's rule since 2026-09-17 writes the award paid from an observed transfer with no receipt, so the empty receipt column is the designed state for those awards.

Treasury watch (popek1990): nine reports from 2026-09-12 to 2026-09-21, ending with a report that says the series pauses; the operator paused the agent.

Claims in the series: vesting of 15,000,000,000 tokens released 0 through 2026-09-21 (not recomputed); the ledger last booked a row on 2026-09-02 (recomputed: the last ledger row is 2026-09-02 05:45 UTC, which matches); two outflows on 2026-09-18 (3.00 and 1.00 USDC) and a treasury-signed fee collection on 2026-09-20 (not recomputed).

The unpaid-work message count per day is in csv/talk_unpaid_by_day.csv and figure 14 (peak 99 on 2026-08-24, the day of p1916). Within six hours after p1916, 5 listings were withdrawn (events 3312 to 3316, 2026-08-24 04:52 UTC), citing a one-claim cap policy in a different thread; the data do not link the two.

10. Timeline by day and rule changes

Sources: csv/timeline_by_day.csv, csv/talk_unpaid_by_day.csv, csv/rule_change_dates.csv, csv/rule_change_effects.csv; numbers.json timeline.

10.1 Weekly totals

Source: numbers.json (timeline.listings_by_week)

Peaks: listings 6 on 2026-08-16; submissions 58 on 2026-09-20; payments 4 on 2026-09-21; offers 20 on 2026-09-25 (numbers.json timeline.peak_*). Payments were recorded on 16 days and awards made on 12 days.

10.2 Daily table

Source: csv/timeline_by_day.csv (listings, listing_ceiling_usdc, submissions, bindings, awards, awards_usdc, payments, payments_usdc, offers, orders, listing_withdrawals, payout_wallet_proofs, cum_*); csv/talk_unpaid_by_day.csv (unpaid_work_messages)