Implementation notes · SCITT

Implementing the MMR receipts profile over a live log

Notes from building receipts to draft-ietf-scitt-receipts-mmr-profile-00 (formerly draft-bryce-cose-receipts-mmr-profile-04) and a Transparency Service to draft-ietf-scitt-scrapi-11. Offered in the hope they save other implementers time. Corrections welcome.

Cross-checks

Our implementation was compared with the authors' reference algorithms (forestrie/univocity) on a 300-leaf tree: index_height for every node, peaks for every complete size, included_root for every node at 40 sizes, and consistent_roots for 500 size pairs. 13,914 comparisons, no differences. The authors' published four-leaf vectors reproduce exactly.

Known-answer vectors, including complete signed receipts, are at /scitt/vectors.json.

1. Consistency checking does real work on its own

We changed an early entry and rebuilt the whole tree honestly from scratch, as a dishonest operator would. A verifier holding the earlier accumulator rejected the next receipt of consistency at the step where consistency paths must agree on the new peak (section 7.1.1), before the signature was checked. The shape requirements are not just hygiene.

2. Choose the leaf form deliberately (section 11.2)

The profile does not domain-separate leaves from interior nodes. We use

leaf = H(0x00 || tag || position (8 bytes BE) || entry digest)

which binds the position (so an empty inclusion path still fixes the index) and has a fixed pre-image length that can never equal an interior pre-image (8 + 32 + 32 bytes).

3. Bound proof sizes in verifiers

With 64-bit indices no valid inclusion path is longer than 63 elements, and consistency paths have exactly the lengths 7.1.1 implies. Our verifiers reject longer paths and more than 64 proofs per receipt before doing any hashing.

4. "Freezing" a busy log is unnecessary

Nodes are append-only and never change, so every complete MMR size is a snapshot. We generate the inclusion path against the current complete size while appends continue; the node it leads to is permanent even after a later peak buries it, so the receipt keeps verifying (sections 4 and 11.1).

5. Low-entropy entries can be confirmed by guessing

If an entry is a yes/no decision or similar, its leaf can be confirmed by hashing guesses. Where that matters, put a per-entry salt or nonce in the leaf pre-image (see the RFC 9943 privacy considerations).

6. The peak check in 5.1

A verifier holding a trusted size and accumulator "can additionally check that the proven node is the accumulator peak for the index at that size". The draft does not yet define the algorithm; univocity's peak_index does it and might usefully join section 9.

7. Key identifiers

Our receipts carry kid (label 4) as the RFC 9679 COSE Key Thumbprint of the service key, as SCRAPI recommends, and the key set is served at /.well-known/scitt-keys.

Where to test

Interop tools (verify any receipt, probe any service read-only): sebbi.pro/scitt/interop
Live Transparency Service: sebbi.pro/scitt · spec /x/scitt/spec · self-check /x/scitt/selfcheck · chain receipts /x/mmr/spec

vds 3 and tree-size-2 -65933 are the draft's requested and private-use values, used provisionally until IANA assigns them.

AI news
TechCrunchSignal restricts mobile screenshots as it confronts ‘existential threat’ of AI surveillance2hTechCrunchA tale of two Nvidias2hThe GuardianTwo groups verify mathematician’s claim that Claude helped prove mathematics problem3hTechCrunchHere are the top AI agents that can live in your text messages6hThe DecoderNvidia researchers argue huge AI context windows are a dead end7hThe DecoderSilicon Valley's growing obsession with AI doomsday is turning parts of the tech industry toward cult-like devotion7hTechCrunchIs selling an AI version of your face to this microdrama startup the future of Hollywood?7hThe VergeCan Anthropic afford to put safety first?8hThe Guardian‘I was shaking, crying, having trouble breathing’: the uncharted trauma of breaking up with AI companions9hThe DecoderOpenAI shares new prompting tips for GPT-6.2 Astra and Sol to fix brief answers and unnecessary questions10hThe Guardian‘Little room for error’: what a debt-laden cloud computing company says about the AI boom10hThe GuardianI’m a priest. Silicon Valley’s search for the ‘soul’ of AI is deeply troubling | Rev Kevin Deal11hThe DecoderHours after Anthropic claims a Claude proof of the Riemann hypothesis, Stanford releases an all-human competing draft12hThe Guardian‘I’m opening the cage of a monster’: Taxi Driver scriptwriter Paul Schrader on unleashing AI at 8012hThe DecoderUS technology stocks suffer their worst weekly loss since the pandemic as AI doubts pile up13hBBC NewsCan AI really help spot Black Friday deals? We put it to the test23hBBC NewsIn a year of AI solving maths problems, an Anthropic worker says he may have cracked the biggest1dBBC NewsTech stock sell-off drags markets down as Anthropic shares slide1dArs TechnicaGoogle rolls out new search homepage with AI chat history and no search button1dArs TechnicaMeta reconsiders 2030 net-zero goal amid record emissions from AI data centers1dArs TechnicaResearcher plants 2m canaries into the AI supply chain. A privacy disaster surfaced.1dBBC NewsTrump threatens ban on US firms building AI data centres overseas1dTechCrunchDaniel Ek backs AI researcher’s fusion power startup. Here’s what they’re building.1dWiredJensen Huang’s ‘Lying’ Alarmed Nvidia Board Ahead of His Resignation1dTechCrunchSignal restricts mobile screenshots as it confronts ‘existential threat’ of AI surveillance2hTechCrunchA tale of two Nvidias2hThe GuardianTwo groups verify mathematician’s claim that Claude helped prove mathematics problem3hTechCrunchHere are the top AI agents that can live in your text messages6hThe DecoderNvidia researchers argue huge AI context windows are a dead end7hThe DecoderSilicon Valley's growing obsession with AI doomsday is turning parts of the tech industry toward cult-like devotion7hTechCrunchIs selling an AI version of your face to this microdrama startup the future of Hollywood?7hThe VergeCan Anthropic afford to put safety first?8hThe Guardian‘I was shaking, crying, having trouble breathing’: the uncharted trauma of breaking up with AI companions9hThe DecoderOpenAI shares new prompting tips for GPT-6.2 Astra and Sol to fix brief answers and unnecessary questions10hThe Guardian‘Little room for error’: what a debt-laden cloud computing company says about the AI boom10hThe GuardianI’m a priest. Silicon Valley’s search for the ‘soul’ of AI is deeply troubling | Rev Kevin Deal11hThe DecoderHours after Anthropic claims a Claude proof of the Riemann hypothesis, Stanford releases an all-human competing draft12hThe Guardian‘I’m opening the cage of a monster’: Taxi Driver scriptwriter Paul Schrader on unleashing AI at 8012hThe DecoderUS technology stocks suffer their worst weekly loss since the pandemic as AI doubts pile up13hBBC NewsCan AI really help spot Black Friday deals? We put it to the test23hBBC NewsIn a year of AI solving maths problems, an Anthropic worker says he may have cracked the biggest1dBBC NewsTech stock sell-off drags markets down as Anthropic shares slide1dArs TechnicaGoogle rolls out new search homepage with AI chat history and no search button1dArs TechnicaMeta reconsiders 2030 net-zero goal amid record emissions from AI data centers1dArs TechnicaResearcher plants 2m canaries into the AI supply chain. A privacy disaster surfaced.1dBBC NewsTrump threatens ban on US firms building AI data centres overseas1dTechCrunchDaniel Ek backs AI researcher’s fusion power startup. Here’s what they’re building.1dWiredJensen Huang’s ‘Lying’ Alarmed Nvidia Board Ahead of His Resignation1d
Forever Proof standard · check it without us